feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
//
|
|
|
|
|
// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
|
|
|
|
|
//
|
|
|
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
|
// you may not use this file except in compliance with the License.
|
|
|
|
|
// You may obtain a copy of the License at
|
|
|
|
|
//
|
|
|
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
|
//
|
|
|
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
|
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
|
// See the License for the specific language governing permissions and
|
|
|
|
|
// limitations under the License.
|
|
|
|
|
//
|
|
|
|
|
|
2026-06-17 13:24:03 +08:00
|
|
|
// Package component — Browser (T3).
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
//
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Browser is an LLM-driven single-shot web extraction canvas node
|
|
|
|
|
// built on `github.com/browserbase/stagehand-go/v3` in local mode.
|
|
|
|
|
// It uses `RunExtract` (not the multi-step agent `RunTask`) to
|
|
|
|
|
// navigate to a page and extract structured content against a
|
|
|
|
|
// `{"type": "string"}` JSON schema.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
//
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// It mirrors the Python `agent/component/browser.py` param surface
|
|
|
|
|
// (`llm_id`, `prompts`, `max_steps`, `headless`, `persist_session`,
|
|
|
|
|
// `upload_sources`, etc.) so the v1 fixture
|
|
|
|
|
// (`internal/agent/dsl/testdata/browser.json`) loads without
|
|
|
|
|
// fixture-side changes.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
//
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// LLM dispatch is delegated to `StagehandInvoker` (see
|
|
|
|
|
// `stagehand_runtime.go`), which owns the stagehand-server child
|
|
|
|
|
// process and the session lifecycle. The component itself is a thin
|
|
|
|
|
// orchestrator: parse → resolve template → look up tenant model
|
|
|
|
|
// config → call runtime.RunExtract → emit Python-shaped outputs.
|
|
|
|
|
//
|
|
|
|
|
// File upload / download and persistent session management are
|
|
|
|
|
// not supported; see [`.claude/plans/tingly-weaving-orbit.md`]
|
|
|
|
|
// for the full deferral list.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
package component
|
|
|
|
|
|
|
|
|
|
import (
|
|
|
|
|
"context"
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
"encoding/json"
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
"errors"
|
|
|
|
|
"fmt"
|
2026-07-13 16:57:56 +08:00
|
|
|
"net/url"
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
"regexp"
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
"strings"
|
|
|
|
|
|
|
|
|
|
"ragflow/internal/agent/runtime"
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
"ragflow/internal/dao"
|
|
|
|
|
"ragflow/internal/entity"
|
2026-07-13 16:57:56 +08:00
|
|
|
|
|
|
|
|
"gorm.io/gorm"
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
)
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
const componentNameBrowser = "Browser"
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
|
2026-07-13 16:57:56 +08:00
|
|
|
var stagehandNativeProviders = map[string]string{
|
|
|
|
|
"anthropic": "anthropic",
|
|
|
|
|
"google": "google",
|
|
|
|
|
"openai": "openai",
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var browserFactoryDefaultBaseURL = map[string]string{
|
|
|
|
|
"302.ai": "https://api.302.ai/v1",
|
|
|
|
|
"anthropic": "https://api.anthropic.com/",
|
|
|
|
|
"astraflow": "https://api-us-ca.umodelverse.ai/v1",
|
|
|
|
|
"astraflow-cn": "https://api.modelverse.cn/v1",
|
|
|
|
|
"avian": "https://api.avian.io/v1",
|
|
|
|
|
"cometapi": "https://api.cometapi.com/v1",
|
|
|
|
|
"dashscope": "https://dashscope.aliyuncs.com/compatible-mode/v1",
|
|
|
|
|
"deepseek": "https://api.deepseek.com/v1",
|
|
|
|
|
"deerapi": "https://api.deerapi.com/v1",
|
|
|
|
|
"futurmix": "https://futurmix.ai/v1",
|
|
|
|
|
"giteeai": "https://ai.gitee.com/v1/",
|
|
|
|
|
"hunyuan": "https://api.hunyuan.cloud.tencent.com/v1",
|
|
|
|
|
"jiekouai": "https://api.jiekou.ai/openai",
|
|
|
|
|
"lingyi-ai": "https://api.lingyiwanwu.com/v1",
|
|
|
|
|
"longcat": "https://api.longcat.chat/openai",
|
|
|
|
|
"minimax": "https://api.minimaxi.com/v1",
|
|
|
|
|
"moonshot": "https://api.moonshot.cn/v1",
|
|
|
|
|
"n1n": "https://api.n1n.ai/v1",
|
|
|
|
|
"novitaai": "https://api.novita.ai/v3/openai",
|
|
|
|
|
"openai": "https://api.openai.com/v1",
|
|
|
|
|
"openrouter": "https://openrouter.ai/api/v1",
|
|
|
|
|
"perfxcloud": "https://cloud.perfxlab.cn/v1",
|
|
|
|
|
"ppio": "https://api.ppinfra.com/v3/openai",
|
|
|
|
|
"siliconflow": "https://api.siliconflow.cn/v1",
|
|
|
|
|
"stepfun": "https://api.stepfun.com/v1",
|
|
|
|
|
"tongyi-qianwen": "https://dashscope.aliyuncs.com/compatible-mode/v1",
|
|
|
|
|
"upstage": "https://api.upstage.ai/v1/solar",
|
|
|
|
|
"zhipu-ai": "https://open.bigmodel.cn/api/paas/v4",
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// browserParam is the static DSL param surface for the Browser
|
|
|
|
|
// component. Mirrors Python `browser.py:LLMParam + browser knobs`:
|
|
|
|
|
//
|
|
|
|
|
// llm_id, model_id (alias), prompts, prompt (alias),
|
|
|
|
|
// max_steps, headless, enable_default_extensions,
|
|
|
|
|
// chromium_sandbox, persist_session, upload_sources.
|
|
|
|
|
//
|
|
|
|
|
// Go-only fields kept for backward compat with the existing test
|
|
|
|
|
// file and the optional-URL form some operators still wire up:
|
|
|
|
|
//
|
|
|
|
|
// url, timeout.
|
|
|
|
|
//
|
|
|
|
|
// v1 does not act on the v1-deferred params; Update accepts them so
|
|
|
|
|
// the v1 fixture loads.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
type browserParam struct {
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
LLMID string `json:"llm_id"`
|
|
|
|
|
ModelID string `json:"model_id"` // alias for llm_id
|
|
|
|
|
Prompts string `json:"prompts"`
|
|
|
|
|
Prompt string `json:"prompt"` // alias for prompts
|
|
|
|
|
MaxSteps int `json:"max_steps"`
|
|
|
|
|
Headless *bool `json:"headless"`
|
|
|
|
|
EnableDefaultExt *bool `json:"enable_default_extensions"`
|
|
|
|
|
ChromiumSandbox *bool `json:"chromium_sandbox"`
|
|
|
|
|
PersistSession *bool `json:"persist_session"`
|
|
|
|
|
|
|
|
|
|
// Go-only fields (kept for backward compat with the existing
|
|
|
|
|
// test file; not used by the stagehand path).
|
|
|
|
|
URL string `json:"url"`
|
|
|
|
|
Timeout int `json:"timeout"`
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// llmIDPattern matches `ModelName@Factory`. The factory part is
|
|
|
|
|
// optional; when absent, the caller's tenant lookup will be
|
|
|
|
|
// `GetByTenantAndModelName` instead of
|
|
|
|
|
// `GetByTenantFactoryAndModelName`.
|
|
|
|
|
var llmIDPattern = regexp.MustCompile(`^(.+)@(.+)$`)
|
|
|
|
|
|
|
|
|
|
// resolveLLMID splits `llm_id` (e.g. "deepseek-v4-pro@DeepSeek") into
|
|
|
|
|
// `(modelName, factory)`. When no `@` is present, factory is empty
|
|
|
|
|
// and the caller must use a single-key lookup.
|
|
|
|
|
//
|
|
|
|
|
// Mirrors the contract of `dao.splitModelNameAndFactory` (private);
|
|
|
|
|
// re-implemented here to keep the component free of an import
|
|
|
|
|
// dependency on a DB-validating private helper.
|
|
|
|
|
func resolveLLMID(llmID string) (modelName, factory string) {
|
|
|
|
|
m := llmIDPattern.FindStringSubmatch(strings.TrimSpace(llmID))
|
|
|
|
|
if m == nil {
|
|
|
|
|
return strings.TrimSpace(llmID), ""
|
|
|
|
|
}
|
|
|
|
|
return strings.TrimSpace(m[1]), strings.TrimSpace(m[2])
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Update copies a fresh param map into the receiver. The
|
|
|
|
|
// `llm_id`/`model_id` and `prompts`/`prompt` alias pairs collapse
|
|
|
|
|
// onto the same field; the first non-empty value wins.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (p *browserParam) Update(conf map[string]any) error {
|
|
|
|
|
if conf == nil {
|
|
|
|
|
conf = map[string]any{}
|
|
|
|
|
}
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
if v, ok := stringFrom(conf, "llm_id"); ok && v != "" {
|
|
|
|
|
p.LLMID = v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := stringFrom(conf, "model_id"); ok && v != "" && p.LLMID == "" {
|
|
|
|
|
p.LLMID = v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := stringFrom(conf, "prompts"); ok && v != "" {
|
|
|
|
|
p.Prompts = v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := stringFrom(conf, "prompt"); ok && v != "" && p.Prompts == "" {
|
|
|
|
|
p.Prompts = v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := intFrom(conf, "max_steps"); ok {
|
|
|
|
|
p.MaxSteps = v
|
|
|
|
|
} else {
|
|
|
|
|
p.MaxSteps = 0
|
|
|
|
|
}
|
|
|
|
|
if v, ok := boolFrom(conf, "headless"); ok {
|
|
|
|
|
p.Headless = &v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := boolFrom(conf, "enable_default_extensions"); ok {
|
|
|
|
|
p.EnableDefaultExt = &v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := boolFrom(conf, "chromium_sandbox"); ok {
|
|
|
|
|
p.ChromiumSandbox = &v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := boolFrom(conf, "persist_session"); ok {
|
|
|
|
|
p.PersistSession = &v
|
|
|
|
|
}
|
|
|
|
|
if v, ok := stringFrom(conf, "url"); ok {
|
|
|
|
|
p.URL = v
|
|
|
|
|
}
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
if v, ok := intFrom(conf, "timeout"); ok {
|
|
|
|
|
p.Timeout = v
|
|
|
|
|
} else {
|
|
|
|
|
p.Timeout = 0
|
|
|
|
|
}
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Check validates the param. The accepted-but-ignored Python
|
|
|
|
|
// fields are NOT validated here — the v1 fixture is allowed to set
|
|
|
|
|
// them; we only reject structurally invalid values for fields we
|
|
|
|
|
// actually use (`llm_id`, `prompts`).
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (p *browserParam) Check() error {
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
if p.LLMID == "" {
|
|
|
|
|
return &ParamError{Field: "llm_id", Reason: "required (or model_id alias)"}
|
|
|
|
|
}
|
|
|
|
|
if p.Prompts == "" {
|
|
|
|
|
return &ParamError{Field: "prompts", Reason: "required (or prompt alias)"}
|
|
|
|
|
}
|
|
|
|
|
if p.MaxSteps < 0 {
|
|
|
|
|
return &ParamError{Field: "max_steps", Reason: "must be non-negative"}
|
|
|
|
|
}
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
if p.Timeout < 0 {
|
|
|
|
|
return &ParamError{Field: "timeout", Reason: "must be non-negative"}
|
|
|
|
|
}
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// AsDict returns the param as a plain map (for serialization / debug).
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (p *browserParam) AsDict() map[string]any {
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
out := map[string]any{
|
|
|
|
|
"llm_id": p.LLMID,
|
|
|
|
|
"model_id": p.LLMID, // alias echoed
|
|
|
|
|
"prompts": p.Prompts,
|
|
|
|
|
"prompt": p.Prompts, // alias echoed
|
|
|
|
|
"max_steps": p.MaxSteps,
|
|
|
|
|
"url": p.URL,
|
|
|
|
|
"timeout": p.Timeout,
|
|
|
|
|
}
|
|
|
|
|
if p.Headless != nil {
|
|
|
|
|
out["headless"] = *p.Headless
|
|
|
|
|
}
|
|
|
|
|
if p.EnableDefaultExt != nil {
|
|
|
|
|
out["enable_default_extensions"] = *p.EnableDefaultExt
|
|
|
|
|
}
|
|
|
|
|
if p.ChromiumSandbox != nil {
|
|
|
|
|
out["chromium_sandbox"] = *p.ChromiumSandbox
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
if p.PersistSession != nil {
|
|
|
|
|
out["persist_session"] = *p.PersistSession
|
|
|
|
|
}
|
|
|
|
|
return out
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// BrowserComponent is the canvas Browser node. Owns its static
|
|
|
|
|
// param; delegates the multi-step agent run to StagehandInvoker.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
type BrowserComponent struct {
|
|
|
|
|
name string
|
|
|
|
|
param browserParam
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// NewBrowserComponent constructs a Browser from the DSL param map.
|
|
|
|
|
func NewBrowserComponent(params map[string]any) (Component, error) {
|
|
|
|
|
p := &browserParam{}
|
|
|
|
|
if err := p.Update(params); err != nil {
|
|
|
|
|
return nil, fmt.Errorf("Browser: param update: %w", err)
|
|
|
|
|
}
|
|
|
|
|
if err := p.Check(); err != nil {
|
|
|
|
|
return nil, fmt.Errorf("Browser: param check: %w", err)
|
|
|
|
|
}
|
|
|
|
|
return &BrowserComponent{
|
|
|
|
|
name: componentNameBrowser,
|
|
|
|
|
param: *p,
|
|
|
|
|
}, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Name returns the registered component name.
|
|
|
|
|
func (b *BrowserComponent) Name() string { return b.name }
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Invoke dispatches a single-shot extraction task via
|
|
|
|
|
// StagehandInvoker.RunExtract with a `{"type": "string"}` schema.
|
|
|
|
|
// The flow:
|
|
|
|
|
//
|
2026-07-13 16:57:56 +08:00
|
|
|
// 1. Pull tenant_id from `state.Sys["tenant_id"]`.
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 2. Resolve the `prompts` template via `runtime.ResolveTemplate`.
|
2026-07-13 16:57:56 +08:00
|
|
|
// 3. Resolve `llm_id` as a tenant_model id or legacy model@factory
|
|
|
|
|
// reference and look up the tenant LLM config (apiKey, baseURL).
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 4. Build `RunExtractRequest` with `ModelName = "openai/<model>"`,
|
|
|
|
|
// the resolved apiKey/baseURL/instruction, and
|
|
|
|
|
// `Schema = {"type": "string"}`.
|
|
|
|
|
// 5. Call `DefaultRuntime.RunExtract` → raw JSON string.
|
|
|
|
|
// 6. Unmarshal the JSON string to get the plain text content.
|
|
|
|
|
// 7. Emit the Python-shaped outputs (`content`,
|
|
|
|
|
// `downloaded_files`) plus Go-native compat keys.
|
|
|
|
|
//
|
|
|
|
|
// File upload/download and session persistence are not supported
|
|
|
|
|
// in this component; they are v1-deferred.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (b *BrowserComponent) Invoke(ctx context.Context, inputs map[string]any) (map[string]any, error) {
|
|
|
|
|
state, _, err := runtime.GetStateFromContext[*runtime.CanvasState](ctx)
|
|
|
|
|
if err != nil {
|
|
|
|
|
return nil, fmt.Errorf("Browser: %w", err)
|
|
|
|
|
}
|
|
|
|
|
if state == nil {
|
|
|
|
|
return nil, errors.New("Browser: nil canvas state")
|
|
|
|
|
}
|
|
|
|
|
|
2026-07-13 16:57:56 +08:00
|
|
|
tenantID, _ := state.Sys["tenant_id"].(string)
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
if tenantID == "" {
|
2026-07-13 16:57:56 +08:00
|
|
|
return nil, errors.New("Browser: tenant_id missing from canvas state (state.Sys[\"tenant_id\"])")
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 1. Resolve prompts template.
|
2026-07-13 16:57:56 +08:00
|
|
|
prompts := b.param.Prompts
|
|
|
|
|
if v, ok := inputs["prompts"].(string); ok && strings.TrimSpace(v) != "" {
|
|
|
|
|
prompts = v
|
|
|
|
|
}
|
|
|
|
|
resolvedPrompts, err := runtime.ResolveTemplate(prompts, state)
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
if err != nil {
|
|
|
|
|
return nil, fmt.Errorf("Browser: resolve prompts template: %w", err)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 2. Look up tenant model config.
|
2026-07-13 16:57:56 +08:00
|
|
|
providerName, modelName, apiKey, baseURL, err := resolveBrowserLLM(tenantID, b.param.LLMID)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
if err != nil {
|
2026-07-13 16:57:56 +08:00
|
|
|
return nil, fmt.Errorf("Browser: tenant llm lookup (%q): %w", b.param.LLMID, err)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
2026-07-13 16:57:56 +08:00
|
|
|
baseURL = strings.TrimSpace(baseURL)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 3. Build RunExtractRequest with single-string schema.
|
|
|
|
|
req := RunExtractRequest{
|
|
|
|
|
TenantID: tenantID,
|
|
|
|
|
LLMID: b.param.LLMID,
|
2026-07-13 16:57:56 +08:00
|
|
|
ModelName: stagehandModelName(providerName, modelName),
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
BaseURL: baseURL,
|
|
|
|
|
APIKey: apiKey,
|
|
|
|
|
Headless: b.param.Headless,
|
|
|
|
|
Instruction: resolvedPrompts,
|
|
|
|
|
Schema: map[string]any{"type": "string"},
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 4. Dispatch via the runtime's RunExtract.
|
|
|
|
|
invoker := getDefaultStagehandInvoker()
|
|
|
|
|
rawJSON, err := invoker.RunExtract(ctx, req)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
if err != nil {
|
2026-07-13 16:57:56 +08:00
|
|
|
return nil, fmt.Errorf("Browser: stagehand extract (model=%q, base_url=%s): %w",
|
|
|
|
|
req.ModelName, browserBaseURLForLog(req.BaseURL), err)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 5. Unmarshal the JSON-string result to get the plain text.
|
|
|
|
|
var content string
|
|
|
|
|
if err := json.Unmarshal([]byte(rawJSON), &content); err != nil {
|
|
|
|
|
return nil, fmt.Errorf("Browser: unmarshal extract result: %w", err)
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// 6. Build the output map.
|
|
|
|
|
out := map[string]any{
|
|
|
|
|
"content": content,
|
|
|
|
|
"downloaded_files": []map[string]any{},
|
|
|
|
|
"url": "",
|
|
|
|
|
"status": 0,
|
|
|
|
|
"size": len(content),
|
|
|
|
|
"model_id": b.param.LLMID,
|
2026-07-13 16:57:56 +08:00
|
|
|
"prompt": prompts,
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
return out, nil
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
|
2026-07-13 16:57:56 +08:00
|
|
|
func browserBaseURLForLog(raw string) string {
|
|
|
|
|
raw = strings.TrimSpace(raw)
|
|
|
|
|
if raw == "" {
|
|
|
|
|
return "<provider default>"
|
|
|
|
|
}
|
|
|
|
|
u, err := url.Parse(raw)
|
|
|
|
|
if err != nil {
|
|
|
|
|
return "<invalid>"
|
|
|
|
|
}
|
|
|
|
|
u.User = nil
|
|
|
|
|
return u.String()
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func stagehandModelName(providerName, modelName string) string {
|
|
|
|
|
modelName = strings.TrimSpace(modelName)
|
|
|
|
|
if modelName == "" {
|
|
|
|
|
return ""
|
|
|
|
|
}
|
|
|
|
|
if strings.Contains(modelName, "/") {
|
|
|
|
|
prefix, _, _ := strings.Cut(modelName, "/")
|
|
|
|
|
if stagehandNativeProviders[strings.ToLower(strings.TrimSpace(prefix))] != "" {
|
|
|
|
|
return modelName
|
|
|
|
|
}
|
|
|
|
|
return "openai/" + modelName
|
|
|
|
|
}
|
|
|
|
|
providerName = strings.ToLower(strings.TrimSpace(providerName))
|
|
|
|
|
if nativeProvider := stagehandNativeProviders[providerName]; nativeProvider != "" {
|
|
|
|
|
return nativeProvider + "/" + modelName
|
|
|
|
|
}
|
|
|
|
|
return "openai/" + modelName
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Stream mirrors Invoke; Browser is a single-shot generator.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (b *BrowserComponent) Stream(ctx context.Context, inputs map[string]any) (<-chan map[string]any, error) {
|
|
|
|
|
out, err := b.Invoke(ctx, inputs)
|
|
|
|
|
if err != nil {
|
|
|
|
|
return nil, err
|
|
|
|
|
}
|
|
|
|
|
ch := make(chan map[string]any, 1)
|
|
|
|
|
ch <- out
|
|
|
|
|
close(ch)
|
|
|
|
|
return ch, nil
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
// Inputs returns the parameter metadata for tooling.
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func (b *BrowserComponent) Inputs() map[string]string {
|
|
|
|
|
return map[string]string{
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
"llm_id": "Required: tenant model id, e.g. \"deepseek-v4-pro@DeepSeek\". model_id accepted as alias.",
|
|
|
|
|
"prompts": "Required: natural-language extraction task. {sys.query} and other canvas refs are resolved. prompt accepted as alias.",
|
|
|
|
|
"max_steps": "Accepted for fixture compat; ignored at Invoke.",
|
|
|
|
|
"headless": "Browser launch mode (default true).",
|
|
|
|
|
"enable_default_extensions": "Accepted for fixture compat; ignored at Invoke.",
|
|
|
|
|
"chromium_sandbox": "Accepted for fixture compat; ignored at Invoke.",
|
|
|
|
|
"persist_session": "Accepted for fixture compat; ignored at Invoke.",
|
|
|
|
|
"url": "Go-only; not used (kept for backward compat).",
|
|
|
|
|
"timeout": "Go-only; not used (kept for backward compat).",
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2026-07-06 19:15:09 +08:00
|
|
|
func (b *BrowserComponent) GetInputForm() map[string]any {
|
|
|
|
|
return map[string]any{
|
|
|
|
|
"prompts": map[string]any{
|
2026-07-13 16:57:56 +08:00
|
|
|
"type": "paragraph",
|
2026-07-06 19:15:09 +08:00
|
|
|
"name": "Prompts",
|
|
|
|
|
},
|
|
|
|
|
"upload_sources": map[string]any{
|
|
|
|
|
"type": "line",
|
|
|
|
|
"name": "Upload sources",
|
|
|
|
|
},
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
// Outputs returns the response surface.
|
|
|
|
|
func (b *BrowserComponent) Outputs() map[string]string {
|
|
|
|
|
return map[string]string{
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
"content": "Extracted plain text (Sessions.Extract result with schema {\"type\":\"string\"}).",
|
|
|
|
|
"downloaded_files": "Always [] (file download not supported).",
|
|
|
|
|
"url": "Go-native compat key; always \"\".",
|
|
|
|
|
"status": "Go-native compat key; always 0.",
|
|
|
|
|
"size": "Bytes in content.",
|
|
|
|
|
"model_id": "Resolved llm_id (echoed back).",
|
|
|
|
|
"prompt": "Resolved prompts (echoed back).",
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2026-07-13 16:57:56 +08:00
|
|
|
// resolveBrowserLLM resolves the Browser's selected model into the model name
|
|
|
|
|
// and credentials required by the stagehand runtime. It first tries a
|
|
|
|
|
// tenant_model.id lookup, then model@factory parsing via resolveTenantLLM.
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
//
|
|
|
|
|
// Tests override the lookup via `tenantLLMLookupForTest` (a
|
|
|
|
|
// package-level function variable) so they don't need a real DB.
|
|
|
|
|
// Production code leaves the variable unset.
|
2026-07-13 16:57:56 +08:00
|
|
|
func resolveBrowserLLM(tenantID, llmID string) (providerName, modelName, apiKey, baseURL string, err error) {
|
|
|
|
|
if tenantLLMLookupForTest != nil {
|
|
|
|
|
oldModelName, factory := resolveLLMID(llmID)
|
|
|
|
|
apiKey, baseURL, err = tenantLLMLookupForTest(tenantID, oldModelName, factory)
|
|
|
|
|
baseURL = browserOpenAICompatibleBaseURL(baseURL, factory)
|
|
|
|
|
return factory, oldModelName, apiKey, baseURL, err
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
providerName, modelName, apiKey, baseURL, err = resolveTenantModelBrowserLLM(tenantID, llmID)
|
|
|
|
|
if err == nil {
|
|
|
|
|
baseURL = browserOpenAICompatibleBaseURL(baseURL, providerName)
|
|
|
|
|
return providerName, modelName, apiKey, baseURL, nil
|
|
|
|
|
}
|
|
|
|
|
modelErr := err
|
|
|
|
|
|
|
|
|
|
oldModelName, factory := resolveLLMID(llmID)
|
|
|
|
|
apiKey, baseURL, oldErr := resolveTenantLLM(tenantID, oldModelName, factory)
|
|
|
|
|
if oldErr == nil {
|
|
|
|
|
baseURL = browserOpenAICompatibleBaseURL(baseURL, factory)
|
|
|
|
|
return factory, oldModelName, apiKey, baseURL, nil
|
|
|
|
|
}
|
|
|
|
|
return "", "", "", "", fmt.Errorf("tenant_model lookup: %v; tenant_llm fallback: %w", modelErr, oldErr)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func resolveTenantModelBrowserLLM(tenantID, modelID string) (providerName, modelName, apiKey, baseURL string, err error) {
|
|
|
|
|
modelRow, err := dao.NewTenantModelDAO().GetByID(modelID)
|
|
|
|
|
if err != nil {
|
|
|
|
|
if errors.Is(err, gorm.ErrRecordNotFound) {
|
|
|
|
|
return "", "", "", "", err
|
|
|
|
|
}
|
|
|
|
|
return "", "", "", "", fmt.Errorf("tenant model id=%s lookup failed: %w", modelID, err)
|
|
|
|
|
}
|
|
|
|
|
if modelRow.Status != "active" {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("tenant model id=%s is disabled", modelID)
|
|
|
|
|
}
|
|
|
|
|
if !entity.ModelType(modelRow.ModelType).Has(entity.ModelTypeChat) {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("tenant model id=%s cannot be used as %s model", modelID, entity.ModelTypeChat.String())
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
provider, err := dao.NewTenantModelProviderDAO().GetByID(modelRow.ProviderID)
|
|
|
|
|
if err != nil {
|
|
|
|
|
if errors.Is(err, gorm.ErrRecordNotFound) {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("provider id=%s not found for model id=%s", modelRow.ProviderID, modelID)
|
|
|
|
|
}
|
|
|
|
|
return "", "", "", "", err
|
|
|
|
|
}
|
|
|
|
|
if provider == nil {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("provider id=%s not found for model id=%s", modelRow.ProviderID, modelID)
|
|
|
|
|
}
|
|
|
|
|
if provider.TenantID != tenantID {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("tenant %s has no access to provider owned by tenant %s", tenantID, provider.TenantID)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
instance, err := dao.NewTenantModelInstanceDAO().GetByID(modelRow.InstanceID)
|
|
|
|
|
if err != nil {
|
|
|
|
|
if errors.Is(err, gorm.ErrRecordNotFound) {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("instance id=%s not found for model id=%s", modelRow.InstanceID, modelID)
|
|
|
|
|
}
|
|
|
|
|
return "", "", "", "", err
|
|
|
|
|
}
|
|
|
|
|
if instance == nil {
|
|
|
|
|
return "", "", "", "", fmt.Errorf("instance id=%s not found for model id=%s", modelRow.InstanceID, modelID)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
apiKey = instance.APIKey
|
|
|
|
|
if strings.TrimSpace(instance.Extra) != "" {
|
|
|
|
|
var extra map[string]string
|
|
|
|
|
if err := json.Unmarshal([]byte(instance.Extra), &extra); err != nil {
|
|
|
|
|
return "", "", "", "", err
|
|
|
|
|
}
|
|
|
|
|
baseURL = extra["base_url"]
|
|
|
|
|
}
|
|
|
|
|
return provider.ProviderName, modelRow.ModelName, apiKey, baseURL, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func browserOpenAICompatibleBaseURL(baseURL, provider string) string {
|
|
|
|
|
baseURL = strings.TrimSpace(baseURL)
|
|
|
|
|
if baseURL != "" {
|
|
|
|
|
return baseURL
|
|
|
|
|
}
|
|
|
|
|
provider = strings.TrimSpace(provider)
|
|
|
|
|
if provider == "" {
|
|
|
|
|
return ""
|
|
|
|
|
}
|
|
|
|
|
return browserFactoryDefaultBaseURL[strings.ToLower(provider)]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// resolveTenantLLM looks up the legacy tenant_llm config and returns
|
|
|
|
|
// (apiKey, baseURL). baseURL may be empty when the tenant's provider doesn't
|
|
|
|
|
// configure a custom endpoint.
|
feat(agent): align Go agent behavior with Python (except retrieval component) (#16225)
## Summary
Aligns the **Go agent runtime/canvas/components/tools** behavior with
the **Python `agent/` implementation** so the same stored canvas DSL
produces the same execution result on either side. Every component,
tool, and runtime primitive in `internal/agent/` is now driven by the
same semantics as its Python counterpart — variable resolution, template
substitution, control flow, error reporting, retry/cancel, and stream
event shapes.
The **retrieval component is the one explicit exception** in this PR. It
is being reworked in a separate change and is excluded from this
alignment pass; the wrapper slot (`universe_a_wrappers.go →
newRetrievalComponent`) is preserved.
## Scope of alignment
### Components (all aligned with `agent/component/`)
`Begin` · `Message` · `LLM` (incl. ChatTemplateKwargs,
MessageHistoryWindowSize, VisualFiles, Cite, OutputStructure,
JSONOutput, TopP, MaxRetries, DelayAfterError, credentials) · `Agent`
(react + tool artifact capture + `Reset()` interface-assert) · `Switch`
(12/12 operators, Python-equivalent semantics) · `Categorize` · `Invoke`
· `Iteration` · `Loop` (macro-expansion through `workflowx.AddLoopNode`)
· `UserFillUp` (Python-equivalent interrupt/resume via eino
`compose.Interrupt`/`ResumeWithData`) · `FillUp` · `DataOperations` ·
`ListOperations` · `StringTransform` · `VariableAggregator` ·
`VariableAssigner` · `Browser` (full stagehand runtime parity) ·
`DocsGenerator` · `ExcelProcessor`.
### Tools (all aligned with `agent/tools/`)
`Retrieval` (wrapper slot only — logic out of scope) · `MCPToolAdapter`
(streamable-HTTP) · `CodeExec` (sandbox bridge with
`code_exec_contract.go` matching Python contract) · `AkShare` · `ArXiv`
· `Crawler` · `DeepL` · `DuckDuckGo` · `Email` · `ExeSQL` · `GitHub` ·
`Google` · `GoogleScholar` · `Jin10` · `PubMed` · `QWeather` · `SearXNG`
· `Tavily` · `Tushare` · `Wencai` · `Wikipedia` · `YahooFinance` —
uniform `eino tool.InvokableTool` interface, SSRF protection, shared
HTTP client.
### Canvas execution engine (`internal/agent/canvas/`)
Aligned with Python's `agent/canvas.py`:
- **Scheduler** (`scheduler.go`): state pre/post handlers, node lambdas,
per-component timeout resolver (4-level: per-class env → per-class table
→ uniform env → 600s fallback), `legacyNoOpNames`.
- **Loop subgraph** (`loop_subgraph.go`): Python-equivalent
`AddLoopNode` macro expansion + condition translation.
- **Multibranch** (`multibranch.go`): `Switch` / `Categorize` routing
via `compose.NewGraphMultiBranch` — same branch selection semantics as
Python.
- **Parallel subgraph** (`parallel_subgraph.go`): matches Python's
parallel fan-out contract.
- **Interrupt/Resume** (`interrupt_resume.go`): `UserFillUpNodeBody` /
`IsInterruptError` / `ExtractInterruptContexts` — replaces the
deprecated Python sentinel chain with eino's native interrupt API,
preserving the same external behavior.
- **Checkpoint** (`checkpoint_store.go`): `RedisCheckPointStore`
Get/Set/Delete, with business metadata (status / canvas_id /
parent_run_id) on a parallel Redis Hash.
- **RunTracker** (`run_tracker.go`): Start / MarkSucceeded / MarkFailed
/ MarkCancelled / AttachCheckpoint — same lifecycle as the Python run
record.
- **Cancel** (`cancel.go`): Redis pub/sub watch.
- **Stream** (`stream.go`): SSE channel with `messages` / `waiting` /
`errors` / `done` events, same shape as Python's `agent.canvas.RunEvent`
payload.
### DSL bridge (`internal/agent/dsl/`)
- `normalize.go`: v1↔v2 collapsed into a single wire format — Python and
Go consume the same stored JSON.
- `reset.go`: per-run state reset matches Python's `Canvas.reset()`
semantics.
- Testdata mirrors Python's `agent_msg.json` / `all.json` / etc.
### Runtime (`internal/agent/runtime/`)
- `CanvasState` / `NewCanvasState` / `GetVar` / `SetVar` / `ReadVars`:
same `{{cpn_id@param}}` resolution model.
- `ResolveTemplate` (regex fast path + gonja fallback) — Python
Jinja-style semantics.
- `selector.go`, `metrics.go`, `component.go`: shared runtime contracts.
## Out of scope (intentionally)
- **`Retrieval` component logic** — wrapped only; full parity lands in a
follow-up PR.
- **Frontend** — only minor dsl-bridge / canvas UX fixes ride along.
- **CLI / admin / model registry** — orthogonal to agent behavior.
## How alignment is verified
`internal/service/agent_run_e2e_test.go` exercises the **full production
chain** against real Python-shaped DSL fixtures:
```
loadCanvasForUser → versionDAO.GetLatest → decodeCanvasFromDSL →
canvas.Compile → cc.Workflow.Invoke → answer extraction
```
using in-memory SQLite + miniredis (no Docker). Covers:
- `TestRunAgent_RealCanvas_BeginMessage` — happy path, `{{sys.query}}`
resolution
- `TestRunAgent_RealCanvas_WaitForUserResume` — two-run resume cycle
(Python-equivalent)
- `TestRunAgent_RealCanvas_CompileFails` — unknown component name →
sanitized error (Python-equivalent)
- `TestRunAgent_RealCanvas_InvokeFails` — unresolvable template ref
(Python-equivalent)
- `TestRunAgent_RunTracker_AttachCheckpoint_CallSequence` —
Start→AttachCheckpoint→MarkSucceeded lifecycle
`internal/handler/agent_test.go` — SSE streaming parity (`Content-Type:
text/event-stream`, `data: {…}\n\n`, trailing `data: [DONE]\n\n`,
OpenAI-compatible non-stream `choices`).
`internal/agent/canvas/fixture_compile_test.go` + per-component tests
pin the Python-equivalent outputs.
```
go test -count=1 -v -run 'TestRunAgent_RealCanvas|TestRunAgent_RunTracker' ./internal/service/
```
## Design reference
`docs/develop/agent-go-port-design.md` (1329 lines, last cross-checked
2026-06-17) — module layout, per-component / per-tool inventory,
corner-case catalogue, and the actionable backlog (Section 14, including
the retrieval alignment follow-up).
---------
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 11:58:29 +08:00
|
|
|
//
|
|
|
|
|
// TODO(v2): this helper can move to `internal/dao` so the LLM
|
|
|
|
|
// component (`llm.go`) and other future components can share it.
|
|
|
|
|
func resolveTenantLLM(tenantID, modelName, factory string) (apiKey, baseURL string, err error) {
|
|
|
|
|
dao := dao.NewTenantLLMDAO()
|
|
|
|
|
var (
|
|
|
|
|
row *entity.TenantLLM
|
|
|
|
|
)
|
|
|
|
|
if factory != "" {
|
|
|
|
|
row, err = dao.GetByTenantFactoryAndModelName(tenantID, factory, modelName)
|
|
|
|
|
} else {
|
|
|
|
|
// No factory suffix on llm_id; fall back to a single-key
|
|
|
|
|
// lookup (errors if the model is registered under multiple
|
|
|
|
|
// factories — caller must use the explicit form).
|
|
|
|
|
row, err = dao.GetByTenantAndModelName(tenantID, "", modelName)
|
|
|
|
|
}
|
|
|
|
|
if err != nil {
|
|
|
|
|
return "", "", err
|
|
|
|
|
}
|
|
|
|
|
if row == nil {
|
|
|
|
|
return "", "", fmt.Errorf("tenant LLM not found")
|
|
|
|
|
}
|
|
|
|
|
if row.APIKey != nil {
|
|
|
|
|
apiKey = *row.APIKey
|
|
|
|
|
}
|
|
|
|
|
if row.APIBase != nil {
|
|
|
|
|
baseURL = *row.APIBase
|
|
|
|
|
}
|
|
|
|
|
return apiKey, baseURL, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// tenantLLMLookupForTest is the test seam for `resolveTenantLLM`.
|
|
|
|
|
// When non-nil, it's called instead of the real DAO lookup.
|
|
|
|
|
// Production leaves this nil; tests set it via `defer ... = nil`.
|
|
|
|
|
var tenantLLMLookupForTest func(tenantID, modelName, factory string) (apiKey, baseURL string, err error)
|
|
|
|
|
|
feat(agent): Go port — canvas engine, 22 components, DSL v2, 13 endpoints (#15952)
Ports the agent canvas subsystem from Python to Go.
## What's included
### Canvas Engine (Phase 0/1)
- State engine, scheduler, variable resolver, Redis checkpoint store,
cancel protocol
- **209 tests** across canvas / component / io packages
### 22 Components (P0–P4)
| Tier | Components |
|---|---|
| P0 T1+T2+T3 | LLM, Agent, ExitLoop, Switch, Categorize, Begin,
Message, Invoke |
| P1 T3 | VariableAggregator, VariableAssigner, StringTransform,
ListOperations, DataOperations |
| P2 T3 | Iteration, IterationItem, Loop, LoopItem |
| P3 T3 | UserFillUp, Fillup |
| P4 T5 | Browser, ExcelProcessor, DocsGenerator |
### DSL v2 Schema (Phase 2.5)
- Typed v2 in-memory model with v1-to-v2 auto-detect converter
- v1 legacy field stripping per plan §2.11.7
### HTTP Endpoints & Bug Fixes (Plans PR1–PR3)
- **DELETE SQL bug fix**: gorm v2 `Where("id = ?", id).Delete(...)`
pattern
- **CreateAgent validation**: title/DSL required, duplicate check, 103
envelope
- **13 new endpoints**: templates, prompts, tags, sessions CRUD,
chat/completions (SSE + non-stream stubs), rerun, test_db_connection,
logs, webhook/logs
- **756 Go unit tests** (745 → 756, +18)
- **17 → 0 Python integration test failures** (test_agents.py +
test_session_management/)
### Tools
21 eino tools: HTTPHelper, search tools, financial/data tools, mandatory
stubs
### Infrastructure
OTel observability, NATS message queue, DeepDoc gRPC client, SSRF
guards, IDOR mitigation
2026-06-12 22:58:28 +08:00
|
|
|
func init() {
|
|
|
|
|
Register(componentNameBrowser, NewBrowserComponent)
|
|
|
|
|
}
|