### What problem does this PR solve?
Adds first-class support for **Mistral OCR** (`POST /v1/ocr`) as a
document parser, and fixes the long-standing bug where selecting
`mistral-ocr-latest` fails with `Can't find model for
<tenant>/image2text/mistral-ocr-latest`.
`mistral-ocr-latest` is Mistral's dedicated document-OCR endpoint, not a
vision-chat (`image2text`) model, but the catalog tagged it `image2text`
— so it resolved to the `CvModel` registry, which has no `Mistral`
entry, and there was no `OcrModel` entry either. This PR registers it
correctly and wires it end to end.
Closes#17056Closes#5782Closes#7075
**What it does**
1. **`MistralParser` + `MistralOcrModel`**
(`deepdoc/parser/mistral_parser.py`, `rag/llm/ocr_model.py`) — a proper
`OcrModel` factory `Mistral OCR`, mirroring the SoMark cloud-OCR
template. Tables stay inline as HTML; the page range maps to Mistral's
native `pages` selector (absolute page indices, billed per selected
page, so multi-task documents do not re-OCR the whole file); documents
over the inline limit go through the `/v1/files` signed-URL flow with
cleanup.
2. **Removes the `image2text` mis-tag** for `mistral-ocr-latest` from
the `Mistral` factory in `conf/llm_factories.json` (it now lives only in
the `Mistral OCR` factory, typed `ocr`). This is what closes the `Can't
find model` path.
3. **`MistralCV`** (`rag/llm/cv_model.py`) — a thin `GptV4` subclass
over Mistral's OpenAI-compatible endpoint, registering a `Mistral` entry
in the `CvModel` registry so Mistral vision models (`pixtral-*`) become
usable as `image2text` at all.
4. **Figure description** — Mistral-OCR-extracted figures are captioned
using the tenant's configured `image2text` model (any provider),
matching MinerU/deepdoc behaviour.
5. **Wires the parser into every chunking method** (`naive`, `paper`,
`book`, `laws`, `manual`, `one`, `presentation`) and the `rag/flow` DAG
path. This also fixes a related latent gap where those chunkers
forwarded only `mineru_llm_name`, so any model-based OCR provider
selected on a non-`naive` method silently fell through.
**Notes on the API contract** (verified against the live Mistral API):
`pages` is a selector (returns absolute `index`, bills only the
requested pages); `include_blocks: true` returns per-block bounding
boxes usable for chunk highlighting and figure cropping; large files use
`POST /v1/files` → signed URL → OCR → `DELETE`.
**Testing**: new unit tests cover the response→sections contract (both
the 2-tuple `naive` path and the typed 3-tuple DAG path), the
position-tag rescale, the HTTP client incl. upload failure/cleanup
paths, `parse_pdf` page-range threading, registry registration, env
config, the suffix normalization, the factory catalog entry, `MistralCV`
registration, and figure-description injection. Verified end to end
against the live Mistral API on real PDFs (table extraction,
page-selector cost avoidance, figure captioning).
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
- Clear stale pipeline IDs and generated data when updating documents
without `pipeline_id`.
- Support tree compilation results in pipeline workflows.
- Update compilation templates in place while preserving existing
template IDs.
- Improve duplicate-template validation messages.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
Co-authored-by: Jin Hai <haijin.chn@gmail.com>
### Summary
Handle searching dataset without embedding model
In this PR, Searching datasets with different embedding models or
searching dataset with/without embedding models are not allowed. We will
improve the behavior later.
### What problem does this PR solve?
Closes#12962
MCPToolCallSessions created during agent execution (in `Agent.__init__`)
are never explicitly closed. Each session starts its own event loop
thread and opens an SSE/HTTP connection to the MCP server. When the
canvas goes out of scope, these threads and connections remain alive
indefinitely, accumulating over time and causing resource exhaustion
after prolonged use.
### Solution
1. Add a `Graph.close()` method that iterates all components, finds
MCPToolCallSessions held by Agent tools, and calls `close_sync()` on
each to properly shut down the event loop, thread, and connection.
2. Call `canvas.close()` in `finally` blocks after `canvas.run()`
completes in `canvas_service.py` and `canvas_app.py`.
3. Move MCP session cleanup to `finally` blocks in `test_tool` endpoint
(`mcp_server_app.py`) and `get_mcp_tools` (`api_utils.py`) to ensure
sessions are closed even on exceptions.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: conflict-resolver <conflict-resolver@local>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
### What problem does this PR solve?
_Briefly describe what this PR aims to solve. Include background context
that will help reviewers understand the purpose of the PR._
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
- [x] Other (please describe):
## Summary
Agent (Canvas) runs previously did not surface token usage in the SSE
stream, and RAGFlow's own Langfuse generations for agent runs were
missing the prompt/completion split and the session/user correlation.
This made it impossible for an external caller (or Langfuse) to
reconcile an agent turn's cost with the upstream provider (e.g.
OpenRouter), because a single turn can issue several distinct LLM calls
(query rewriting / cross-language translation, multi-round tool
reasoning, nested sub-agents, and the final answer).
This PR introduces a per-run token usage sink so that **every** LLM call
in a run is aggregated and reported once, and enriches Langfuse
generations with the prompt/completion split plus session/user
attributes.
## What changes
### 1. Per-run token usage sink (`common/token_utils.py`)
- Adds two `contextvars`: `token_usage_sink` (a mutable per-run
accumulator) and `langfuse_run_attrs` (session_id/user_id for the run).
- Adds `record_run_token_usage(...)` (thread-safe via a lock, because
`thread_pool_exec` copies the context into worker threads that share the
sink dict) and `usage_from_response(...)` which extracts a
`{prompt_tokens, completion_tokens, total_tokens}` split from
OpenAI/OpenRouter-style responses.
### 2. Provider layer captures the prompt/completion split
(`rag/llm/chat_model.py`)
- `LiteLLMBase` and `Base` now store `self.last_usage`
(prompt/completion/total) for the most recent chat call, in both the
plain and tool-calling paths.
- Streaming requests set `stream_options.include_usage = True` (LiteLLM
path) so the authoritative usage arrives on the final chunk; this is
read even on the usage-only chunk that carries no `choices`.
- Fixes a multi-round accounting bug in `*_with_tools`: token totals
were **overwritten** by each round (`total_tokens = tol`) instead of
accumulated, undercounting multi-round tool conversations. Each round is
now committed to a running aggregate.
### 3. LLMBundle reports usage once, per call
(`api/db/services/llm_service.py`)
- New `_report_usage(total_tokens)` records the call's usage into the
active run sink and returns the prompt/completion/total split for
Langfuse. The split is only used when it is consistent with the
authoritative total; otherwise only the total is reported.
- All three chat entry points (`async_chat`, `async_chat_streamly`,
`async_chat_streamly_delta`) now emit `usage_details` with
`input`/`output`/`total` instead of total-only.
- `_start_langfuse_observation` now applies `session_id`/`user_id` from
the per-run context (`langfuse_run_attrs`) so agent-run generations are
correctly grouped, even though agent LLMBundles are constructed without
those attributes.
### 4. Canvas installs the sink and emits the aggregate
(`agent/canvas.py`)
- `Canvas.run()` installs a fresh `token_usage_sink` and
`langfuse_run_attrs` (from `user_id`/`session_id`) at the start of every
turn.
- `message_end` now includes an aggregated `usage` object:
`{prompt_tokens, completion_tokens, total_tokens, calls}` covering all
LLM calls in the run.
### 5. Pass session id into the run
(`api/db/services/canvas_service.py`)
- `completion()` forwards `session_id` to `Canvas.run()` for Langfuse
session correlation.
## Why a context variable
LLM calls in an agent run originate from many places that each build
their own `LLMBundle` (e.g. `cross_languages`/`keyword_extraction`
helpers, the Agent component, and nested sub-agents invoked as tools). A
run-scoped context variable is the only non-invasive chokepoint that
captures all of them exactly once, including nested agents (which run in
the same async context) and thread-pool tools (the executor copies the
context).
## Behavior / compatibility
- No public API or wire-format removal: `message_end` gains an
additional optional `usage` field; existing consumers are unaffected.
- When a provider does not return authoritative usage, behavior falls
back to the previous token estimate (total only, no split).
- Non-agent flows (Dataflow `Pipeline`, sync `Graph.run`) are untouched.
## Testing
- [x] Simple agent answer: `message_end.usage.total_tokens` matches
provider usage.
- [x] Agent with cross-language retrieval: aggregate equals the sum of
both provider calls.
- [x] Tool-calling agent (multi-round): total accumulates across rounds.
- [x] Nested agent (agent-as-tool): sub-agent tokens included in the
parent run total.
- [x] Langfuse: agent generations show input/output split and are
grouped by session/user.
---------
Co-authored-by: yzc <yuzhichang@gmail.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
## Summary
- Treat `max_tokens=0` as unset (`or 8192`) when building model context
budgets, fixing agents that silently zeroed prompts when a vLLM model
had `max_tokens: 0` in tenant config
- Replace trailing same-role canvas history in `LLM._sys_prompt_and_msg`
instead of skipping the current user prompt
- Add `LLM.fit_messages()` validation after `message_fit_in` on agent
paths so empty user content fails fast with a clear error instead of
reaching vLLM
Fixes#16411
## Root cause
Agent canvas workflow called `message_fit_in` with `int(max_length *
0.97)`. When `max_length` was `0`, both system and user content were
trimmed to empty strings. The `[HISTORY STREAMLY]` log showing only
`{"role":"user","content":""}` matches this. A secondary bug skipped
appending the formatted user prompt when history ended with a `user`
role message.
## Test plan
- [x] Added `test/unit_test/agent/component/test_llm_prompt.py` for
role-replace, validation, and zero-budget fitting
- [x] Added
`test_message_fit_in_zero_budget_preserves_non_empty_messages` in
`test_generator_message_fit_in.py`
- [ ] CI unit tests
- [ ] Manual: agent canvas `begin → Retrieval → Agent → Message` with
vLLM Qwen3; confirm user message reaches LLM
Made with [Cursor](https://cursor.com)
---------
Co-authored-by: Taranum Wasu <taranumwasu@Taranums-MacBook-Pro.local>
Co-authored-by: Cursor <cursoragent@cursor.com>
### What problem does this PR solve?
GET /api/v1/agents (list_agents) already supports filtering by
canvas_category, keywords, tags, and owner_ids, but it does not support
canvas_type — even though canvas_type is a persisted field on UserCanvas
and is already accepted on agent create/update APIs.
This gap causes two issues:
Filtering — clients cannot list agents by business category (e.g.
Marketing, Agent, Ingestion Pipeline) without fetching all agents and
filtering client-side.
Response payload — list_agents did not return canvas_type in each canvas
item, so consumers had to call GET /api/v1/agents/{id} per agent to read
it.
This PR adds optional canvas_type query parameter support and includes
canvas_type in the list response.
### Type of change
- [√] Bug Fix (non-breaking change which fixes an issue)
- [ ] New Feature (non-breaking change which adds functionality)
- [ ] Documentation Update
- [ ] Refactoring
- [ ] Performance Improvement
- [ ] Other (please describe):
