## Summary
- LangChain adapter now always uses `randomId()` for message stream IDs
instead of relying on `lc_kwargs.id` (which GoogleGenerativeAI sets to
`"0"` for all messages)
Closes#2929
## Test plan
- [x] `@copilotkit/runtime` tests pass
🤖 Generated with [Claude Code](https://claude.com/claude-code)
## Summary
- Default `max_tokens` was hardcoded to 1024, causing truncated
responses for most use cases
- Increased default from 1024 to 4096 while still respecting
user-provided `maxTokens`
Fixes#2354
## Test plan
- [x] Added test verifying default max_tokens is 4096 when not specified
- [x] Added test verifying user-provided maxTokens is passed through
- [x] All existing anthropic adapter tests pass
## Summary
- Providers like `@ai-sdk/openai-compatible` emit sequential IDs
(`txt-0`, `reasoning-0`, `msg-0`) that are treated as unique but collide
across requests
- The existing check only caught the literal `"0"` — expanded to match
the `^(txt|reasoning|msg)-0$` pattern
- Both `text-start` and `reasoning-start` event handlers now use
`randomUUID()` for these non-unique IDs
## Test plan
- [x] Added `provider-id-collision.test.ts` with 4 tests covering txt-0,
reasoning-0, msg-0, and legitimate ID preservation
- [x] All 1226 runtime tests pass (including existing
config-tools-execution tests)
Fixes#3410, #3623
## Summary
- `onAfterRequest` middleware callback was called with `{}` instead of
actual hook parameters
- Now forwards `threadId`, `runId`, `inputMessages`, `outputMessages`,
and `url` from the v2 runtime's `hookParams`
- The `onBeforeRequest` handler at line 566 already did this correctly;
this fix makes `onAfterRequest` consistent
## Test plan
- [x] Added `on-after-request.test.ts` verifying the callback receives
threadId and runId (not empty object)
- [x] All 1226 runtime tests pass
Fixes#2124
## Summary
Fixes#2450
Token trimming in the Anthropic adapter could remove assistant messages
containing `tool_use` blocks while keeping the corresponding user
messages with `tool_result` blocks. Anthropic's API rejects orphaned
`tool_result` messages. This adds a post-processing step that removes
orphaned `tool_result` blocks after trimming.
## Test plan
- [ ] Verify token trimming preserves tool_use/tool_result pairing
- [ ] Verify orphaned tool_result blocks are removed
- [ ] Verify messages with mixed content (tool_result + text) retain
non-orphaned content
## Summary
Fixes#2405
Both OpenAI and Anthropic adapters used a hardcoded token limit for
message trimming. This adds an optional `maxInputTokens` constructor
parameter to both adapter classes, allowing consumers to override the
default context window limit.
## Test plan
- [ ] Verify default behavior unchanged when `maxInputTokens` not
provided
- [ ] Verify custom `maxInputTokens` is passed through to
`limitMessagesToTokenCount`
## Summary
Fixes#2504
When Anthropic returns a `tool_use` block for a tool not registered in
the current action set, the adapter would crash trying to process it.
This adds a check against known action names and skips unknown tool_use
blocks gracefully.
## Test plan
- [ ] Verify known tools are still processed normally
- [ ] Verify unknown tool_use blocks are skipped without crashing
## Summary
- langchain-mcp-adapters sends tool call result content as an array of
`{type:"text", text:string}` objects
- Extracts and joins text parts so downstream consumers always receive a
plain string
Closes#2922
Closes#1936
`extractParametersFromSchema` now preserves enum values as structured
data on string parameters, recursively extracts nested object
attributes, and converts object arrays to `object[]` type with
attributes. This ensures complex MCP tool schemas survive the
`Parameter[]` -> Zod conversion without losing nested structure.
Split from #3838.
The compatibility layer correctly handles v6 (same API surface as v5),
but the version range ^5.0.0 excluded v6.x. Changed to >=5.0.0 so
users on the latest openai SDK don't get peer dependency warnings.
- Remove all `as any`, `as Function`, and `as Record<string, unknown>` casts
from production code in the OpenAI compatibility layer
- Introduce typed `OpenAIV4Beta` interface and `hasV4BetaChat` type guard
for safe runtime detection of v4 vs v5 clients
- Extract `retrieveThreadRun` and `submitToolOutputsStream` helper functions
into utils.ts with properly typed generic signatures, moving the v4/v5
dispatch logic out of the assistant adapter
- Give `getChatCompletionsForStreaming` an explicit return type so callers
no longer need secondary casts
- Update tests to exercise the new helpers directly instead of
duplicating dispatch logic inline
Cover isOpenAIV5 detection, getChatCompletionsForStreaming dispatch,
and named-path-param calling conventions for runs.retrieve and
submitToolOutputsStream.
OpenAI SDK v5 removed beta.chat (promoted to chat.completions) and
changed multi-path-param methods to use named params (e.g.
runs.retrieve(runId, { thread_id }) instead of positional args).
- Add isOpenAIV5() detection and getChatCompletionsForStreaming() helper
- Migrate OpenAIAdapter to use helper for streaming completions
- Migrate OpenAIAssistantAdapter runs.retrieve() and
submitToolOutputsStream() to branch on SDK version
- Add openai ^4.85.1 || ^5.0.0 to peerDependencies
The V1 CopilotRuntime constructor did `{...endpointAgents, ...agents}`
which silently spread a factory function to `{}`, losing all agents.
Anyone using the V1 API with a factory function got zero agents and no
error. This wraps factory functions so endpoint agents are merged at
resolution time instead of construction time.
- Add `satisfies OnAfterRequestOptions` for compile-time type safety (consistent with onBeforeRequest)
- Replace `(m: any)` casts with type predicates and explicit `as unknown as Message[]`
- Add TODO comment for hardcoded empty properties
- Expand test to verify all 6 fields of OnAfterRequestOptions
- Add edge case tests for undefined messages and missing threadId/runId
Move hasReceivedContent flag inside each content block type handler
so that skipped unknown tool_use blocks do not prevent the fallback
response mechanism from activating. Add tests for unknown tool
skipping behavior.
Cover the MCP adapter scenario where content arrives as an array of
{type:"text", text:string} objects, and verify non-text parts are
filtered out during normalization.
createLogger() was called on every request when debug was enabled,
instantiating a new pino + pino-pretty stream each time. Now creates
a single logger in the runtime constructor and passes it to handlers.
Added test verifying pre-created logger is reused.
Fixes#1979
## Summary
- Add `isOpenAIV5()` detection helper that checks whether `beta.chat`
was removed (v5 promoted it to `chat.completions`)
- Add `getChatCompletionsForStreaming()` helper that routes to the
correct completions namespace per SDK version
- Migrate `OpenAIAdapter` to use the new helper instead of direct
`beta.chat.completions` access
- Migrate `OpenAIAssistantAdapter` `runs.retrieve()` and
`submitToolOutputsStream()` to use v5 named path params (e.g.
`retrieve(runId, { thread_id })` instead of positional
`retrieve(threadId, runId)`)
- Add `openai` to peerDependencies with `^4.85.1 || ^5.0.0` range
- Add 11 unit tests covering v4/v5 detection, streaming dispatch, and
named path param calling conventions
## Notes
The remaining `openai.beta.threads.*` calls (`threads.create`,
`messages.create`, `runs.stream`) do NOT need migration because:
1. `beta.threads` still exists in v5 (only `beta.chat` was removed)
2. These methods have single path params, so their signatures are
unchanged
## Test plan
- [x] All 11 v5 compat tests pass (`nx run @copilotkit/runtime:test`)
- [x] Runtime package builds successfully (`nx run
@copilotkit/runtime:build`)
- [x] Pre-commit hooks pass (lint, format, publint, attw)
## Summary
- The V1 `CopilotRuntime` constructor did `{...endpointAgents,
...agents}` which silently spread a factory function to `{}`, destroying
all agents with no error or warning
- Anyone using the V1 API with a factory function (introduced in #3854)
got zero agents — complete silent data loss
- This wraps factory functions so endpoint agents are merged at
resolution time instead of construction time, matching how V2 already
handles it via `resolveAgents()`
Fixes the V1 path regression from #3854 (per-request agent factory,
issue #2941).
## Test plan
- [x] New test: factory function is preserved through constructor (not
spread to `{}`)
- [x] New test: factory resolves different agents per-request based on
headers
- [x] New test: endpoint agents are correctly merged with
factory-resolved agents
- [x] New test: static agent records still work (backward compat)
- [x] New test: promised agent records still work (backward compat)
- [x] Red-green verified: 4/5 tests fail without the fix, all 5 pass
with it
- [x] Full runtime test suite passes (1234 tests)
- [x] Full build passes
The compatibility layer correctly handles v6 (same API surface as v5),
but the version range ^5.0.0 excluded v6.x. Changed to >=5.0.0 so
users on the latest openai SDK don't get peer dependency warnings.
- Remove all `as any`, `as Function`, and `as Record<string, unknown>` casts
from production code in the OpenAI compatibility layer
- Introduce typed `OpenAIV4Beta` interface and `hasV4BetaChat` type guard
for safe runtime detection of v4 vs v5 clients
- Extract `retrieveThreadRun` and `submitToolOutputsStream` helper functions
into utils.ts with properly typed generic signatures, moving the v4/v5
dispatch logic out of the assistant adapter
- Give `getChatCompletionsForStreaming` an explicit return type so callers
no longer need secondary casts
- Update tests to exercise the new helpers directly instead of
duplicating dispatch logic inline
- Count all emitted events unconditionally (eventCount), track separately
how many were logged (loggedEventCount) — completion log shows both
- Remove dead `?? undefined` on runtime.debug (non-optional type)
- Update test to verify eventCount reflects actual events; add test with
events: true to verify loggedEventCount
## Summary
- Widen `TanStackChatMessage.content` from `TanStackContentPart[]`
to `any[]` so messages from `convertInputToTanStackAI` are directly
passable to any TanStack AI adapter without `as any` casts
- Split `TanStackContentPart` into a proper discriminated union with
separate variants per modality
- Add `env.d.ts` with `vite/client` reference for CSS import types
- Fix `onError` callback shape in the example
## Test plan
- [x] All 17 multimodal TanStack tests pass
- [x] All 294 agent tests pass
- [ ] Verify no TS errors in react-router example IDE
🤖 Generated with [Claude Code](https://claude.com/claude-code)
- New ThreadStoreRegistry suite (8 tests) covers register/get round-trip,
replacement on duplicate id, unregister no-op, and subscriber events for
both register and unregister
- handle-threads suite gains handleClearThreads (InMemory path + intelligence
path) and handleGetThreadMessages (InMemory, unknown thread, intelligence
delegation, 422 fallback) describe blocks
- in-memory-runner suite: clearThreads() in beforeEach fixes GLOBAL_STORE
isolation; vacuous empty-array test replaced with a meaningful post-clear
assertion
- use-threads suite: new test asserts registerThreadStore is called on mount
and unregisterThreadStore is called on unmount
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- CopilotKitCore gains a ThreadStoreRegistry (register/unregister by agentId)
and a new onAgentRunStarted subscriber event so the inspector can subscribe
before agent.runAgent() snapshots the subscriber list
- Runtime gains handleListThreads, handleUpdateThread, handleArchiveThread,
handleDeleteThread, handleSubscribeToThreads, and handleGetThreadMessages
handlers; all mutations are authenticated via identifyUser (request body
userId is ignored)
- InMemoryAgentRunner now stores thread history for the local-dev fallback
path; debug console.log removed; InMemoryThread uses literal types for
constant-value fields (organizationId: "", createdById: "", archived: false)
- useThreads hook registers its store with CopilotKitCore on mount and
unregisters on unmount so the inspector can read thread state
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Cover isOpenAIV5 detection, getChatCompletionsForStreaming dispatch,
and named-path-param calling conventions for runs.retrieve and
submitToolOutputsStream.
OpenAI SDK v5 removed beta.chat (promoted to chat.completions) and
changed multi-path-param methods to use named params (e.g.
runs.retrieve(runId, { thread_id }) instead of positional args).
- Add isOpenAIV5() detection and getChatCompletionsForStreaming() helper
- Migrate OpenAIAdapter to use helper for streaming completions
- Migrate OpenAIAssistantAdapter runs.retrieve() and
submitToolOutputsStream() to branch on SDK version
- Add openai ^4.85.1 || ^5.0.0 to peerDependencies
The V1 CopilotRuntime constructor did `{...endpointAgents, ...agents}`
which silently spread a factory function to `{}`, losing all agents.
Anyone using the V1 API with a factory function got zero agents and no
error. This wraps factory functions so endpoint agents are merged at
resolution time instead of construction time.
- Restore getCapabilities() return type to Promise<AgentCapabilities>
with ?? {} fallback, honoring AbstractAgent base class contract
- Simplify type annotation in get-runtime-info.ts to explicit
AgentCapabilities | undefined instead of complex Awaited<ReturnType<...>>
- Add console.warn logging to per-agent capabilities error catch block
- Fix useCapabilities JSDoc to accurately describe behavior during
runtime handshake
- Add test for empty capabilities object {} (truthy, included in response)
- Update error isolation test to verify warning is logged
- Add per-agent error isolation in /info handler so a single
getCapabilities() failure doesn't 500 the whole endpoint
- Normalize getCapabilities() to return undefined (not {}) when no
capabilities are set, matching the sync getter
- Replace instanceof ProxiedCopilotRuntimeAgent with duck-type check
in useCapabilities hook for extensibility
- Add JSDoc warning about shallow-merge behavior on capabilities config
- Add useCapabilities hook tests (5 cases covering both branches)
- Add per-agent error isolation test for get-runtime-info
Alem Tuzlak