mirror of https://github.com/langchain-ai/langgraphjs.git synced 2026-07-22 00:55:26 -04:00

Files

T

Christian Bromann 3529e3831a fix(sdk): align AssembledToolCall typing with pre-v1 expectations (#2421 )

## Summary
- Split tool-call handles by consumer:
- **Client SDK** (`ThreadStream.toolCalls`, subgraph/subagent
projections): `ClientAssembledToolCall` with a promise-only `output`
(resolves on success, rejects on error). Still exported as
`AssembledToolCall` from `@langchain/langgraph-sdk/client` for script
usage.
- **Framework SDKs** (`stream.toolCalls`, `useToolCalls`,
`injectToolCalls`): `AssembledToolCall` with plain reactive fields —
`output: T | null`, `status`, and `error` — updated in place as events
arrive so React/Vue/Svelte/Angular can render from snapshots without
`await`, effects, or Suspense around promises.
- Add generic `AssembledToolCall<TName, TInput, TOutput>` plus
`id`/`args` aliases; point `InferToolCalls` at assembled streaming
handles and add `AssembledToolCallFromTool` (exported as
`ToolCallFromTool` from `@langchain/react`, `@langchain/vue`,
`@langchain/svelte`, and `@langchain/angular`).
- Rework `ToolCallAssembler` around a mutable internal handle and
`toClientAssembledToolCall()` for client projections; framework stores
the reactive handle directly.
- Remove redundant `InferAssembledToolCalls` and deprecated `StateOf`;
wire typed `toolCalls` / selector generics across all four framework
packages.
- Expand and align `createAgent`, `createDeepAgent`, and `langgraph`
type tests across React, Vue, Svelte, and Angular; update examples,
protocol-v2 integration tests, and Vue migration docs.

2026-05-22 06:38:06 -07:00