FrameworksOfficial

AG-UI LangGraph 0.0.46 reads native interrupt schemas, unwraps serialized tool results, and declares protocol version at run start

AG-UI released @ag-ui/langgraph 0.0.46 on November 13. The adapter reads Interrupt.response_schema on LangGraph 1.2.12 and later while retaining value-level fallbacks. It also unwraps outputs from serialized LangChain ToolMessage kwargs, skips tool-end events without a call ID, and declares protocolVersion on RUN_STARTED.

Why it matters

Human confirmation, tool results, and recovery surfaces depend on frontend and backend agreeing on response structure, call identity, and protocol version. Native interrupt schemas let the host collect answers with the intended controls, while correct ToolMessage decoding and explicit versioning reduce lost results, malformed events, and version guessing.

A closer look

Version 0.0.46 prefers LangGraph's native Interrupt.response_schema and falls back to responseSchema or response_schema. On tool end it unwraps LangChain-serialized ToolMessage outputs under kwargs; outputs without a tool-call ID are skipped instead of throwing or emitting an undefined ID. The release also refreshes Gemini thinking demos and raises @ag-ui/client and related peer dependencies to at least 1.0.0.

Who this is for

Frontend, agent-platform, and integration teams using AG-UI with LangGraph, LangChain interrupts, tools, or human confirmation.

Original sources

Revisions & corrections (1)

— Backfilled a missed evidence-complete official adapter release using its true November 13 event date; documented the peer-dependency break and the limits of event-transport correctness.

Related events

Frameworks

assistant-ui's November 13 releases extend durable AG-UI across history ownership, concurrent threads, and cross-worker cancellation

assistant-ui released @assistant-ui/react-ag-ui 0.1.10 and Python assistant-stream 0.0.38 on November 13. Frontend fixes cover persistence of snapshot-selected answers, joined-step ordering, unique tool-call ownership, and streamless resume. The server adds compare-and-set thread claims, per-run context, SSE keepalive, and cross-worker cancellation.