Fixed Bedrock reasoning block handling to normalize and safely replay reasoning blocks, preserving signatures and omitting incomplete signature-only reasoning during replay.
LangChain.js
Bedrock reasoning blocks are now emitted with standard reasoning block signatures, incomplete signature-only blocks are omitted during replay, and compatibility with legacy and redacted Bedrock reasoning is retained.
Patch Changes
Fixed chat completions API calls by filtering out content blocks that the OpenAI API rejects as input, including tool_call content blocks.
Patch Changes
Patch Changes
Patch Changes
- Updated dependencies []:
- @langchain/classic@1.0.40
Patch Changes
- Updated dependencies [
988ca7d]:- @langchain/openai@1.5.4
Fixed OpenAI integration to emit output_text for assistant content in responses input.
Patch Changes
- Updated dependencies []:
- @langchain/classic@1.0.39
Patch Changes
- Updated dependencies [
988ca7d]:- @langchain/openai@1.5.4
Patch Changes
- Updated dependencies [
988ca7d]:- @langchain/openai@1.5.4
Patch Changes
- Updated dependencies [
988ca7d]:- @langchain/openai@1.5.4
Fixed a crash when constructing an AIMessage with response_metadata.output_version === "v1" and string content, which prevented contentBlocks.push operations and occurred frequently in serialized LangGraph stream payloads.
Patch Changes
- Updated dependencies [
988ca7d]:- @langchain/openai@1.5.4
Fixed an issue where malformed tool input schemas were unrecoverable.


