LangChain fixes tool-call replay and subagent identity
Two agent-library patches address how tool calls and interrupted subagents are restored. Retrospective brief for 15:00–15:59 UTC on 29 September 2026.
Top AI stories from the last hour
Copy MarkdownThe Anthropic integration preserves invalid tool calls on replay
The langchain-anthropic 1.7.5 patch serializes invalid tool calls as tool-use blocks when replaying a conversation. This changes how the integration reconstructs tool-call history for Anthropic model requests. The same release includes Sonnet 5.5 compatibility, already noted in the preceding core release.
The QuickJS integration preserves subagent identity through interrupts
The langchain-quickjs 0.3.8 patch preserves subagent identity when replaying interrupts. Agent applications that resume interrupted execution need the replay to refer to the same subagent; this release explicitly addresses that behavior.