LangChain: Tool Call Reliability Fixes
Three core fixes tighten handling of broken and miscounted tool calls. Finalized streamed messages and history filtering now treat invalid calls consistently, and token estimation no longer crashes on plain functions.
Duration: PT2M4S
Episode overview
This episode is a short developer briefing from LangChain.
It explains recent repository work in plain language.
- Show: LangChain
- Published: 2026-10-18T13:02:35Z
- Audio duration: PT2M4S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning, it's October 18th, 2026, and this is your LangChain developer briefing.
The through line today is trust around tool calls — three small core fixes that stop the framework from papering over broken or miscounted tool usage.
First, finalized streamed messages are now honest about truncation. During a stream, lenient parsing makes sense so you can see partial arguments as they arrive. But that same lenient logic was also applied when assembling the final message, healing truncated JSON into a valid-looking call with only half an input.…
That pairs directly with the second fix. The filter messages utility with the exclude tool calls option used to strip valid calls and Anthropic tool use blocks by ID, but it ignored invalid calls, so a broken call you explicitly excluded stayed in history. PR 41293 closes that gap by filtering invalid calls the same…
Finally, a reliability fix for token estimation. The approximate token counting utility documented support for plain Python functions in its tools parameter, but it crashed trying to read a schema attribute that only wrapped tools have. PR 41288 routes plain functions through Open A I tool conversion, leaving the…
What this…
Nearby episodes from LangChain
- State Leak Fixes and Python 3.10 Removal
- Token Counting Fix for Plain Functions
- Weekly Recap - Streaming Efficiency and Agent Control
- Streaming Fixes, Retry Misalignment, and Python 3.10 Removal
- Cheaper Streaming and Safer Agent Recovery
- Per-Call Approvals and Leaner Streaming
- Agent Steering, Shell Reliability, and Content Fixes
- Mid-Conversation Releases and Flexible Summarization