LangChain: Mid-Conversation Profiles and Splitter Fixes
Model capability checks move from hard-coded names into model profiles, while parallel fixes target text splitter correctness and runnable retry reliability.
Duration: PT2M12S
Episode overview
This episode is a short developer briefing from LangChain.
It explains recent repository work in plain language.
- Show: LangChain
- Published: 2026-10-10T13:02:06Z
- Audio duration: PT2M12S
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 Saturday, October 10th, 2026, and this is your LangChain daily briefing.
The lead story is a shift in how chat models declare what they can do mid-conversation, moving the logic out of model-name lists and into model profiles.
That change landed in PR 41180, also recorded as commit b060a74. The model profile gains two flags for mid-conversation system messages and mid-conversation tools. In practice, Chat Anthropic now checks its profile instead of matching model names, which previously missed Bedrock model IDs and forced downstream…
Second theme: text splitter correctness. Three reports cluster here. PR 41189 fixes the HTML semantic splitter mistaking user text that looks like its internal preserved marker, which could silently replace content and duplicate preserved text. Separately, PRs 41187 and 41188 both address the same issue, where…
Third: core execution reliability. PR 41178 reports batch with retry can return the wrong result for permanently failing inputs, handing a recovered value to a still-failing slot with no exception. PR 41181 would let runnables with fallbacks accept non-dictionary inputs like strings or lists when an exception key is…
What's…
Nearby episodes from LangChain
- Tool Message Fixes and Mid-Conversation Releases
- Weekly Recap - Profiles, Decisions, and Message Reliability
- Safer Agent Sandboxing and Measurement Fixes
- Decisions API Beta and Agent Reliability Fixes
- Reasoning Fixes and Decisions API Beta
- Streaming Fidelity and Core Edge Cases
- Sharper Agent Guardrails and Credential Hardening
- Credential Leak Fixes and Required OpenAI Model