LangChain: Safer Agent Sandboxing and Measurement Fixes
Four independent changes landed: Vetto kernel sandboxing for agent tools, a beta shared decision-model interface, and fixes for splitter start indexes and approximate token counting. The highest impact is safer shell execution without requiring Docker.
Duration: PT2M17S
Episode overview
This episode is a short developer briefing from LangChain.
It explains recent repository work in plain language.
- Show: LangChain
- Published: 2026-10-11T13:01:20Z
- Audio duration: PT2M17S
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 Sunday, October 11th, 2026.
Today's activity doesn't form a single pattern — it's four independent changes — but the highest-impact is safer execution for autonomous agents.
First, containment. Pull request 41195 adds Vetto kernel sandboxing to tools and middleware. The context is familiar: agents running shell commands without process and filesystem isolation risk modifying the host, leaking local credentials and secrets, abusing network sockets, and leaving runaway background…
Second, a shared decision interface. Pull request 41192 introduces a beta base decision model so you don't have to rewrite question definitions when switching providers, in this case between OpenAI Decisions and TypeSafe Jev. It standardizes three question shapes: true-or-false probability, a multiple-choice…
Finally, two quiet accuracy fixes. Pull request 41197 fixes the start index option in the recursive character splitter, which could point to an earlier duplicate when the prior chunk was shorter than the overlap. Offsets are now tracked during splitting and merging, with chunk text unchanged. And pull request 41198…
What's next: test sandbox policies in staging before tightening…
Nearby episodes from LangChain
- Tool Message Fixes and Mid-Conversation Releases
- Weekly Recap - Profiles, Decisions, and Message Reliability
- Mid-Conversation Profiles and Splitter 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