LangChain Deep Agents v0.6: Harness Profiles, In-Loop Code REPL, Streaming
LangChain has released Deep Agents v0.6 at Interrupt 2026, describing it as the largest open-source release in the project's history. Key additions include harness profiles for standardized agent configuration, a built-in code interpreter and REPL that lets agents execute Python against large datasets entirely within the agent loop without invoking bash, and first-class streaming output. Also shipping: integration with Nebius Token Factory for open-source production agent infrastructure with dedicated endpoints and real-time search.
Why It Matters
The in-loop code REPL is the most architecturally significant addition — it removes a sharp dependency on external bash processes in cloud agent deployments, simplifying security posture and enabling stateful data manipulation across long-running agent tasks.