LangChain Ships deepagents-cli Open-Model Profiles, Fleet Routing, and LangGraph 1.2
LangChain released a coordinated suite of production agent infrastructure: deepagents-cli gains a profiles API that bundles per-model prompts, middleware, and settings so open-weight models (Kimi, Qwen, GLM) can drive agent loops at closed-frontier quality; LangSmith Fleet adds per-sub-agent model routing so simple tasks hit fast/cheap models and complex steps use stronger ones; LangGraph 1.2 alpha ships node-level error handlers, static and dynamic timeouts, delta channels for optimized storage, new streaming primitives, and live agent update support without breaking running threads. Community tool Promptloop (built on deepagents) automates the full prompt-eval loop in a single chat session.
Why It Matters
Together, these releases form a complete open-model orchestration stack — harness, routing, runtime, and evaluation — significantly lowering the cost of running production multi-agent systems on open-weight models versus closed APIs.