An engineer who shipped a customer-facing LangGraph agent published a beta book, 'LangGraph in Small Pieces: Building AI Agents from Blocks to Production', freely readable at lgisp.ai. It teaches LangGraph fundamentals first, then production agent concerns like tool calling, MCP, RAG, memory, evaluation, serving, and guardrails. All transcripts come from real model runs, with a local model as the default setup so readers can follow along without API costs.
A free online book, 'LangGraph in Small Pieces: Building AI Agents from Blocks to Production', walks readers from LangGraph fundamentals through production deployment. It distinguishes itself by using real, unedited transcripts from local models and ending each chapter with graded challenges and worked solutions.
A free online book, 'LangGraph in Small Pieces: Building AI Agents from Blocks to Production', walks readers from LangGraph fundamentals through production deployment. It distinguishes itself by using real, unedited transcripts from local models and ending each chapter with graded challenges and worked solutions.
A developer released Agnos, an open-source (MIT) infrastructure layer that sits under LangGraph agents to handle model fallbacks, per-agent spend caps, key management, and audit trails. The post frames LangGraph as strong at the agent loop but lacking production concerns, and seeks community critique of the architecture.
A developer released Agnos, an open-source (MIT) infrastructure layer that sits under LangGraph agents to handle model fallbacks, per-agent spend caps, key management, and audit trails. The post frames LangGraph as strong at the agent loop but lacking production concerns, and seeks community critique of the architecture.
FreshCtx is an Apache-2.0 Python library that adds a validation step immediately before protected tools or nodes execute in agent workflows. It lets developers declare which external evidence a tool depends on so resumed LangGraph runs fail or re-check when that evidence is stale. It targets agent builders whose LLM decisions were sound when made but act on changed external state by execution time.
FreshCtx is an Apache-2.0 Python library that adds a validation step immediately before protected tools or nodes execute in agent workflows. It lets developers declare which external evidence a tool depends on so resumed LangGraph runs fail or re-check when that evidence is stale. It targets agent builders whose LLM decisions were sound when made but act on changed external state by execution time.
langhost, an independent open-source project, released a self-hosted LangGraph Agent Server that uses production Postgres and Redis, offering the same SDK and Studio as LangGraph with zero code changes. It aims to provide a free alternative to LangChain's paid offerings, and the project is not affiliated with or endorsed by LangChain.
langhost, an independent open-source project, released a self-hosted LangGraph Agent Server that uses production Postgres and Redis, offering the same SDK and Studio as LangGraph with zero code changes. It aims to provide a free alternative to LangChain's paid offerings, and the project is not affiliated with or endorsed by LangChain.
An analysis of langgraph-supervisor source shows the supervisor runs after every worker handoff, re-reading the full shared message history and prompt each time. One production team found their supervisor consumed as many monthly tokens as all 8 of its Qwen3-14B specialist agents combined. The post offers concrete fixes, though LangChain has archived the repo and now recommends the supervisor pattern via tools.
An analysis of langgraph-supervisor source shows the supervisor runs after every worker handoff, re-reading the full shared message history and prompt each time. One production team found their supervisor consumed as many monthly tokens as all 8 of its Qwen3-14B specialist agents combined. The post offers concrete fixes, though LangChain has archived the repo and now recommends the supervisor pattern via tools.