DATE:
AUTHOR:
The LangChain Team
LangGraph

LangGraph Workflow Updates (Python & JS)

DATE:
AUTHOR: The LangChain Team

We’ve rolled out a batch of new features across both LangGraph.js and LangGraph for Python — giving you faster development cycles, more efficient executions, and greater control over your agentic workflows.


Workflow Engine Improvements (Python & JS)

  • Node Caching
    Cache the results of individual nodes to skip redundant computation and speed up runs — especially useful during iterative development.
    Python docs | JS docs

  • Deferred Nodes
    Delay execution until all upstream paths complete — perfect for map-reduce, consensus, and collaborative agent workflows.
    Python docs | JS docs


Prebuilt ReAct Agent Upgrades

  • Pre/Post Model Hooks
    Add custom logic before or after model calls. Great for controlling context bloat (pre) and inserting guardrails or human-in-the-loop checks (post).
    Python docs | JS docs

  • Built-in Provider Tools
    Use web search, RemoteMCP, and more out of the box by simply adding a tool spec dict to your ReAct agent.


LangGraph.js Specific Enhancements

  • Resumable Streams
    Streams now resume automatically after reloads or connection drops with reconnectOnMount.
    JS docs

  • Dev Experience Improvements

    • .stream() is now fully type-safe.

    • .addNode({ ... }) and .addSequence({ ... }) make building workflows cleaner.

    • .invoke() and "values" stream modes now return interrupts directly.

Powered by LaunchNotes