Skip to main content
Use the LangGraph adapter for stateful graph-based agents.

​
Manifest

runtime:
  adapter:
    type: "langgraph"
    entry_point: "src.main:graph"

​
Runtime Behavior

Charm loads the compiled graph and invokes it with the input payload from the Store. Thread or state behavior depends on how your graph is configured.

​
Best For

  • Multi-step workflows.
  • Graph-based agent control flow.
  • Stateful assistants.
  • Agents with conditional routing.

​
State and Auto-Checkpointing

If your graph requires durable state, the Charm SDK provides automatic checkpointing when used with an external memory provider. When you configure runtime.memory.provider: supabase, the platform automatically injects a CharmSupabaseCheckpointer into your LangGraph execution. This eliminates the need for custom database logic—every step of your graph’s state is transparently serialized and persisted. In the event of a system upgrade, network disconnect, or runner restart, the next invocation automatically rehydrates the latest graph state to resume the asynchronous workflow seamlessly, ensuring strict fault tolerance.

​
Tips

  • Keep graph inputs aligned with interface.input.
  • Validate state shape before publishing.
  • Use interactive lifecycle for chat UX and daemon lifecycle for persistent background behavior.