langgraph-store

LangGraph Store memory attribution

This example wraps a LangGraph JS BaseStore style memory surface so each store operation becomes a signed atrib record while the store remains host-owned.

Run it from the repo root:

pnpm --filter @atrib/integration langgraph-store-smoke

What the smoke proves:

• A real @langchain/langgraph@1.3.3 entrypoint can receive an attributed InMemoryStore.
• Workflow code still calls getStore().put, getStore().get, and getStore().search.
• LangGraph routes those workflow calls through BaseStore.batch; public records use langgraph.store.batch.
• The wrapped store returns the same values LangGraph expects.
• Public atrib records carry hashes, operation names, signatures, and chain roots.
• Raw memory text and the underlying put/get/search batch payloads stay in local sidecars, not public records.

What it does not prove yet:

• Python LangGraph store or checkpointer parity.
• A hosted LangGraph Platform deployment.
• A current LangChain or LangGraph review target.

The integration point is intentionally narrow: wrap the store boundary, not the workflow. Existing LangGraph code keeps its storage backend and graph shape.