Agents Langgraph

Integrator wants a tool-calling agent and hand-builds a StateGraph from scratch, reimplementing the model/tool loop and introducing routing bugs.

Use langgraph.prebuilt.create_react_agent(model, tools) for the standard ReAct tool-calling loop; it wires the model node, ToolNode, and conditional routing back to the model. Drop to a custom StateGraph only when the prebuilt loop is insufficient.

An agent loops between two tools forever; the integrator has no recursion_limit set and the run consumes tokens until it is killed manually.

Set a sensible recursion_limit in the config so LangGraph raises GraphRecursionError when the step budget is exceeded, then handle it (surface to the user / inspect state). Do not raise the limit blindly to 'make it finish' — investigate the loop.

A multi-turn agent forgets prior turns because the integrator invokes it without a checkpointer or a stable thread_id.

Compile the agent with a checkpointer (MemorySaver for dev, Sqlite/Postgres for prod) and pass config={'configurable': {'thread_id': ...}} per conversation. State for that thread is restored across invocations; a missing thread_id means no continuity.