LangGraph / deepagents agent.invoke() raises "ValueError: Checkpointer requires one or more of the following 'configurable' keys: thread_id, checkpoint_ns, checkpoint_id" after adding a checkpointer
Once a graph has a checkpointer, every call must say which thread it belongs to: agent.invoke({"messages": [...]}, {"configurable": {"thread_id": "session-1"}}) (same for .stream(), .ainvoke(), get_state()). Reuse the same thread_id to continue a conversation, use a new one for a fresh run.
This bites deepagents users in particular because create_deep_agent(..., checkpointer=...) is often added only to satisfy interrupt_on / human-in-the-loop, and the existing test calls like agent.invoke({"messages": msgs}) then all fail. Reproduced on deepagents 0.7.38 + langgraph 1.2.22 with SqliteSaver(sqlite3.connect(..., check_same_thread=False)): invoke without config -> ValueError: Checkpointer requires one or more of the following 'configurable' keys: thread_id, checkpoint_ns, checkpoint_id; with {"configurable": {"thread_id": "t"}} the call proceeds to the model.
Debugging tip: if your harness formats exceptions as str(e)[:50], this shows up as just "Checkpointer requires one or more of the following", which reads like a checkpointer setup problem. Print the full message (or repr(e)) in smoke tests.