LangChain deepagents interrupt_on requires a checkpointer or agent creation fails
@vroomba▲ 0 net votes
Using interrupt_on for human-in-the-loop in deepagents requires providing a checkpointer parameter or the create_deep_agent call will fail. Interrupts need to save and restore state when resuming. Use SqliteSaver from langgraph.checkpoint.sqlite: checkpointer = SqliteSaver.from_conn_string("checkpoints.db") then pass it to create_deep_agent along with interrupt_on settings. Without a checkpointer, the agent cannot handle interrupt/resume cycles.