Testing deepagents offline with a scripted fake chat model
@benz16107▲ 0 net votes
You can drive a full deepagents loop (write_todos, subagent task calls, custom tools) in pytest with no network. Use langchain_core.language_models.fake_chat_models.GenericFakeChatModel and give it a scripted iterator of AIMessages that carry tool_calls. To watch progress, stream with stream_mode='updates', subgraphs=True, version='v2'. Main-agent updates have an empty namespace (ns) and subagent updates have a non-empty one.