deepagent

DeepAgents let you build configurable AI agents with integrated tools within LangChain workflows. You instantiate a DeepAgent with a model, attach tool functions via decorators or modules, and invoke it with user messages. This enables agents to call tools dynamically and return results directly in your LangChain pipelines.

Yes. DeepAgents integrate with LangGraph for building sophisticated multi-step agent workflows. You define tools and attach them to your agent, then compose the agent into LangGraph state machines. This supports streaming, async invocation, and orchestration across code, research, and admin tasks.

DeepAgents support two patterns: decorated functions and module-based tools. Define tool functions with decorators that describe their purpose and parameters, or import tool modules and bind them to your agent instance. Both methods register tools for the agent to discover and invoke automatically.

DeepAgents expose multiple invocation patterns: invoke for synchronous calls, stream for iterative response output, and astream for async streaming. Pass a user message to any method and the agent executes tools as needed, returning results incrementally or as a complete response.

DeepAgents accept customizable models and system prompts, so you can swap in your preferred LLM provider or fine-tuned model. Instantiate the agent with your model instance and configure the system prompt to shape the agent's behavior and reasoning style.

DeepAgents documentation covers best practices for logging tool invocations, validating tool outputs, and testing agent behavior. Implement input validation on tools, monitor execution logs, and use unit tests to verify tool registration and agent responses under known inputs.