Deep Agents is an open-source agent harness from the LangChain ecosystem designed for long-horizon, multi-step work. It provides an opinionated agent that runs out of the box while allowing developers to extend, override, or replace its components rather than treating the default architecture as fixed.

The harness supports planning, file operations, context management, tools, sub-agents, model choice, and production capabilities inherited from LangGraph such as streaming, persistence, checkpointing, tracing, and evaluation. It is model-agnostic and can work with compatible frontier, open-weight, or local models that support tool calling.

Deep Agents is useful for developers who want a stronger starting point than assembling an agent loop from basic primitives. It targets coding, research, and other complex tasks that require sustained execution, while preserving enough modularity for teams to customize tools, prompts, models, and orchestration for their own applications.