Deer Flow

Deer Flow (Deep Exploration and Efficient Research Flow) is an open-source super-agent harness from ByteDance. It started as an internal deep-research framework at ByteDance and was released to the community, where it has grown into one of the most popular agent projects available, accumulating tens of thousands of GitHub stars. Built on LangGraph and LangChain, it ships batteries-included and fully extensible — everything an agent needs out of the box. The framework's defining trait is long-horizon autonomy. Rather than handling single prompts, Deer Flow orchestrates tasks that take minutes to hours through the coordination of multiple specialized sub-agents — such as a Researcher, a Coder, and a Writer — each with its own role in a multi-agent topology. It equips these agents with sandboxes for safe code execution, persistent memory, tools, skills, and the ability to spawn subagents, so it can research a topic, write and run code, crawl the web, and produce a polished report end-to-end. Deer Flow places a strong emphasis on human-in-the-loop control: research plans can be modified interactively through natural language, or auto-accepted for fully streamlined runs. This makes it suitable both for fully autonomous workflows and for cases where a human wants to steer or correct the agent mid-task. The result is a modular, community-driven framework that combines language models, search engines, web crawlers, and even text-to-speech into a coherent deep-research pipeline. It is aimed at developers and teams who want a powerful, extensible foundation for building deep-research and long-horizon agent applications — an open alternative to stitching together LangGraph primitives from scratch.