deer-flow
An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours.
An AI agent framework that orchestrates long-running automation tasks, research, coding, planning, across multiple tools, sub-agents, and memory, deployable via Docker or locally.
DeerFlow is an open-source AI agent framework built by ByteDance that lets you create powerful, long-running automation systems capable of researching topics, writing code, and producing outputs over extended periods — tasks that might take minutes or even hours to complete. The problem it solves is that standard AI chatbot interactions are short and stateless; DeerFlow enables AI agents that can plan, remember context, delegate to sub-agents, and use external tools in a coordinated way.
At its core, DeerFlow acts as a harness that orchestrates multiple components together: skills and tools the agent can call (such as web search or code execution), sub-agents that handle parallel or specialized work, a sandbox environment for safely running code, and long-term memory so the agent can retain information across steps. Users configure which AI model to use — the README mentions support for models like GPT-4o, DeepSeek, Gemini, and others via a YAML configuration file — then run the system either through Docker or a local development setup. The frontend is a Node.js web interface while the backend logic runs in Python. DeerFlow is version 2.0, described as a complete rewrite from its earlier deep-research-focused v1. You would use this if you need an AI that can autonomously complete multi-step research or coding tasks, for example writing a full technical report, building a small codebase, or aggregating information from many sources. The tech stack is Python 3.12 on the backend with Node.js 22 on the frontend, deployable via Docker.
Where it fits
- Build an AI agent that researches a topic and writes a complete technical report over several hours.
- Create a system that autonomously writes and tests code across multiple files in a sandbox environment.
- Set up a multi-agent workflow where specialized sub-agents handle different parts of a complex task in parallel.
- Deploy a long-running automation that aggregates information from web searches and external APIs into structured outputs.