DeerFlow CLI
ByteDance's open agent harness for long tasks: research, code, reports and slides with sub-agents, memory and sandboxes
- 84,629 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs code and bash commands in a sandbox or on the host depending on settings
- Stores model and search provider keys in .env
Install
- Works in
Manual install
git clone https://github.com/bytedance/deer-flow.git
cd deer-flow
make setupThe wizard creates config.yaml and .env. Then start with make dev or make docker-init and make docker-start.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs code and bash commands in a sandbox or on the host depending on settings
- Stores model and search provider keys in .env
This is third-party code. Review the repository files before installing.
What it does
DeerFlow 2.0 is a self-hosted agent with a web UI that splits work across sub-agents, keeps memory and runs code in isolated sandboxes. Its abilities come from SKILL.md skills: built-ins include deep research, data analysis, charts, slide, image, video and podcast generation, and academic paper review. Skills load only when needed, can be invoked explicitly with /skill-name, replaced or extended. A claude-to-deerflow skill connects it from Claude Code.
Who it is for. For teams and researchers who need their own agent for hours-long tasks with reports, not just an editor assistant.
Good fit when
- You need deep research on a topic ending in a report or slides
- The task runs long and needs sub-agents and a sandbox
- You want your own agent deployment with custom skills
Not a fit when
- You need a quick code edit in an IDE
- No server resources: local runs need at least 4 vCPU and 8 GB RAM
Example request
Research the open-source vector database market and produce a report with a comparison table and slidesLimitations
Requires Python 3.12+, Node.js 22+ and ideally Docker. Models connect via your provider key; web search is configured separately. Availability from Russia depends on the chosen model and search providers.
How to disable. Stop services with make stop, make docker-stop or make down depending on how you run it, then delete the project folder.
FAQ
Is this a continuation of DeerFlow 1.x?
No, 2.0 is a ground-up rewrite that shares no code with v1, which lives on the 1.x branch.
Can I add my own skills?
Yes, drop in a folder with SKILL.md; built-in skills can be disabled or replaced too.
Discussion
No questions or reviews yet.
Related
CLI toolsAn open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A desktop app where Claude Code, Codex, Gemini and other agents work as a team of specialists in a shared project
Shared cloud memory for a team's agents: stores session traces and turns repeated patterns into skills
Self-hosted system for a team of AI agents: org chart, tasks, budgets and oversight of Claude Code, Codex and OpenClaw
pydantic-deepagents
pydantic-deepagents (pydantic-deep)
An open-source terminal AI assistant and Python framework with a Docker sandbox, agent tools and subagents
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
README in short
The README opens with a one-line prompt for a coding agent to clone and bootstrap the project, then covers manual setup: clone, the make setup wizard for model, search and sandbox mode, and make doctor for checks. It supports local dev, Docker and production via make up with the UI on port 2026. Skill sections explain progressive loading and explicit invocation. MIT licensed.