
DeerFlow
ByteDance's self-hosted general agent — a finished application, no longer the deep-research framework it began as
What is DeerFlow?
Version 2.0 is a complete application behind one port — web UI, gateway backend and all — where a lead agent spawns sub-agents, keeps persistent memory, executes work in sandboxes from local Docker to Kubernetes and E2B, and produces reports, slide decks, web pages and video through Markdown-defined skills. Models wire up through LangChain provider classes, search works keyless via DuckDuckGo and swaps for six alternatives, and the agent answers from Telegram, Slack, Feishu and other chat channels. Two facts should precede adoption: the celebrated v1 deep-research pipeline is frozen on a separate branch and is not what installs today, and the README scopes the design to loopback-only trusted use — gateway admin is equivalent to code execution on the host.
What can you do with DeerFlow?
- One make command stands the whole thing up — make setup interviews you and writes the configuration, make doctor verifies it, and make docker-start serves the web UI on a single port; a TUI and an embeddable Python client cover the exceptions. The sizing table is honest — it starts at 4 vCPUs and 8 GB of RAM and recommends double for Docker development.
- Sub-agents work inside sandboxes you choose — The lead agent delegates to sub-agents whose execution lands in a sandbox — local process, Docker, Kubernetes or E2B — so letting an agent run code becomes a deployment decision rather than an act of faith.
- Skills define what a run can produce — Research reports, slide decks, web pages, images and video come from Markdown-defined skills that load progressively, activate with a slash command, and install as archives. Model wiring reaches OpenAI-compatible gateways, vLLM, and CLI-backed providers such as Claude Code's own sign-in.
- Reachable from the chat apps you already run — Connectors for Telegram, Slack, Discord, Feishu, WeChat, WeCom and DingTalk let the agent take work from a chat thread without the server ever being exposed to the internet.
Before you choose DeerFlow
- The deep-research DeerFlow that made the project famous is v1, wrapped up on the main-1.x branch — installing today gets the 2.0 rewrite, which shares no code with it.Reported in#824
- Security is scoped to trusted loopback-only deployment by the README's own account, and skill gating is behavioural rather than a hard boundary — a shell command can reach what a disabled skill could not.Reported in#4107
- There has been exactly one release: users track the main branch of a rewrite that went GA in June 2026, moving at over a hundred commits in the first three weeks of August.
Star history
19 Aug to 30 Oct · +4.4k
Frequently asked questions
Is DeerFlow free for commercial use?
DeerFlow is released under the MIT licence — OSI-approved open source, which permits commercial use.
How can DeerFlow be deployed?
DeerFlow is available as Self-hosted / Runs locally.