Price terms as published
Each line is the vendor’s own wording, split from the price capture checked on 2026-10-16. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.
| # | Plan / term | Checked |
|---|---|---|
| 1 | Free (MIT) | 2026-10-16 |
DeerFlow hit GitHub Trending at #1 on February 28, 2026, and as of October 16, 2026 the GitHub API reports 83,925 stars, 11,699 forks, last push the same morning. Chinese tech media calls it “ByteDance’s Deep Research.” That description is wrong in three ways worth understanding before you spend a weekend on it. This is a terms-and-repo analysis, not a hands-on benchmark — our hands-on budget is currently spent on the self-evolving agent OS reviewing next door.
What it actually is (hype problem #1: it’s not Deep Research anymore)
DeerFlow 2.0 is a ground-up rewrite that shares no code with v1 — the original Deep
Research framework lives on, frozen-ish, in the 1.x branch. Version 2.0 is a general
“super agent harness”: it orchestrates sub-agents, long-term memory, sandboxed execution,
extensible skills, and a message gateway, aimed at tasks that “take minutes to hours.”
Built on LangChain/LangGraph, so most of the plumbing is familiar to anyone who has used
either.
The feature list (all verified in README/docs on 2026-08-29): sandboxed file system with bash/file-write toggles, scheduled tasks, MCP server support, IM channel integration, a terminal workbench, Claude Code integration, session goals, and manual context compaction.
What it actually costs (hype problem #2: “free” is doing a lot of work)
The software is genuinely MIT-licensed — fully permissive, commercial use fine, no custom-license asterisk (we checked; plenty of trending repos hide one). But a harness has no brain: you bring model API keys, and a long-horizon agent burns tokens across sub-agents for every task. ByteDance’s README recommends Doubao-Seed-2.0-Code, DeepSeek v3.2, and Kimi 2.5 — and links its own Volcengine/BytePlus “coding plan” subscriptions. Realistic all-in cost is your inference bill, same as any agent framework. For scale: this site’s own agent workload runs roughly $40/month in DeepSeek API calls for a few hundred pages of writing and code — and a long-horizon harness orchestrating sub-agents is the multi-x scenario, not the baseline.
The quiet commercial pull (hype problem #3)
DeerFlow’s README integrates and promotes InfoQuest, BytePlus’s search/crawl toolset, and the Volcengine plan links above. None of it is hidden, but “ByteDance’s open-source repo” is partly a well-built funnel toward ByteDance Cloud. Use DeepSeek or any OpenAI-compatible endpoint instead — multi-provider configs are documented — and the funnel’s harmless.
Who should self-host it
You need tasks that run past the half-hour mark (research sweeps, codebase migrations),
want your data on your own box, and are comfortable with Docker (recommended), Python 3.12+,
Node 22+, and the rough edges of a project sitting on 502 open issues. Setup is a guided
make setup wizard — README claims ~2 minutes plus your keys. If you just want research
answers now, a hosted $20/mo assistant beats installing anything. One more flag: the repo
ships its own security notice that improper deployment (it can run bash and write files)
introduces real risk — treat that section as required reading, not fine print.
Verdict
A credible, honestly-licensed harness where the marketing articles are stale — but it is infrastructure, not an appliance. We do not score it: the review above is a terms-and-repo analysis, not a measured run.
Methodology & sources
Star/fork/issue counts and license field from the GitHub API on 2026-10-16; all feature and configuration claims checked against the repo’s README and linked docs the same day; the #1 Trending date is from the repo’s own announcement. No hands-on execution was performed for this article; treat it as a documentation review with a publish-ready structure.