DeerFlow
Open-source SuperAgent harness that researches, codes, and creates inside a persistent Docker sandbox — MIT licensed and self-hosted.
DeerFlow 2.0 RC is a serious open-source SuperAgent build, and the persistent Docker All-in-One Sandbox — Browser, Shell, File, MCP, VSCode Server in one container — plus progressively loaded SKILL.md files is the real draw for developers who want control without lock-in. The April 2026 2.0 RC added planning, sub-tasking, and cross-session memory, which pushed it past its deep-research origins. The tradeoff is honest: you maintain Docker, you debug the setup, and no vendor SLA is waiting for you. Pick it if your team is technical and self-hosting is a feature, not a chore; look at managed agent platforms instead if you want hosting handled for you. It is MIT licensed, so there is no licence
Verified 7d ago · liveness 63/100 · cite: rightaichoice.com/tools/deerflow
- Developers building custom self-hosted AI agents with full data control
- Research teams running long, multi-hour deep-research workflows
- Technical power users automating multi-step coding and content creation
- Open-source contributors who want to extend the agent with SKILL.md files
- Non-technical users who want a no-code agent they run in a browser
- Teams that require managed cloud hosting, uptime SLAs, or vendor support
- Buyers expecting ready-made Slack or Zapier connectors
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip DeerFlow if your team wants a managed agent platform with uptime guarantees and no Docker to maintain, rather than a self-hosted MIT-licensed framework you configure and operate yourself.
Self-hosting means you pay for the compute, storage, and staff time to run the Docker sandbox and maintain it — there is no licence fee, but there is an infrastructure and ops bill.
DeerFlow is MIT licensed and self-hosted, so there is no software licence cost at any team size — the budget line is infrastructure and engineering time instead. That makes it cheaper on paper than managed agent platforms and cloud deep-research services, which bundle hosting and support into a subscription, and more expensive in staff hours than a no-code tool. Compare it to committing an engineer to a platform subscription, not to a per-seat SaaS fee.
In short
DeerFlow — Open-source SuperAgent harness that researches, codes, and creates inside a persistent Docker sandbox — MIT licensed and self-hosted. Best for Developers building custom self-hosted AI agents with full data control, Research teams running long, multi-hour deep-research workflows, Technical power users automating multi-step coding and content creation. Free to use.
What's new in DeerFlow
Checked 7 days agoAcross the latest 2 updates: 1 launch and 1 changelog entry.
DeerFlow 2.0 RC
DeerFlow 2.0 RC released, evolving from a deep research agent into a full-stack SuperAgent with planning, sub-tasking, memory, and extensible skills. It also introduces the All-in-One Sandbox combining Browser, Shell, File, MCP, and VSCode Server.
Weekly - 2026-04-06
English weekly update covering recent improvements and community contributions, including notes on how tool-using agents should handle provider safety termination signals.
What people actually say about DeerFlow — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
12 mentions across 1 source (Hacker News) · researched Jul 3, 2026.
Average across the 1 source that answered — each source counts once, not each post.
- +Fully open-source under MIT license, free to self-host.
- +Docker sandbox with persistent filesystem for long-running tasks.
- +Supports multiple LLMs: OpenAI, Gemini, DeepSeek, Doubao, etc.
- +Progressive skill loading allows extensibility via custom SKILL.md files.
- +Handles diverse tasks: research, coding, video/image generation.
- −Setup requires Docker and technical know-how, not trivial.
- −Multi-agent orchestration may not outperform single-agent for many tasks.
- −Community showcase lacks impressive real-world results.
- −Documentation and support are community-driven, potentially sparse.
- −ByteDance origin may cause privacy concerns despite open-source.
- • Infrastructure costs for running Docker and LLM API fees
Viability Score
How well maintained and how widely used is DeerFlow? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Deep research with multi-step orchestration
- Planning and sub-tasking with sequential or parallel execution
- Short-term and long-term memory across a session
- Persistent Docker All-in-One Sandbox
- Sandbox bundles Browser, Shell, File, MCP, and VSCode Server
- Isolated, safe, persistent, mountable filesystem
- Long-running task execution in a sandboxed environment
- Progressive loading of Agent Skills
- Custom skill files authored as SKILL.md
- Multi-model support across Doubao, DeepSeek, OpenAI, and Gemini
- Code generation and execution inside the sandbox
- Image generation from text or scenes
- Video generation from text or scenes
- Web scraping and data collection
- Summarization and full research report generation
About DeerFlow
DeerFlow is an MIT-licensed, self-hosted SuperAgent harness — a runtime for building agents that research, code, and produce content end to end. Instead of wrapping a chat model, it hands the agent a real computer: a persistent, isolated, mountable Docker sandbox that bundles Browser, Shell, File, MCP, and a VSCode Server in a single All-in-One container, so the agent can execute commands, write 156-line programs, install packages like pygame 2.5.0, and grind through jobs that run for minutes to hours. The 2.0 RC release, announced April 8 2026, moved DeerFlow from a deep-research agent to a full-stack SuperAgent. It adds planning and sub-tasking so work can run sequentially or in parallel, short- and long-term memory across a session, and Agent Skills that load progressively — only what a task needs, when it needs it. Skills are plain SKILL.md files you can write yourself, covering domains such as deep-search, biotech, computer science, physics, and frontend design. The docs split cleanly into two layers: the DeerFlow Harness (SDK and runtime for building your own agent system) and the DeerFlow App (a reference application for deployment, operations, and end-user workflows). Model support is pluggable across Doubao, DeepSeek, OpenAI, and Gemini environments, and the docs include a model-provider reference. Documented outputs include multi-step research reports, generated images and video from text or scenes, web scraping, and exploratory data analysis on datasets. The docs were last updated July 11 2026. Who it's for: developers and research teams who want a custom agent stack with no vendor lock-in and are comfortable running Docker. Who it isn't for: teams that need managed hosting, uptime SLAs, or support contracts, or non-technical users who want a browser-based no-code agent.
Behind the Verdict
DeerFlow's differentiator is architectural, not cosmetic. Most 'agent' products are a chat surface over a hosted model; DeerFlow ships a runtime. The All-in-One Sandbox is a single Docker container combining Browser, Shell, File, MCP, and VSCode Server, with an isolated, persistent, mountable filesystem. That is what lets the agent install pygame 2.5.0, write 156 lines of code, load sprites, and report stable 60 FPS — the homepage demo is a concrete illustration of stateful, multi-step execution rather than a text completion. The 2.0 RC release (April 8 2026) is the material upgrade. Planning and sub-tasking let the agent decompose complexity and run steps sequentially or in parallel. Short- and long-term memory carry context across a session. Agent Skills load progressively rather than all at once, and you can extend the built-in library with your own SKILL.md files across deep-search, biotech, computer science, physics, and frontend design. Modalities documented in the scrape include code generation and execution, image generation from text or scenes, and video generation from text or scenes. The documentation is the strongest signal of intent: it explicitly separates the Harness (core SDK and runtime for building your own agent system) from the App (a reference application for deployment, operations, and end-user workflows), with tutorials for first conversation, first harness, tools and skills, memory, and deploying your own. That is a framework posture, not a product posture, and it means adoption has a real learning curve. Weaknesses are structural. It is self-hosted: Docker setup and technical maintenance are on you. The release is a Release Candidate, so stability may vary. Non-technical buyers are out of scope — there is no no-code browser path documented. And if your plan was to bolt DeerFlow onto Slack or Zapier, nothing in the scrape suggests those connectors exist; DeerFlow's integration story is MCP plus your own wiring. Where it fits: developers integrating agent capabilities into their own system, research teams running multi-hour analysis workflows, and open-source contributors who want to write skills. Where it doesn't: teams that need managed hosting and uptime commitments, or anyone who wants an agent running in five minutes without touching a container runtime.
Researching DeerFlow? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas DeerFlow actually fits — and what changes day-one when you adopt it.
Clones DeerFlow, spins up the All-in-One Docker sandbox, wires their preferred model provider (Doubao, DeepSeek, OpenAI, or Gemini), and reads the Harness Quick Start to create their first harness before wiring it into their own system.
Outcome: A running agent runtime they control, with the browser, shell, file, MCP, and VSCode Server tooling already inside one container rather than assembled by hand.
Uses planning and sub-tasking to break a research question into parallel branches, relies on memory to keep context across the session, and lets the sandbox grind through scraping and analysis over hours.
Outcome: A consolidated report assembled from multiple sources, produced without babysitting each step and with the working filesystem preserved in the sandbox.
Authors a SKILL.md file for a domain DeerFlow does not yet cover and relies on progressive skill loading so the agent only pulls that skill in for tasks that need it.
Outcome: A shared capability that becomes part of the built-in library, keeping the agent lean for unrelated tasks while extending what the community can do.
Use Cases
- Run multi-step deep research on a topic and assemble the findings into a comprehensive report.
- Watch a YouTube video, then conduct follow-up deep research on the ideas it raises.
- Build a webpage around a deep research report forecasting 2026 agent technology trends.
- Search a specific scene from a novel and generate a reference image plus a video from it.
- Generate a comic strip that explains the MoE architecture to teenagers.
- Run exploratory data analysis on a dataset such as the Titanic set and surface the key survival factors.
- Scrape and summarize a public figure's podcast appearances from the last six months.
- Deploy a frontend design skill agent that writes and executes its own code.
Models Under the Hood
as of 2026-10-04
Limitations
- DeerFlow 2.0 is currently in Release Candidate (M1 RC), so stability may vary.
- It is self-hosted and MIT licensed: Docker setup and ongoing maintenance are the user's responsibility, with no vendor support contract documented.
- Deployment relies on the All-in-One Docker sandbox combining Browser, Shell, File, MCP, and VSCode Server, which requires container and agent-runtime familiarity.
as of 2026-10-03
Verification history
We have re-verified DeerFlow 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where DeerFlow's pricing actually pencils out — and where peers do it cheaper.
DeerFlow is MIT licensed and self-hosted, so there is no software licence cost at any team size — the budget line is infrastructure and engineering time instead. That makes it cheaper on paper than managed agent platforms and cloud deep-research services, which bundle hosting and support into a subscription, and more expensive in staff hours than a no-code tool. Compare it to committing an engineer to a platform subscription, not to a per-seat SaaS fee.
Setup time & first value
How long it actually takes to get something useful out of DeerFlow — broken out by persona, not the marketing-page minute.
Developers with Docker experience: expect a working sandbox and first harness within an afternoon of following the Harness Quick Start and model-provider setup. Research teams deploying the App: budget roughly a day for deployment and workspace configuration before handing it to colleagues. Non-technical buyers: there is no realistic first-value path here without engineering help on the container
Switching to or from DeerFlow
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a thin research wrapper or chat UI: port your prompt logic into the DeerFlow Harness and move tool access into the All-in-One Sandbox so tasks can run as multi-step agent jobs.
- →From a chat model used manually: replan the workflow as sub-tasks so the agent can run steps in sequence or in parallel instead of turn-by-turn.
- →From a managed agent platform: replicate your tool definitions through MCP and the sandbox Shell/Browser/File tools, then repoint your existing skills at SKILL.md files.
- ↗To a managed agent platform: if you no longer want to run Docker, expect to rewrite tool access as that platform's native connectors and lose the persistent sandbox filesystem.
- ↗To a scripted pipeline: export your SKILL.md logic into your own orchestration code, trading DeerFlow's planning and memory layer for something you own outright.
- ↗To a hosted deep-research service: move report generation to a subscription product and retire the Harness and sandbox setup entirely.
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “DeerFlow”, and we withheld 6: 6 could not be judged, because “DeerFlow” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about DeerFlow.
Official links
Tools that pair well with DeerFlow
Common stack mates teams adopt alongside DeerFlow, with the specific reason each pairing earns its keep.
Imbue
Imbue is an open AI lab publishing modular, open-source coding-agent tools you run and inspect yourself.
Mastra
Open-source TypeScript framework for building durable AI agents and workflows, with a hosted platform for observability and cloud deployment.
Refact.ai
Open-source autonomous AI coding agent that plans, executes, and deploys tasks inside VS Code and JetBrains IDEs.
Featured Head-to-Head Comparisons
Deerflow vs Presto Voice
If you're a developer or power user needing a free, open-source multi-agent platform for research and coding, DeerFlow is the clear choice. For QSR chains wanting a proven voice AI solution to automate drive-thru orders and increase revenue (with backing from Dairy Queen and others), Presto Voice is the enterprise pick. These tools serve completely different markets — choose based on your domain: agent automation vs. restaurant operations.
Deerflow vs Praktika
DeerFlow and Praktika serve completely different needs. DeerFlow is a powerful open-source agent harness for developers who want to automate research, coding, and creation with full control. Praktika is a mobile app for language learners to practice speaking with AI tutors. Choose DeerFlow if you need a customizable AI agent sandbox; choose Praktika if you want to improve your conversational fluency.
Deerflow vs Truleo
If you're a law enforcement agency needing to connect siloed data for case leads, Truleo is purpose-built with jail call analysis, BWC analysis, and report writing features. For developers seeking a free, open-source multi-agent harness for research and coding, DeerFlow is far more flexible and cost-effective. Choose based on your domain: policing or programming.
Alternatives to DeerFlow
View all
Frequently Asked Questions
Used DeerFlow? Help shape our editorial sentiment research.