DeerFlow

A multi-agent harness with Docker sandbox, long-term memory, skills and sub-agents.

Code & Development No-code & Automation
#Agents autonomes #Agents IA #Automatization workflows #Open source

Overview of DeerFlow

https://deerflow.tech/
Screenshot of DeerFlow
Visit DeerFlow →

Présentation détaillée

DeerFlow is an __open source__ framework developed by ByteDance for building __SuperAgents__ capable of carrying out long tasks lasting several hours. Built on top of __LangChain__ and __LangGraph__, it provides out of the box a file system, __short and long-term memory__, skills, sub-agents, an isolated Docker sandbox and a message gateway. Version 2.0 adds site creation, presentation generation and orchestration of multiple sub-agents in parallel, all under the MIT license.

What is DeerFlow?

DeerFlow is a harness, that is, a software framework that orchestrates everything an agent needs to function over time. It relies on LangChain and LangGraph for the orchestration layer, and additionally provides a file system, Docker sandbox, skills, short and long-term memory, and a message gateway. The whole is designed to allow an agent to plan, launch sub-agents, execute code, retain memory and iterate without permanent human intervention.

Key Features

DeerFlow shines through its functional richness. The Docker sandbox ensures each agent has a persistent environment where it can write files, install packages and execute scripts safely. Short-term memory tracks the context of the current task, while long-term memory stores user profiles, preferences and cross-session knowledge. Skills are Markdown files that describe how to accomplish a type of task: they are loaded on demand so the agent only solicits truly necessary skills. Planning allows agents to break down their missions and spawn sub-agents to execute sub-tasks. On connectivity, DeerFlow supports multiple LLM providers and can integrate with external tools via API. Version 2.0 adds site creation, presentation generation and parallel sub-agent orchestration, which significantly broadens the scope of action.

Use Cases

DeerFlow’s use cases span a wide spectrum. Researchers use it to conduct multi-source deep research with structured output. Software engineers use it to generate code, launch test suites or refactor legacy in parallel. Content studios entrust it with producing mini-sites, presentations and reports by leveraging its custom skills. On the enterprise side, DeerFlow also serves as the foundation for internal autonomous agent platforms, where you can audit each execution thanks to sandbox persistence and logs.

Advantages

The first benefit of DeerFlow is autonomy: MIT license, self-hosting, full control over data and models. The second is functional richness: few frameworks cover as many needs (sandbox, memory, skills, sub-agents) in a single package. The third is extensibility: the philosophy of skills as Markdown facilitates adding new capabilities without touching core code. The last is ecosystem: backed by LangChain, DeerFlow benefits from rapid evolution in the agentic ecosystem.

Pricing

DeerFlow is completely free. It is published under the MIT license, which authorizes commercial use without restriction. The only costs are your infrastructure (VPS, cluster or private cloud) and your LLM API calls to your chosen provider. This cost structure often proves more economical than SaaS, provided you have the skills to operate it.

Conclusion

DeerFlow is an excellent choice for anyone who wants to build true autonomous agents without sacrificing data sovereignty. Its complete approach, open source ecosystem and permissive license make it a reference framework for ambitious technical teams.

✅ Strengths

  • 100% open source under MIT license, self-hostable
  • Persistent Docker sandbox for safe code execution
  • Native short and long-term memory cross-session
  • Skills system Markdown loaded on demand
  • Orchestration of sub-agents in parallel (deep research)
  • Compatible multi-LLM (OpenAI, Anthropic, local models)

⚠️ Limits

  • Requires technical skills for self-hosting
  • No no-code interface ready to use
  • Cloud and LLM API costs at your charge
  • Initial setup longer than turnkey SaaS
  • Technical documentation, not beginner-friendly
👤 GOOD CHOICE?

DeerFlow est-il fait pour vous ?

✓ Ideal if you…

  • Équipes tech qui veulent un framework open source
  • Chercheurs et data scientists faisant du deep research
  • Développeurs construisant des agents autonomes custom
  • Entreprises sensibles à la souveraineté de leurs données
  • Startups SaaS bâtissant leur propre couche agentique

✗ To avoid if you…

  • Utilisateurs sans aucune compétence DevOps
  • PME cherchant un produit clé en main
  • Personnes voulant un agent en quelques clics
  • Équipes qui ne veulent pas gérer d’infrastructure

🎯 Our verdict

DeerFlow is rapidly establishing itself as one of the most complete open source frameworks for building ambitious AI agents. Version 2.0 transforms the tool into a true platform of SuperAgents capable of chaining research, code, sites and presentations on long tasks. Its combination of Docker sandbox, persistent memory, Markdown skills and sub-agent orchestration is rarely as well thought-out in the open source world. It’s also an excellent compromise for organizations that don’t want to be locked into a SaaS and that demand sovereignty over their data. The tradeoff is inevitable: you need to know how to handle Docker, configure your LLMs and accept a technical learning curve. For good teams, the power-to-cost ratio is unbeatable, especially with the MIT license that imposes no commercial usage limits.

❓ FREQUENT QUESTIONS

FAQ — DeerFlow

What is DeerFlow?
DeerFlow is an open source framework created by ByteDance for building SuperAgents capable of carrying out long tasks, leveraging LangGraph, Docker sandboxes, memory and sub-agents.
Is DeerFlow really free?
Yes. DeerFlow is under MIT license and 100% free. You only pay for your hosting infrastructure and LLM API calls.
What models can I use?
DeerFlow is multi-LLM: it works with OpenAI, Anthropic, Google Gemini, local models (Qwen, Llama) and any LangChain-compatible provider.
Do I need to be a developer to use it?
Yes. DeerFlow is a code framework, without ready-to-use no-code interface. Good command of Python and Docker is recommended.
How does it compare to AutoGen or CrewAI?
DeerFlow offers a more complete approach: persistent sandbox, Markdown skills, short and long-term memory, and parallel sub-agent orchestration, which brings it closer to a platform ready for very long tasks.
★★★★½ 4.7/5 (71 avis)
✅ Verified by Comparateur-IA
Code & Development No-code & Automation

A multi-agent harness with Docker sandbox, long-term memory, skills and sub-agents.

💰 Rate Free (open source)
🆓 Free trial Yes
🌐 Languages ANGLAIS, MULTILINGUE
Visit the site →
This site is registered on wpml.org as a development site. Switch to a production site key to remove this banner.