AI · GitHub repo
DeerFlow
Open-source long-horizon agent harness with Docker sandboxes, memory, and sub-agents
DeerFlow is ByteDance's open-source autonomous agent harness built to execute long-running tasks spanning minutes to hours. It pairs a lead agent with parallel sub-agents inside isolated Docker sandboxes featuring full filesystem access, cross-session memory, and progressive skill loading. It works across major LLM providers including Claude, GPT-4, Gemini, DeepSeek, and local models via Ollama.
What it's for
- Conduct deep web research with cited sources and structured visual reports
- Build complete websites and web apps directly inside isolated execution sandboxes
- Generate slide decks and presentations from multi-topic research queries
- Execute arbitrary Python and bash code safely for automated data processing tasks
Features
- Isolated Docker sandboxes with full filesystem access and bash execution
- Parallel sub-agent orchestration coordinated by a lead agent
- Progressive skill loading to reduce LLM context bloat
- Cross-session memory for user preferences, workflows, and output styles
- Message gateway supporting chat channels like Telegram, Feishu, and DingTalk
- Support for OpenAI-compatible APIs, Anthropic Claude, Google Gemini, DeepSeek, and Ollama
Pricing
Open source. Free and open source under the MIT license.
Details
- Type
- GitHub repo
- Category
- AI
- Topic
- Autonomous Agents
- Open source
- Yes
- Platforms
- Self-hosted, Web, Linux, macOS, Windows
- For
- Developers and AI engineers looking for an extensible, self-hosted harness for long-horizon autonomous tasks.
- Alternative to
- Manus, AutoGPT, OpenHands, Devin
- Tags
- ai agents, superagent, docker, deep research, langchain, llm
Why it was saved
- @heynavtoor: A full agentic framework that actually executes (its own sandboxed computer), not just suggests — useful as a base for building automated research and content workflows.
Compare
Related
Open-source, self-hosted AI agent with persistent memory and messaging integrations.
Open-source, fully local autonomous AI agent that browses and writes code
Open-source platform and framework for building autonomous AI agents
A Python multi-agent pattern pairing Opus orchestration with Haiku worker instances.
All-in-one AI workspace with autonomous agent tools for content, research, and analysis.
Open vision-to-action foundation model for generalist autonomous game playing
Updated