DeerFlow

DeerFlow is an open-source SuperAgent harness built by ByteDance that combines deep research, code execution, and media generation into a single autonomous agent platform. It uses Docker-based sandboxes, long/short-term memory, and a modular skill system to handle tasks ranging from minutes to hours in duration. DeerFlow 2.0 evolves beyond a simple research agent into a full-stack Super Agent capable of planning, sub-tasking, and executing complex workflows sequentially or in parallel. With over 29,000 GitHub stars, it is self-hosted under the MIT License and supports multiple LLM backends.

• Agent Skills System — Load skills progressively; extend DeerFlow with custom skill files or use the built-in library covering deep search, frontend design, deployment, and domain-specific knowledge.
• AIO Sandbox — Provides an isolated, Docker-based All-in-One Sandbox combining Browser, Shell, File, MCP, and VSCode Server for safe and persistent code execution.
• Long/Short-term Memory — Context engineering enables the agent to remember user preferences and prior context across sessions for more personalized and coherent task execution.
• Planning and Sub-tasking — The agent plans ahead, reasons through complexity, and executes tasks sequentially or in parallel using a structured sub-agent architecture.
• Video and Media Generation — Generate videos, images, and rich reports from research tasks, including scene-based video creation from literary sources.
• Multi-Model Support — Compatible with Doubao, DeepSeek, OpenAI, Gemini, and other LLM providers, giving users flexibility in model selection.
• Deep Research — Conducts comprehensive web research, synthesizes findings, and produces structured reports with visualizations and insights.
• Open Source & Self-Hosted — Licensed under MIT, fully self-hostable with complete control over data and infrastructure; get started by cloning the repository and running the Docker-based setup.