AI Briefingdeer-flow: Super Agent Harness Equipped with Sub-agents and Sandboxes
bytedance/deer-flow
About the project
DeerFlow 2.0, released by ByteDance, is a super agent harness that integrates sub-agents, memory, and sandboxes to perform complex tasks. It is designed to handle various automation tasks beyond simple research through extensible skills. It was newly developed as a complete separation from the Deep Research framework of the previous 1.x version.
It integrates with coding agents such as Claude Code and Codex, and supports MCP servers and IM channels. It features built-in complex context engineering capabilities, including session goal setting, manual context compression, and long-term memory management. It provides a built-in Python client and a terminal workbench (TUI) to ensure a flexible execution environment.
It supports various LLM providers such as Doubao-Seed-2.0-Code, DeepSeek v3.2, and Kimi 2.5, and also allows integration with gateways like vLLM and OpenRouter. Docker deployment is recommended, and in local development environments, setup can be completed in 2 minutes via an interactive wizard. To reduce security risks, sandbox mode and file system isolation features are provided by default.
bytedance/deer-flow
An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours.
Python
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.