DeerFlow is an open-source super agent harness designed to orchestrate complex tasks through sub-agents, memory, and sandboxes, catering to long-horizon research and development processes.
Source: Official README and linked documentation View on GitHub →DeerFlow is being discussed because its coverage to managing complex tasks, addressing the pain points of inefficient workflows and the need for a system to handle long-duration tasks. Its unique technical choices, such as the integration of various skills and sub-agents, differs through the open-source field.
Source: Official README and linked documentationOrchestrates sub-agents, memory, and sandboxes to handle complex tasks ranging from minutes to hours, supporting long-horizon research and development processes.
Source: Official README and linked documentationSupports integration of various skills and tools, enhancing the agent's capabilities and adaptability to different tasks.
Source: Official README and linked documentationEnables the creation and management of sub-agents to perform specific tasks, contributing to the efficiency of the system.
Source: Official README and linked documentationUtilizes sandboxes and a managed file system to ensure secure and controlled execution of tasks, preventing interference and data leakage.
Source: Official README and linked documentationMaintains a long-term memory system to store and retrieve information over extended periods, aiding in the continuity and context-awareness of tasks.
Source: Official README and linked documentationThe architecture of DeerFlow is modular, with distinct components for skills, sub-agents, memory, and sandboxes. It employs design patterns such as the agent-based system and the use of sandboxes for task isolation. Data flow is managed through a message gateway, and key technical decisions include the integration of various LLM providers and the emphasis on security and efficiency.
Source: Official repository tree and dependency filesNot enough informationDeerFlow is suitable for developers and research teams working on complex, long-horizon projects that require the orchestration of multiple tasks and the management of large amounts of data. It is useful in scenarios such as AI research, software development, and content creation, where tasks are interdependent and require a high degree of automation and control.
Source: Official README and linked documentationThe latest official release is v2.0.0, released on 2026-06-25, and the logo 2.x is officially rewritten. The current main has continued to the back-end 2.1.0-rc0; production deployment should be fixed to v2.0.0 or reviewed specific commits, and main cannot be used as a stable version.
Source: github.com/bytedance/deer-flow/releases/tag/v2.0.0status: docs-only | checked_at: 2026-09-23 | environment: Check official README, Install.md, deployment documentation, backend/pyproject.toml, mit license and v2.0.0 release | limitations: No local installation or runtime verification was performed. The repository was not cloned, make setup/doctor/up was not performed, the model, sandbox, gateway, database or message channel was not started, and resource occupancy and permission isolation were not verified | observed_output: The latest official release of GitHub is v2.0.0, released on 2026-06-25; the current main's backend/pyproject.toml flag is 2.1.0-rc0, which requires Python 3.12 +
Source: github.com/bytedance/deer-flow/releases/tag/v2.0.0OpenHands Agent Canvas is a self-managed control center for coding agents and automation, focusing on ACP Agents such as OpenHands, Claude Code, and Codex, project directories, and engineering tasks. DeerFlow emphasizes general long tasks, skills, memory, sub-agent, content generation, and multi-message channels. The main goal is for the software engineering team to look at OpenHands when scheduling; to include research and content processes in the same harness before evaluating DeerFlow.
Source: github.com/OpenHands/OpenHandsDeerFlow is a project for teams requiring a extensible system to manage complex, long-duration tasks. Its unique features and modular architecture make it a tool for research and development, particularly in fields that demand high levels of automation and data management.
The latest official release of GitHub is v2.0.0, released on 2026-06-25; the current main's backend/pyproject.toml flag is 2.1.0-rc0, which requires Python 3.12 + No local installation or runtime verification was performed.
The review environment was: Check official README, Install.md, deployment documentation, backend/pyproject.toml, mit license and v2.0.0 release deer-flow behavior outside that scope still needs a local check.