Deer Flow

DeerFlow is an open source agent harness, built on LangChain, that researches, writes code and creates content on a user's behalf. Version 2.0 extends the project from a deep research agent into what its authors call a full stack super agent: it combines sandboxes, memory, tools, skills and subagents to handle tasks that can take anywhere from minutes to hours. It is aimed at developers and technical teams who want a self run agent for multi step work. Showcased tasks include producing a web page from a deep research report on agent technology trends, generating a video and reference image from a scene in a novel, drawing an explanatory comic strip, running an exploratory data analysis with visualisations, researching the content of a YouTube video, and collecting and summarising a person's recent podcast appearances. Agent skills load progressively, only when a task needs them, and teams can add their own skill files alongside the built in library. DeerFlow is given its own computer: a Docker based sandbox where it executes commands, manages files and runs long tasks. The project recommends the open source All in One Sandbox, which combines a browser, shell, file system, MCP and VS Code Server in a single container that is isolated, persistent and mountable. DeerFlow is free, open source software available from GitHub. The site publishes no pricing or hosted plan; running it means supplying your own infrastructure and model access. The project is developed in the open, with a community programme inviting contributions to its roadmap.