Overview
DeerFlow is an open-source framework for building and operating agent systems that can research, code, and create content. Its runtime combines memory, tools, skills, sandboxes, and subagents to handle tasks ranging from short interactions to longer multi-step work.
The project separates reusable agent infrastructure from a complete application experience, allowing teams to adopt the layer that fits their development or deployment needs.
Key Capabilities
- Memory preserves useful context for agent work.
- Tools and skills extend the actions an agent can perform.
- Sandboxes provide isolated environments for browser, shell, file, MCP, and development-server tasks.
- Subagents let a system divide work among specialized agent processes.
- Persistent, mountable filesystems support longer-running workflows.
Harness and Application
DeerFlow Harness is the core SDK and runtime layer. It is intended for developers integrating agent capabilities into an existing system or creating a custom agent product, with documentation for configuration, customization, memory, tools, skills, sandboxing, and broader integrations.
DeerFlow App is a reference application built on the Harness. It packages those runtime capabilities into an end-user product and covers local operation, deployment, application configuration, workspaces, agents, threads, maintenance, and troubleshooting.
Use Cases
DeerFlow supports research and creation workflows that combine several types of work. Examples include producing research reports and webpages, analyzing datasets with visualizations, collecting and summarizing podcast material, reviewing video content for a research task, and creating images or videos based on written scenes.
This breadth makes the framework relevant both to teams building their own agent systems and to operators who want to deploy a ready application. The modular Harness path emphasizes integration, while the App path emphasizes practical user workflows and operations.