DeerFlow

Summary: DeerFlow is an open-source AI SuperAgent framework developed by ByteDance that helps users automate complex, multi-step tasks involving research, coding, data analysis, and content creation. Powered by agent orchestration, subagents, persistent memory, sandboxed execution, and customizable skills, it can plan tasks, delegate work, use tools, and generate practical deliverables such as research reports, presentations, websites, dashboards, and code. Built on LangGraph and LangChain, DeerFlow 2.0 is designed for long-running workflows that may take minutes or hours to complete. It is available under the MIT License and supports self-hosted deployment.

What is DeerFlow?

DeerFlow is an AI agent platform that goes beyond ordinary chatbot interactions. Instead of simply responding to prompts, it provides an execution environment where AI agents can plan tasks, perform multiple actions, use external tools, manage files, and produce completed work. It is designed to handle complex tasks that require several stages of reasoning and execution.

One of its key capabilities is multi-agent orchestration. DeerFlow can divide a large task into smaller assignments and delegate them to specialized subagents. These agents can work independently or in parallel, allowing the main agent to coordinate research, analysis, coding, and other activities without overwhelming a single conversation context.

The platform also includes a customizable Skills and Tools system. Users can extend its capabilities with built-in or custom skills for deep research, report generation, slide creation, website development, image and video generation, and data analysis. It also supports external tools through Model Context Protocol (MCP) integrations and custom Python functions.

Another important feature is its sandboxed execution environment. Agents can read and write files, execute commands, run code, and create deliverables within an isolated workspace. Combined with persistent memory and context management, this allows DeerFlow to maintain useful information and work through longer projects across multiple steps.

DeerFlow provides two main components: DeerFlow Harness, the core runtime and SDK for developers building custom agent systems, and DeerFlow App, a reference application for deploying and using those capabilities through a workspace interface. This makes it useful both for developers creating AI-powered products and teams looking to operate an AI agent environment.

Key Features

1. Autonomous Task Planning: Breaks complex requests into smaller steps and coordinates their execution.

2. Multi-Agent Orchestration: Delegates specialized tasks to multiple subagents, including parallel execution.

3. Deep Research: Searches for information, gathers sources, and creates structured research reports.

4. AI Coding & Execution: Writes code, works with files, runs commands, and supports software development workflows.

5. Custom Skills & Tools: Extends agent capabilities through reusable skills, MCP servers, and Python functions.

6. Sandboxed Execution: Provides an isolated workspace for file operations, code execution, and project development.

7. Persistent Memory: Retains useful information across conversations and sessions.

8. Content & Presentation Generation: Supports workflows for creating reports, slide decks, websites, and other content.

9. Data Analysis: Helps process datasets, perform analysis, and generate visualizations.

10. Multi-Model Support: Can be configured to work with different language-model providers.

11. Agentic Browser Control: Supports browser navigation, clicking, typing, and interacting with dynamic websites through its optional browser tools.

12. Scheduled Tasks: Supports one-time, recurring, and interval-based task scheduling in its application.