deer-flow
by bytedance
Research-first agentic framework for building reproducible, tool-enabled agent pipelines
What It Does
Orchestrates research-focused agentic workflows that combine language models with web search, crawling, and Python execution. Provides a modular framework and primitives for building reproducible pipelines, tool-augmented agents, and long-running experiments. Distinctive features include built-in connectors for data collection, task orchestration, and a community-driven collection of research recipes.
The Value Proposition
Ideal For
Researchers and engineers prototyping complex agent pipelines that need reproducible experiments, web/tool integration, and clear execution traces. Supports web/tool integration for tool integration and provides clear execution traces to audit agent behavior.
Real-World Examples
- Orchestrate multi-step research experiments that mix LLMs, web crawling, and code execution
- Capture detailed traces of tool calls and inputs for reproducible evaluation and debugging
- Prototype agent pipelines that require external data collection or long-running workflows