deerflow

DeerFlow - Deep Research Agent (Go Port)

This is a Go implementation of the ByteDance DeerFlow deep research agent, built using langgraphgo and langchaingo.

DeerFlow is a multi-agent system designed to conduct deep research on a given topic. It plans a research strategy, executes search steps (simulated or real), and synthesizes a comprehensive report.

Features

• Multi-Agent Architecture: Orchestrates Planner, Researcher, and Reporter agents using a state graph.
• Web Interface: A modern, dark-themed web UI with real-time status updates using Server-Sent Events (SSE).
• CLI Support: Can be run directly from the command line for quick queries.
• Extensible: Built on langgraphgo, making it easy to add new nodes, tools, or complex control flows.

Prerequisites

• Go 1.23 or higher
• An OpenAI-compatible API Key (e.g., OpenAI, DeepSeek)

Configuration

Set the following environment variables:

export OPENAI_API_KEY="your-api-key"

# Optional: If using DeepSeek or another compatible provider
export OPENAI_API_BASE="https://api.deepseek.com/v1" 

Usage

Web Interface (Recommended)

Build and run the application:

go build -o deerflow ./showcases/deerflow
./deerflow

Open your browser and navigate to http://localhost:8080.

Command Line Interface (CLI)

Run a query directly from the terminal:

./deerflow "What are the latest advancements in solid state batteries?"

Project Structure

• main.go: Entry point. Handles CLI arguments and starts the HTTP server.
• graph.go: Defines the State struct and the Graph topology (Nodes and Edges).
• nodes.go: Contains the implementation logic for Planner, Researcher, and Reporter.
• web/: Contains the frontend assets (HTML, CSS, JS).

Architecture

The agent follows a sequential workflow:

1. Planner: Decomposes the user's query into a step-by-step research plan.
2. Researcher: Iterates through the plan, gathering information for each step.
3. Reporter: Synthesizes all gathered information into a final, formatted report.

License

MIT