Redigg 🦎

Autonomous Research Agent for Scientific Discovery

English | 中文

🇬🇧 English

📖 Introduction

Redigg is an advanced autonomous research agent designed to accelerate scientific discovery. It acts as a tireless research assistant that can autonomously search for literature, analyze papers, explain complex concepts, and generate comprehensive PDF reports. It features a self-evolving memory system that learns from every interaction.

✨ Key Features

• Autonomous Research: Performs deep literature reviews, finding and summarizing relevant papers from the web.
• Auto-Research Loop: Continuously improves research reports through iterative planning, critiquing, and refinement.
• Memory & Evolution: Remembers past interactions and evolves its skills over time.
• Multi-Modal Output: Generates structured markdown summaries and professional PDF reports.
• Code & Paper Analysis: Can analyze local codebases and specific scientific papers in depth.

🚀 Modes

1. Research Chat: Have a natural conversation with the agent. It will use its tools (search, analysis) as needed to answer your questions.
2. Literature Review: Ask for a review on a specific topic (e.g., "Literature review on LLM agents"). Redigg will scour the web for papers and synthesize a report.
3. Auto-Research: Enable "Auto Mode" to let Redigg iteratively refine a document. It will draft, critique, improve, and generate new PDF versions in a loop until satisfied.
4. Concept Explainer: Ask "Explain [Concept]" for a detailed, pedagogical breakdown of complex topics.

🛠️ Quick Start

Prerequisites: Node.js >= 22.0.0

Option 1: Install via NPM (Recommended)

npm install -g @redigg/redigg
redigg start

This will start the Gateway on http://localhost:4000 (serving the Dashboard).

Option 2: Run from Source

1. Clone & Install

   git clone https://github.com/redigg/redigg.git
   cd redigg
   npm install

2. Configure Copy .env.example to .env and add your OpenAI API Key:

   cp .env.example .env
   # Edit .env file

3. Run Start both the backend gateway and frontend UI with a single command:

   npm run dev

   • UI: http://localhost:5173
   • Gateway: http://localhost:4000

🔌 A2A Integration

Redigg supports the Agent-to-Agent (A2A) protocol, allowing it to communicate with other agents or platforms like OpenClaw.

1. Endpoints

• Agent Card: http://localhost:4000/.well-known/agent-card.json
• JSON-RPC: http://localhost:4000/a2a/jsonrpc

2. Connect with OpenClaw

To use Redigg as a node within an OpenClaw network:

1. Start Redigg: Ensure Redigg is running (redigg start).
2. Configure OpenClaw: Add Redigg to your OpenClaw agents.yaml or configuration:

   agents:
     - name: "redigg"
       url: "http://localhost:4000/.well-known/agent-card.json"

3. Interact: You can now route tasks to Redigg via OpenClaw, e.g., "Ask redigg to perform a literature review on X".

💡 Examples

• "Perform a literature review on multi-agent reinforcement learning." -> Generates a summary and list of papers.
• "Explain the concept of Transformer architecture." -> Provides a detailed explanation.
• "Analyze this paper: [Title]" -> Deep dives into a specific paper.
• "Auto-research: Future of AI in Healthcare (3 iterations)" -> Produces a refined PDF report after 3 rounds of self-improvement.

🗺️ Roadmap

• Enhanced Survey Skill: Support for in-depth surveying, data plotting, and chart generation.
• Skill Ecosystem Expansion: Integrate more research-oriented skills to accelerate paper writing and full-link research capabilities.
• Coding Agent Integration: Connect with coding agents (e.g., Cursor, Claude Code) for autonomous code writing, debugging, and execution.
• Research Infrastructure: Access to computational and experimental infrastructure for autonomous scientific experiments.
• Multi-Agent Collaboration: Enable 24/7 fully autonomous research operations through multi-agent collaboration and task orchestration.

🇨🇳 中文 (Chinese)

📖 简介

Redigg 是一个专为加速科学发现而设计的先进自主研究智能体。它就像一位不知疲倦的研究助手，能够自主搜索文献、分析论文、解释复杂概念，并生成专业的 PDF 报告。它具备自进化记忆系统，能够从每一次交互中学习并变得更强。

✨ 核心功能

• 自主研究: 进行深度的文献综述，从网络上搜索并总结相关论文。
• 自动研究闭环 (Auto-Research): 通过迭代式的规划、批判和优化，持续改进研究报告质量。
• 记忆与进化: 能够记住过去的交互，并随着时间推移进化其技能。
• 多模态输出: 生成结构化的 Markdown 摘要和专业的 PDF 报告。
• 代码与论文分析: 支持分析本地代码库结构以及深度解读特定科学论文。

🚀 运行模式

1. 研究对话 (Research Chat): 与智能体进行自然对话。它会根据需要自动调用工具（搜索、分析）来回答你的问题。
2. 文献综述 (Literature Review): 指定一个主题（例如：“关于 LLM 智能体的文献综述”），Redigg 将全网搜寻论文并合成报告。
3. 自动研究 (Auto-Research): 开启“自动模式”，让 Redigg 迭代打磨文档。它会循环执行“起草-批判-改进-生成 PDF”的流程，直到达到满意的效果。
4. 概念解释 (Concept Explainer): 发送“Explain [概念]”，它会像教授一样详细拆解复杂的科学概念。

🛠️ 快速开始

前置要求: Node.js >= 22.0.0

方式 1: 通过 NPM 安装 (推荐)

npm install -g @redigg/redigg
redigg start

这将在 http://localhost:4000 启动网关和界面。

方式 2: 源码运行

1. 克隆与安装

   git clone https://github.com/redigg/redigg.git
   cd redigg
   npm install

2. 配置 复制 .env.example 为 .env 并填入你的 OpenAI API Key：

   cp .env.example .env
   # 编辑 .env 文件

3. 运行 使用一条命令同时启动后端网关和前端界面：

   npm run dev

   • 界面 (UI): http://localhost:5173
   • 网关 (Gateway): http://localhost:4000

🔌 A2A 集成

Redigg 支持 Agent-to-Agent (A2A) 协议，允许与其他智能体或平台（如 OpenClaw）进行通信。

1. 端点地址

• Agent Card: http://localhost:4000/.well-known/agent-card.json
• JSON-RPC: http://localhost:4000/a2a/jsonrpc

2. 连接到 OpenClaw

要将 Redigg 作为 OpenClaw 网络中的一个节点使用：

1. 启动 Redigg: 确保 Redigg 正在运行 (redigg start)。
2. 配置 OpenClaw: 将 Redigg 添加到 OpenClaw 的 agents.yaml 或配置文件中：

   agents:
     - name: "redigg"
       url: "http://localhost:4000/.well-known/agent-card.json"

3. 交互: 现在你可以通过 OpenClaw 向 Redigg 分发任务，例如："Ask redigg to perform a literature review on X"。

💡 使用示例

• "Perform a literature review on multi-agent reinforcement learning." -> 生成论文摘要和列表。
• "Explain the concept of Transformer architecture." -> 提供详细的概念解释。
• "Analyze this paper: [Title]" -> 深入分析特定论文。
• "Auto-research: Future of AI in Healthcare (3 iterations)" -> 经过 3 轮自我优化后生成一份精炼的 PDF 报告。

🗺️ 路线图 (Roadmap)

• 增强 Survey Skill: 支持深度调研、数据绘图、图表生成等。
• 扩展技能生态: 接入更多科研导向的技能，全链路增强科研能力，实现更快的论文撰写。
• 接入 Coding Agent: 连接 Coding Agent（如 Cursor、Claude Code 等），实现自主代码编写、调试和执行。
• 接入科研基建: 能够调用计算和实验基础设施，进行自主科研实验。
• 多 Agent 协同: 实现多种 Agent 协同工作，达成 7x24 小时全自主科研运行。