Sagi is a production-ready, open-source LLM agent framework that combines advanced technologies to create powerful AI agents. It features:

• 🧠 Deep-research workflow architecture, more details refers to the doc
• 🛠️ MCP (Model Context Protocol) integration, the guide of the usage of MCP refers to the doc
• 📊 GraphRAG-powered retrieval system, HiRAG mcp server refers to the repo

• [2025/04] 🔥 Sagi is publicly released!

• Interactive Web UI
• Advanced file chunking & indexing supports
• Improved documentation with more demos
• Chat history & workflow state management

1. System Requirements

   • Docker and Docker Compose (Installation Guide)

2. Clone the Repository

   git clone https://github.com/Kasma-Inc/Sagi.git
   cd Sagi
   

3. Environment Setup

   cp .env.example .env
   

   Configure the following in your .env file:

   • OPENAI_API_KEY - Your OpenAI API key
   • BRAVE_API_KEY - Your Brave Search API key
   • VOYAGE_API_KEY - (Optional) Your Voyage API key. Required for the reranker component when using the rag_agent.
   • DOC2X_API_KEY - (Optional) Your DOC2X API key. Required for the PDF parsing component when using the rag_agent.
   • OPENAI_BASE_URL - (Optional) Custom endpoint URL
   • POSTGRES_URL_NO_SSL - (Optional) Custom PostgreSQL URL without SSL. You can use the https://neon.com/ to create a free database. We will add the on-premise database support in the docker-compose.yml file soon.
   • POSTGRES_URL_NO_SSL_DEV - (Optional) Custom PostgreSQL URL without SSL for development.

1. Build the Docker Container

   chmod +x dev/setup.sh
   ./dev/setup.sh
   

2. Access the Container (Choose one method)

   • Option A: Use VSCode Remote Container (Recommended). You can run bash dev/install_vscode_extensions.sh to install the basic plugins for the development.
   • Option B: Access via terminal:

     docker exec -it "$(whoami)_sagi-dev" /bin/bash
     

3. Install Dependencies

   • Option A: pip install

   pip install -e .
   

   • Option B: uv install

   uv venv
   source .venv/bin/activate
   uv pip install -e .
   

4. Start CLI

   python cli.py
   

We welcome contributions! Please feel free to submit a pull request. We also welcome any questions via Github Issues or via the community group, please join us.

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.