Gurubase provides an open-source Retrieval Augmented Generation (RAG) system for creating AI-powered Q&A assistants, or "Gurus," from various data sources. It targets developers and organizations looking to enhance their documentation, support, or knowledge bases with an "Ask AI" feature, offering instant, referenced answers and reducing hallucinations.

How It Works

Gurubase employs a RAG architecture involving indexing, embedding, and retrieval. Data sources (web pages, PDFs, YouTube, GitHub repos, Jira, Zendesk) are processed and chunked, then converted into vector embeddings stored in Milvus. When a question is asked, relevant context is retrieved from Milvus, and an LLM generates an answer, with an evaluation mechanism to minimize hallucinations. This approach allows for accurate, context-aware responses grounded in the provided data.

Quick Start & Requirements

• Install via curl -fsSL https://raw.githubusercontent.com/Gurubase/gurubase/refs/heads/master/gurubase.sh -o gurubase.sh && bash gurubase.sh.
• Detailed instructions are in INSTALL.md.
• Minimum requirements: 4 CPU cores, 8GB RAM, 10GB SSD storage, Linux or macOS (WShell2 for Windows).
• See Gurubase Documentation for more.

Highlighted Details

• Supports web pages, PDFs, YouTube videos, GitHub repositories, Jira issues, and Zendesk tickets/articles as data sources.
• Offers integrations via a website widget, Slack bot, Discord bot, and GitHub bot.
• Features "Binge" for visualizing and navigating learning paths.
• Provides a self-hosted option for full deployment control.
• Microservices architecture: Next.js frontend, Django backend, Milvus vector store, RabbitMQ, Redis, PostgreSQL.

Maintenance & Community

• Active community via Discord.
• Follow on X, Mastodon, and Bluesky.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Content generated aligns with the license of the datasources used.

Limitations & Caveats

Currently, only the Gurubase team can create new Gurus on Gurubase.io; users must submit requests via GitHub issues. While self-hosting is supported, the initial quick install script is a basic entry point, with detailed setup and upgrades managed via INSTALL.md. Periodic reindexing for all data sources is planned but not yet implemented.