Khoj is a self-hostable AI second brain designed to augment user capabilities by retrieving information from the web and local documents. It targets individuals and enterprises seeking a personalized, autonomous AI assistant that can interact with various LLMs and integrate with common workflows.

How It Works

Khoj leverages advanced semantic search and retrieval-augmented generation (RAG) to provide answers from diverse data sources, including web pages and local files (PDF, Markdown, Word, Notion). It supports a wide array of local and cloud-based LLMs, allowing users to customize their AI's persona, tools, and knowledge bases to create specialized agents for research and automation.

Quick Start & Requirements

• Install: pip install khoj
• Prerequisites: Python 3.8+, Docker (for self-hosting).
• Demo: https://app.khoj.dev
• Docs: https://docs.khoj.dev

Highlighted Details

• Supports over 15 LLMs including GPT, Claude, Gemini, Llama, Mistral, Qwen.
• Processes various document types: PDF, Markdown, Word, Notion, Org-mode.
• Integrates with Browser, Obsidian, Emacs, Desktop, and WhatsApp.
• Enables creation of custom agents with tunable personality, tools, and knowledge bases.

Maintenance & Community

• Active development with a public contributors project board.
• Community support via Discord: https://discord.gg/BDgyabRM6e

Licensing & Compatibility

• License: MIT. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The README mentions an "experimental research mode" and "new start any message with /research," suggesting ongoing development and potential for breaking changes in these features.