// the find
khoj-ai/khoj
Your AI second brain. Self-hostable. Get answers from the web or your docs. Build custom agents, schedule automations, do deep research. Turn any online or local LLM into your personal, autonomous AI (gpt, claude, gemini, llama, qwen, mistral). Get started - free.
Khoj is a self-hostable personal AI that indexes your local documents and the web, then lets you chat against them, build custom agents, and schedule automated research tasks. It works with any LLM backend — cloud or local via Ollama/llama.cpp — and has clients for browser, Obsidian, Emacs, desktop, and WhatsApp. The target audience is developers and knowledge workers who want a private, flexible alternative to ChatGPT with their own data in the loop.
The LLM backend abstraction is genuinely useful: swap between GPT-4, Claude, Gemini, and local Llama models without touching your data setup or workflows. The Obsidian and Emacs integrations are first-class, not afterthoughts — there's a full Elisp client with its own test suite. The automations feature (scheduled research → email digest) fills a real gap that most AI chat tools ignore entirely. At 35k stars with active CI, evals workflow, and a Docker path that actually works, this is production-quality open source, not a demo.
The Python Django backend means your self-hosted instance will be heavier than you expect — not something you drop on a 512MB VPS. Document sync is pull-based and polling; if you're used to Notion or Obsidian sync being instant, the latency will bother you. The Android client is a TWA wrapper around the web app, so offline use is essentially unsupported despite the 'on-device AI' framing. Search quality depends heavily on which embedding model you configure, but there's no guidance in the docs on which one works best for which use case — you're expected to experiment.