Khoj

Meant to extend what one person can do, Khoj runs as a personal AI app that scales from a laptop up to a deployment serving a whole company. The gap it closes is the one between what your documents know and what the internet knows.

It talks to whichever model you point it at, local or hosted, naming llama3, qwen, gemma, mistral, gpt, claude, gemini and deepseek among them. Access is not tied to one window either: the same assistant is reachable from a browser, Obsidian, Emacs, a desktop app, a phone or WhatsApp.

The features cluster around finding, asking and automating.

• Document grounding: answers drawn from images, PDFs, Markdown, org mode, Word and Notion files.
• Semantic search: relevant documents surfaced quickly rather than by exact keyword.
• Custom agents: agents given their own knowledge, persona, chat model and tools.
• Automations: repetitive research handled on a schedule, delivering personal newsletters and notifications to your inbox.
• Generation: images created from prompts, messages played back, replies spoken aloud.

Khoj is open source and self hostable, with setup documented for running it privately on your own computer, and a hosted app is available for anyone who would rather skip installation entirely. For organizations, the project is offered as a cloud service, on premises or as a hybrid arrangement.