Khoj

Khoj positions itself as a personal AI second brain: it indexes PDFs, Markdown, org-mode, Word, and Notion content, then answers questions over that corpus and the web through semantic search and chat. Users build custom agents with their own knowledge bases, tools, and personas, schedule automations such as newsletters and smart notifications, run multi-step research sessions, and generate images or hear responses through text to speech. Models are pluggable: local ones through Ollama including llama3, qwen, gemma, and mistral, or hosted ones like GPT, Claude, Gemini, and DeepSeek. Clients span the browser, a desktop app, an Obsidian plugin, an Emacs package, phones, and WhatsApp. Self-hosting works via Docker or pip with no GPU required unless running local models, and a managed cloud at app.khoj.dev is free to use, with enterprise on-premises and hybrid deployments sold through khoj.dev/teams. The code is AGPL-3.0 and the repository holds 36.1k stars; the YC-backed project is popular with note-takers who want private, searchable memory across their files.