What it does Khoj is a personal AI layer that sits between you and any local or online LLM. It ingests your documents—PDFs, Markdown, Notion, Word, org-mode, images—and lets you query them via chat, semantic search, or scheduled automations. You reach it through a browser, Obsidian, Emacs, desktop, phone, or WhatsApp. Self-hosting is the default posture; a managed cloud version exists if you’d rather not.

The interesting bit The Emacs and Obsidian integrations are unusual. Most “AI second brain” projects chase the Notion or Roam crowd; Khoj also speaks fluent org-mode and will ping you on WhatsApp. The agent builder lets you pin custom knowledge, personas, and tools to specific tasks—less “chatbot,” more “intern with a reading list.”

Key highlights

• Plugs into local models (llama3, qwen, mistral, etc.) or API ones (GPT, Claude, Gemini, DeepSeek)
• Reads PDFs, images, Markdown, Word docs, Notion exports, and org-mode files
• Access via browser, Obsidian plugin, Emacs package, desktop app, phone, or WhatsApp
• Custom agents with their own knowledge bases, personas, and tool access
• Scheduled research automations: newsletters, notifications, recurring queries
• Semantic search across your ingested documents
• Image generation, voice I/O, and audio playback
• Dockerized self-hosting with docs; cloud option at app.khoj.dev

Caveats

• The README is heavy on feature lists and light on architecture or performance specifics; “excellent performance on modern benchmarks” is claimed but no numbers are shown
• Enterprise and hybrid deployment details live off-repo on a marketing page
• The project also promotes Pipali, a separate “AI coworker” repo, which suggests the boundary between Khoj and its sibling products is still being defined

Verdict Good fit if you want a single self-hosted backend that feeds LLM access into Emacs, Obsidian, and WhatsApp without writing glue code. Skip if you need transparent benchmarking data or a minimal, single-purpose tool—this is a broad kitchen-sink system.