Personal AI that talks to your docs, Emacs, and WhatsApp

What it does

Khoj is a personal AI platform that ingests your local files—PDFs, Markdown, Notion pages, Word docs, even org-mode—and lets you query them alongside live web search through any local or online LLM you choose. It wraps everything in a web app, desktop client, and plug-ins for Obsidian and Emacs, with WhatsApp and phone access added for good measure. You can also build custom agents with their own knowledge bases and personas, or automate research tasks like scheduled newsletters and smart inbox notifications.

The interesting bit

The project refuses to pick a lane: it aims to be both your private, offline-ready notebook assistant and your cloud-scale enterprise AI, scaling from a laptop to a server cluster without splitting the codebase. That breadth means you are not locked into a single interface or model provider, which is rare in a space where most tools force one or the other.

Key highlights

• Plugs into local and online LLMs (Llama 3, Qwen, Mistral, GPT, Claude, Gemini, DeepSeek) without forcing a subscription.
• Reads an unusually wide document menagerie: PDF, Markdown, Word, Notion, org-mode, and images.
• Surfaces everywhere: browser, desktop, phone, WhatsApp, Obsidian, and Emacs.
• Custom agents can carry their own knowledge, persona, model, and tool set.
• Runs fully offline or as a managed cloud service; enterprise on-premises and hybrid deployments are also available.

Caveats

• The README claims “excellent performance on modern retrieval and reasoning benchmarks” but never names the benchmarks or scores, so you will have to read the linked blog post for specifics.

Verdict

Developers who want a single self-hosted hub for searching their own documents, querying the web, and running automated agents across Obsidian, Emacs, or WhatsApp should take a look. If you just need a simple chatbot wrapper, this is overkill.