Khoj is a self-hostable, open-source AI second brain written in Python that turns any local or online large language model into a personal research, automation, and agent platform for individuals, small teams, and enterprises that want their documents and queries kept under their own control.

What it is

Khoj is a personal AI application that lives in the Python ecosystem and is distributed as a PyPI package and a Docker image. It sits in the note-taking and knowledge-management category, and it scales deliberately: the README describes the same project running on a single person's computer as an on-device assistant and serving cloud-scale enterprise deployments. It ships both as a self-hosted application and as a hosted service at app.khoj.dev, and its clients span a browser interface, Obsidian, Emacs, desktop, phone, and WhatsApp.

The problem it solves is fragmentation. Instead of running one tool to chat with a model, another to index personal notes, and a third to schedule recurring research, Khoj combines retrieval, generation, and scheduling in one place. It answers from the public internet and from a personal corpus that can include PDFs, Markdown, Notion pages, Word documents, org-mode files, and images, using semantic search rather than keyword matching. For anyone already keeping notes in Obsidian or Emacs, that means querying the material in place rather than exporting it to another service.

Key capabilities

• Chat with any local or online LLM, including llama3, qwen, gemma, mistral, gpt, claude, gemini, and deepseek.
• Ground answers in web results and personal documents, with support for PDF, Markdown, Notion, Word, and org-mode files plus images.
• Create agents with custom knowledge, persona, chat model, and tools so a single assistant can take on a specific role.
• Schedule automations that handle repetitive research, deliver personal newsletters, and send smart notifications to an inbox.
• Find relevant documents through advanced semantic search.
• Generate images, speak answers aloud, and play back messages.
• Reach the same account from a browser, Obsidian, Emacs, desktop, phone, or WhatsApp.

Who uses it and how

• Individuals who run Khoj privately on their own computer against a local model, keeping both documents and prompts on their own hardware.
• Teams and individuals who want the product with no setup and use the hosted cloud app at app.khoj.dev instead of self-hosting.
• Enterprises that need the same assistant as a cloud service, on-premises, or as a hybrid deployment.
• Researchers and knowledge workers who automate recurring literature or news gathering as scheduled agents rather than repeated manual searches.
• Obsidian and Emacs users who query existing vaults and notes in place through the native clients.

Getting started

Self-hosting starts from the setup guide at docs.khoj.dev/get-started/setup, with the project also published as the khoj package on PyPI and as a container image for Docker. Anyone who would rather run nothing can use the hosted app at app.khoj.dev with no setup.

How it compares

The facts provided name no paid products that Khoj replaces, so it stands alone in this registry as a self-hostable AI second brain. ChatGPT appears in the topic list as one of the models Khoj can talk to rather than as a competing product, and Obsidian and Emacs appear as clients it plugs into rather than alternatives it displaces.

When to use it — and when not to

A self-hoster takes on real operational work: running the application, keeping the document index current, and connecting language model endpoints, which the README defers to the setup documentation rather than reducing to a single command. The AGPL-3.0 licence matters for anyone who wants to embed Khoj inside a closed-source product, and the README itself stays high-level, pointing to the docs for the full feature list. Anyone who wants nothing at all to operate should use the hosted app rather than self-host.