Khoj
Khoj is an open-source AI second brain under AGPL-3.0 that answers from the internet and from your own documents, including PDFs, Markdown, Word and Notion files. It chats with any local or hosted model, runs semantic search over your files, builds custom agents with their own knowledge and tools, and automates research into scheduled newsletters. It is self-hostable, with a hosted cloud app and an enterprise edition from its makers.
What does Khoj do?
Khoj calls itself "your AI second brain." It chats with "any local or online LLM", naming Llama 3, Qwen, Gemma, Mistral, GPT, Claude, Gemini and DeepSeek, and gets answers "from the internet and your docs", covering images, PDF, Markdown, Org-mode, Word and Notion files. Semantic search finds relevant documents across what you have given it.
You can create agents with their own knowledge, persona, chat model and tools. Automations deliver personal newsletters and smart notifications to your inbox. It generates images and talks out loud. Clients cover the browser, Obsidian, Emacs, a desktop app, phones and WhatsApp, and the README says it scales from an on-device personal AI to a cloud enterprise AI. It states "Khoj is open-source, self-hostable. Always." The README does not say which web search engine it uses or describe citations.
Key facts
- Licence: AGPL-3.0, unmodified.
- Models: any local or online model; the README names Llama 3, Qwen, Gemma, Mistral, GPT, Claude, Gemini and DeepSeek.
- Answers from the internet and from your documents: images, PDF, Markdown, Org-mode, Word and Notion.
- Semantic search over your documents.
- Custom agents with their own knowledge, persona, chat model and tools; automations that email newsletters and notifications.
- Image generation and voice.
- Clients: browser, Obsidian, Emacs, desktop, phone and WhatsApp; a hosted cloud app at app.khoj.dev and Khoj Enterprise for cloud, on-premises or hybrid deployment.
- Latest release when read: 2.0.0-beta.28 on 2026-03-25; last push 2026-08-01; 37,396 stars on 2026-09-17.
What does it replace, and where does it fall short?
Khoj replaces Perplexity for someone whose questions reach into their own notes and files as well as the web, and ChatGPT for someone who wants custom agents and scheduled research emails on a model they choose. Its Obsidian and Emacs clients make it the pick for people who keep their knowledge in those tools on the open-source alternatives to Perplexity page.
Where it falls short: the README does not say which search engine it uses or describe citations, which are the core of Perplexity's answers; Vane and Morphic document both. Self-hosting instructions are in the documentation, not the README. The latest release is labelled beta, and the README now leads with a different product from the same team.
How does Khoj run?
The README points to the documentation for self-hosting rather than giving commands; its badges point to a Docker image and a PyPI package. Or use the hosted app at app.khoj.dev with no setup. Khoj Enterprise is offered as a cloud service, on-premises or hybrid; the README states no prices.
Who is Khoj for?
Someone who wants one assistant across their notes, documents and the web, reachable from Obsidian, Emacs or WhatsApp, with agents and scheduled research. Someone who wants a plain cited web answer engine should start with Vane.
What limits does the README state?
None stated in the README.
Questions people ask
Is Khoj open source?
Yes. Khoj is AGPL-3.0, unmodified, and the README says "Khoj is open-source, self-hostable. Always." Its makers also run a hosted app and sell an enterprise edition for cloud, on-premises or hybrid use.
Can Khoj search my own documents?
Yes. It answers from your documents as well as the internet, covering images, PDF, Markdown, Org-mode, Word and Notion files, with semantic search to find the relevant ones. That is its main difference from a web-only answer engine.
Which models does Khoj support?
Any local or online model, by the README, which names Llama 3, Qwen, Gemma, Mistral, GPT, Claude, Gemini and DeepSeek. Each custom agent can use its own chat model, so one agent can run locally while another uses a hosted model.
How does Khoj compare with Perplexity?
Khoj adds your own documents, custom agents and scheduled research emails to web answers, with clients in Obsidian, Emacs and WhatsApp. Perplexity is a hosted web answer engine with citations as its core. The open-source alternatives to Perplexity page compares Khoj with Vane and Morphic.
Where can I use Khoj?
In a browser, in Obsidian, in Emacs, in a desktop app, on a phone, or through WhatsApp, against either your own self-hosted server or the hosted app at app.khoj.dev. The README lists all six clients.
Sources
- Khoj README and AGPL-3.0 LICENSE: github.com/khoj-ai/khoj, read 2026-09-17.
Compared with the others
On the open-source alternatives to Perplexity page, Khoj is the pick for answers from your documents as well as the web. AGPL-3.0, any model, semantic search over your PDFs, Markdown, Word and Notion files, custom agents and scheduled research by email.
Also on that page: Vane for the closest perplexity shape, self-hosted, Morphic for a richer answer page with accounts.
On the open-source alternatives to ChatGPT page, Khoj is the pick for your notes, documents and the web together. AGPL-3.0 assistant that answers from your files and the web with any model, builds agents and emails scheduled research, with clients for Obsidian, Emacs and WhatsApp.
Also on that page: Open WebUI for a private chatgpt for a whole team, LibreChat for every provider in one mit platform, AnythingLLM for chatting with your documents.
On the open-source alternatives to Glean page, Khoj is the pick for notes, files and the web together. AGPL-3.0, answering from your documents and the internet with any model, with agents, scheduled research and clients for Obsidian and WhatsApp.
Also on that page: Onyx for company-wide search with agents, Open WebUI for documents plus roles and single sign-on, AnythingLLM for document chat with citations.
More answer engines
An MIT answer engine, formerly Perplexica, that searches the web through a bundled SearXNG and answers with cited sources, using local models through Ollama or hosted ones, in speed, balanced and quality modes, on your own hardware.
An Apache-2.0 AI search engine with a generative interface that answers with citations, using OpenAI, Anthropic, Google, Ollama or any OpenAI-compatible model and Tavily, SearXNG, Brave or Exa for search, with sign-in and shareable results.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/khoj-ai/khoj.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.