Khoj review covering its open-source AI second brain, self-hosting, supported platforms, and how it compares to NotebookLM and Obsidian AI plugins.
Khoj is an open-source, self-hostable personal AI application founded by Debanjum Singh and Saba Imran, backed by Y Combinator, and built as an 'AI second brain' that lets users search and chat with their own notes and documents alongside the open web. The project is distributed under the AGPL-3.0 license and has attracted a large open-source community, with tens of thousands of GitHub stars.
Functionally similar in concept to Google's NotebookLM, Khoj indexes personal documents into semantic embeddings and answers natural-language questions grounded in the user's own files, while additionally supporting custom AI agents, scheduled research automations, image generation and text-to-speech, all of which can run against either local open models or hosted commercial LLMs.
Khoj indexes a wide range of document types, including images, PDFs, Markdown, Org-mode, Word documents and Notion exports, and lets users query them conversationally through semantic search. Users can build custom agents with distinct personas, knowledge bases, chat models and tool access, and schedule automated research tasks with smart notifications.
The platform is model-agnostic, supporting both local on-device models (Llama 3, Qwen, Gemma, Mistral, DeepSeek) and hosted commercial models (GPT, Claude, Gemini), giving users control over privacy and cost trade-offs. Khoj is accessible via browser, desktop app, mobile, and native integrations for Obsidian, Emacs, and WhatsApp.
Khoj's self-hosted version is free and open-source, aside from the user's own hardware or hosting costs, reflecting its privacy-first, community-driven design. The project's official hosted cloud service, previously offered at app.khoj.dev with a free trial and paid plans including student discounts, has since been deprecated.
Users who want a cloud-hosted deployment today typically self-host on third-party infrastructure such as Railway, rather than through an officially maintained Khoj-run cloud service, since the company no longer offers its own hosted subscription.
Yes, the self-hosted version of Khoj is free and open-source, aside from the user's own hardware or hosting costs. The official hosted cloud service has since been deprecated.
Khoj was founded by Debanjum Singh and Saba Imran, and is backed by Y Combinator.
Khoj can index and search images, PDFs, Markdown, Org-mode files, Word documents and Notion exports, alongside general web search.
Yes, Khoj supports both local open models such as Llama 3, Qwen, Gemma and Mistral, and hosted commercial models like GPT, Claude and Gemini.
Khoj is accessible via web browser, desktop app, mobile, and has native integrations with Obsidian, Emacs and WhatsApp.
Khoj offers similar retrieval-augmented-generation functionality over personal documents, but unlike NotebookLM it is open-source, self-hostable and model-agnostic rather than a closed, hosted-only Google product.