The archive · AI & Models · Product decision · 2023–2026
Khoj's open-source second-brain bet: self-host your AI, 35k GitHub stars
Ex-Microsoft AI builders make an open-source personal assistant for your own docs and notes: AGPL self-host or cloud, YC W24, 35.5k GitHub stars by June 2026.
Khoj
What the business is
Khoj is an open-source (AGPL-3.0) personal AI application that runs semantic search and chat over a user's own documents — PDFs, Markdown, org-mode, Word, Notion — and connects any local or cloud LLM (Ollama, GPT, Claude, Gemini, DeepSeek). Clients span browser, desktop, phone, Obsidian, Emacs and WhatsApp; the product adds custom agents, scheduled research automations, a code-execution sandbox and privacy-first web search, with self-hosting free and enterprise cloud, on-prem or hybrid sold on top.
How it started
The two founders met while building enterprise personal AI assistance at Microsoft, where the team scaled it to tens of millions of daily active users, and concluded no equivalent existed for individuals. Khoj launched as an open-source project in 2023, joined Y Combinator's W24 batch, and announced itself to the batch with 'Your superhuman companion', pitching an AI that can answer questions from an outdated insurance PDF or a stray note from a hundred standups ago.
What happened
Khoj grew from doc search into a personal agent platform: custom agents with knowledge bases and personas, scheduled newsletters and smart notifications, image generation, text-to-speech, and deep research combining SearxNG web search with a Terrarium code sandbox. By 2026 the self-host path became the stated center: Railway's 2026-04-26 template page says the hosted Khoj Cloud is being deprecated and self-hosting is the primary deployment path, while the repository keeps promising enterprise cloud, on-prem and hybrid offerings.
How it ended up
Still running and expanding as of 2026-09-06: the repository remained active (last update 2026-06-24 in the DEV.co snapshot), the README promises 'open-source, self-hostable. Always', and the company sells cloud and enterprise tiers; the reviewed sources disclose no revenue or funding beyond YC.
Background
Khoj, a Y Combinator W24 company founded by two former Microsoft enterprise-AI builders, sells an open-source personal AI that searches and chats over a user's own documents — PDFs, notes, org-mode files, Notion exports — using any local or cloud LLM. Its bet, visible in the repo's 'open-source, self-hostable. Always' promise, is that personal AI belongs on your own data rather than in a cloud walled garden.
The product spreads through the tools people already use: an Obsidian plugin, an Emacs package, desktop and phone apps, a browser, and WhatsApp access, with semantic search over files the cloud never sees. On top it adds custom agents, scheduled research and newsletters, a code sandbox and privacy-first web search, all under AGPL-3.0 with one-command Docker self-hosting.
Traction is measured in stars and surfaces rather than revenue: 34k+ GitHub stars on 2026-04-26 per Railway, 35.5k with 2.3k forks per the 2026-06-24 DEV.co snapshot. By 2026 the strategy converged on self-hosting as the primary path — Railway's template page notes the hosted Khoj Cloud is being deprecated — with enterprise cloud, on-prem and hybrid as the paid layers.
What has to be true
- People's most valuable data — PDFs, meeting notes, old emails — is exactly what cloud chatbots cannot see, so the wedge needs no education.
- An Obsidian or Emacs plugin lets Khoj ride an existing knowledge-management habit instead of asking users to migrate.
- AGPL self-hosting turns the privacy argument into code: the user's data never leaves infrastructure they control.
- Multi-LLM support removes the model-vendor risk, so the product stays useful as frontier models change.
What can be applied
A personal AI's moat is not the model but permission to sit on your data — shipping an AGPL self-host path first builds the trust a cloud chat can never earn.
Aftermath
As of 2026-09-06 Khoj is active with a clear direction: the hosted cloud is being deprecated in favor of self-hosting as the primary path, with enterprise on-prem, hybrid and cloud tiers still offered. Its 35.5k-star community (2026-06-24) and multi-surface clients give it distribution most personal-AI startups lack, but the reviewed sources disclose no revenue, funding beyond YC or paying-customer counts. The open question is whether individuals will run their own AI as cloud rivals improve, or whether the enterprise tier becomes the real business.
Sources
- Khoj: Self-Hosted AI Assistant & RAG Platform (35.5k stars, 2026-06-24 snapshot)
- Deploy & Host Khoj | Open Source Personal AI Assistant (34k+ stars, YC W24)
- Khoj: AI for humans — Y Combinator company profile
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card