Key Takeaways
- AGPL 3.0 license — free and open-source, but copyleft with a network-use clause (explained in detail below)
- Self-hosting is now the only supported way to run Khoj: Khoj Cloud (app.khoj.dev), the paid hosted version, was sunset on April 15, 2026
- The core open-source project is not affected — GitHub commits continued past the cloud shutdown, and the repository is not archived
- Syncs Obsidian vaults through a dedicated plugin, and Notion workspaces through direct sync from the web app
- Chats with local models (through Ollama, llama-cpp-server, or any OpenAI-API-compatible server) or cloud models (OpenAI, Anthropic, Google, and others) side by side
- Also does semantic document search, custom agents with scoped knowledge, scheduled research automations, image generation, and text-to-speech
📍 In One Sentence
Khoj is an open-source (AGPL 3.0), self-hostable personal AI assistant that indexes your notes, documents, and Notion workspace so you can chat with and search them using local or cloud LLMs, with Obsidian, Emacs, desktop, and WhatsApp clients.
💬 In Plain Terms
Point Khoj at your Obsidian vault, your PDFs, or your Notion workspace, and it builds a searchable index you can ask questions against — the same idea as Notion AI or a document chatbot, except the code runs on your own machine and stays free forever.
📌Note: Do not sign up expecting a hosted Khoj Cloud account — that product no longer exists. Every current Khoj deployment is self-hosted.
Is Khoj Still Maintained?
Yes, with one important change: Khoj Cloud, the paid hosted version at app.khoj.dev, was sunset on April 15, 2026. The team behind Khoj stated that the underlying project "remains fully open-source" and pointed users to self-hosting instead, while the company shifted its commercial focus to two newer products, an academic research tool and an AI co-worker for automation tasks.
The open-source repository itself shows no sign of abandonment: it is not archived, commits have continued past the cloud shutdown date, and the issue tracker is active. That makes Khoj a different situation from a project whose company wound down entirely — here, only the convenience of a hosted, pay-for-it version disappeared. Self-hosting was always an option and is now the default one.
Khoj Cloud sunset
- Date:
- 2026-04-15
- What it means:
- app.khoj.dev hosted service stopped; no new signups or paid plans
Open-source repo status
- Date:
- ongoing
- What it means:
- Not archived; commits and releases continue on GitHub
Company focus shift
- Date:
- 2026
- What it means:
- Team now also builds a research-paper tool and an automation-focused AI agent, alongside Khoj
📌Note: Ignore any pricing page or sign-up flow that still references Khoj Cloud — it is discontinued. The only supported path today is self-hosting.
What Is Khoj?
Khoj is an open-source (AGPL 3.0 license, github.com/khoj-ai/khoj) personal AI application that indexes your own documents and lets you chat with them, search them, and build automations on top of them. The GitHub repository has passed 36,000 stars. It is designed to scale from a single person running it on a laptop to a self-hosted deployment shared across a small team.
- Chat: talk to local or cloud LLMs about your own notes and documents, with citations back to the source file
- Search: semantic (meaning-based, not keyword) search across PDFs, Markdown, Notion pages, Word documents, org-mode files, and images
- Agents: create custom agents with a scoped knowledge base and a defined persona for a specific task
- Automations: schedule recurring research tasks that deliver a summary or notification on a schedule
- Extras: image generation and text-to-speech are built in
- Clients: browser web app, Obsidian plugin, Emacs client, desktop app, phone/PWA, and a WhatsApp integration
What Does Khoj's AGPL 3.0 License Actually Mean?
Khoj is licensed under the GNU Affero General Public License version 3 (AGPL 3.0) — a copyleft license, not a permissive one. This is a meaningfully different license from the Apache 2.0 or MIT licenses used by most tools this article compares Khoj to, and the difference matters for two different groups of people in opposite ways.
If you just self-host Khoj for yourself or your team
The AGPL's copyleft obligations are triggered by distributing a modified version, or by running a modified version as a network service that other people (outside your own use) interact with. Running Khoj unmodified, or with private tweaks, purely for your own personal or internal use, does not require you to publish anything. This is the normal case for almost every reader of this article.
If you modify Khoj and offer it to others as a service
This is where AGPL differs sharply from the permissive licenses (Apache 2.0, MIT) used by tools like AnythingLLM or Open WebUI. Under a permissive license, a company can take the code, modify it, and sell access to a closed, proprietary hosted version without ever releasing the changes — that is legal and common. AGPL Section 13 closes that path for network software: if you run a modified version of Khoj and let other people interact with it over a network (for example, as your own paid SaaS product), you must make your modified source code available to those users. This is often called the "SaaS loophole" that GPL left open and AGPL closes.
Does Khoj Support Local LLMs, RAG-Style Search, and Agents?
Yes to all three, and they are the core of what Khoj does.
- Local LLMs: point Khoj at Ollama, llama-cpp-server, vLLM, or any OpenAI-API-compatible local server through the admin panel's AI Model API settings
- Cloud LLMs: add API keys for OpenAI, Anthropic, or Google (and others) to chat with hosted models through the same interface, switching between local and cloud per conversation
- Document search: Khoj indexes PDFs, Markdown, Notion pages, Word docs, org-mode files, and images into a searchable, semantic index — the same retrieval pattern used by dedicated RAG tools, but built around personal knowledge management rather than business documents
- Obsidian sync: the Obsidian plugin keeps your vault indexed automatically as you edit; the Emacs client and desktop app do the same for their respective file sets
- Notion sync: your Notion workspace is synced directly from the Khoj web app, rather than through a live plugin
- Agents: define a custom agent scoped to a subset of your knowledge base, with its own system prompt and personality, for a repeated task (e.g., a research assistant that only searches your work notes)
How Do You Self-Host Khoj?
Self-hosting is the only supported path now that Khoj Cloud is gone. Khoj documents two installation routes: Docker (recommended for most people) and a pip package for Python users who want a local install without containers.
- 1Install Docker and Docker Compose if you do not already have them.
- 2Download the docker-compose.yml file from github.com/khoj-ai/khoj, and set KHOJ_ADMIN_PASSWORD, KHOJ_DJANGO_SECRET_KEY, and any LLM provider API keys (OpenAI, Anthropic, Gemini) you plan to use in the environment file.
- 3To use a local model instead of (or alongside) a cloud one, uncomment and set OPENAI_BASE_URL to point at Ollama or another OpenAI-API-compatible server running on your host.
- 4Start the stack with docker-compose up from the directory containing your configuration file.
- 5Open http://localhost:42110, sign in with the admin credentials you set, and connect your Obsidian vault (via the Obsidian plugin), your Notion workspace (via direct sync in the web app), or drag files in directly.
- 6Alternative for Python users: skip Docker and run python -m pip install 'khoj[local]', then start it with USE_EMBEDDED_DB="true" khoj --anonymous-mode. Add CMAKE arguments during install for GPU acceleration (CUDA, ROCm, Metal, or Vulkan) if you plan to run local models.
What hardware does self-hosted Khoj need?
The documented minimums are 8 GB RAM and 5 GB of free disk space. A GPU (Nvidia, AMD, or Apple Silicon M1 or later) is optional but speeds up local model inference and semantic indexing; Khoj's own guidance suggests 16 GB of VRAM if you want to run larger local models comfortably.
Can I run Khoj entirely offline?
Yes, if you configure it to use a local model server (Ollama, llama-cpp-server, or vLLM) instead of a cloud provider. Search and chat against your indexed documents then run entirely on your own hardware with no external API calls.
Who Should Use Khoj?
Khoj fits a specific reader: someone who wants a personal, searchable AI layer over notes and documents they already keep in Obsidian, Notion, or plain files, and who is comfortable running the server themselves.
Use Khoj if
- You keep notes in Obsidian and want an AI chat/search layer over your vault without leaving your files in someone else's cloud
- You use Notion and want to search or chat with that workspace using a model you choose, rather than only Notion's built-in AI
- You want to switch between local models (for privacy or cost) and cloud models (for capability) inside one interface
- You are comfortable self-hosting via Docker or pip, and reading a license (AGPL 3.0) before deciding whether to build a commercial product on top of the code
Avoid Khoj if
- You want a fully hosted, zero-setup product — Khoj Cloud is gone, and self-hosting requires basic comfort with Docker or Python
- You need a permissively licensed (Apache 2.0/MIT) codebase because you plan to build and resell a closed, modified hosted version of the tool
- Your only knowledge base is Notion and you are fully satisfied with Notion AI's built-in search and chat
- You need enterprise vendor support, an SLA, or a compliance-ready managed hosting option — none of that currently exists for Khoj
Khoj vs. Alternatives
Khoj's closest comparisons are other self-hosted document-chat and RAG tools, plus the built-in AI features of the note apps it syncs with.
Tool | Interface | License | Best For | Maintenance |
|---|---|---|---|---|
| Khoj | Chat/search + Obsidian/Emacs plugins | AGPL 3.0 | Personal notes + Notion sync | Active |
| AnythingLLM | Desktop/web RAG app | MIT | General document RAG | Active |
| Open WebUI | Web chat UI | BSD-3 | Ollama-based chat frontend | Active |
| Notion AI | Built into Notion | Proprietary | Notion-only, no self-hosting | Active (cloud SaaS) |
Common Mistakes When Evaluating Khoj
These mistakes come from carrying assumptions over from Khoj Cloud, or from confusing AGPL with the permissive licenses used by most similar tools.
Mistake 1: Trying to sign up for Khoj Cloud
app.khoj.dev no longer accepts new hosted accounts — the service was sunset on April 15, 2026. If a search result or pricing page still points you there, treat it as stale and go straight to self-hosting via docs.khoj.dev.
Mistake 2: Assuming AGPL 3.0 works like Apache 2.0 or MIT
Apache 2.0 and MIT let anyone fork the code into a closed, proprietary product with no obligation to publish changes. AGPL 3.0 does not — if you modify Khoj and offer the modified version as a network service to others, you must release your modified source. Self-hosting privately for yourself or your own team is unaffected either way.
Mistake 3: Expecting a live two-way Notion sync like the Obsidian plugin
The Obsidian integration is a dedicated plugin that keeps your vault indexed as you edit. Notion sync currently works differently — it pulls your workspace directly from the Khoj web app rather than through a persistent background plugin, so treat it as a sync-on-demand step, not an always-on live connection.
Frequently Asked Questions
Is Khoj still maintained?
Yes. The open-source Khoj project on GitHub is actively developed, with commits continuing past the April 15, 2026 shutdown of Khoj Cloud, the paid hosted version. Only the hosted product was discontinued — the self-hosted software was not.
What happened to Khoj Cloud?
Khoj Cloud, the hosted version at app.khoj.dev, was sunset on April 15, 2026. The team stated the underlying project "remains fully open-source" and pointed users to self-hosting via docs.khoj.dev instead.
What license is Khoj released under?
AGPL 3.0 (GNU Affero General Public License version 3), a copyleft license. Self-hosting privately does not obligate you to publish anything; offering a modified version of Khoj as a network service to others does require releasing your modified source, under AGPL Section 13.
Is AGPL 3.0 the same as GPL?
No. Standard GPL only requires releasing source if you distribute the software itself. AGPL adds a network-use clause: running a modified version as a service others interact with over a network also triggers the source-release requirement, closing what is sometimes called the SaaS loophole.
Does Khoj sync with Obsidian?
Yes, through a dedicated Obsidian plugin that keeps your vault indexed automatically as you edit notes.
Does Khoj sync with Notion?
Yes. Your Notion workspace can be synced directly from the Khoj web app, which is a different mechanism than the always-on Obsidian plugin — treat it as a sync step rather than a live background connection.
Can Khoj use local LLMs?
Yes. Khoj connects to Ollama, llama-cpp-server, vLLM, or any OpenAI-API-compatible local server through the admin panel, so chat and search can run entirely on your own hardware without external API calls.
What are the minimum hardware requirements to self-host Khoj?
Khoj's documented minimums are 8 GB RAM and 5 GB free disk space. A GPU is optional; the project suggests 16 GB of VRAM if you want to run larger local models comfortably.
How is Khoj different from Notion AI?
Notion AI is a proprietary, cloud-only feature built into Notion itself. Khoj is a separate, self-hosted, open-source application that can sync with Notion (and Obsidian, and plain files) and lets you choose which LLM — local or cloud — answers your questions.
Is Khoj free to use?
The self-hosted, open-source version is free — there is no license fee to run it. You are responsible for your own hosting, and for any API costs if you choose to connect a paid cloud LLM provider instead of a local model.