Khoj hosting

A personal AI that searches your own notes, documents and email.

AI assistants & chatAGPL-3.0from $11.99/mo
In short

Khoj hosting at AIHostBox means a dedicated, single-tenant Khoj instance on our infrastructure: we install it, pin it to a reviewed release, put it behind TLS on your domain, back it up nightly and keep it patched. Plans start at $11.99 per month and the admin account, the data and any API keys are yours.

Notes searchPDF & MarkdownSemantic searchAutomationsLocal models

Khoj is an open-source “AI second brain”, licensed AGPL-3.0 and built by the khoj-ai team. It indexes the documents you give it — PDF, Markdown, org-mode, Word, images and Notion pages — and lets you chat with a local or hosted model against that index, mixing in web results when a question needs them. Most chat front-ends treat documents as an add-on; Khoj is organised around your own notes, with semantic search as a first-class feature and clients for Obsidian, Emacs, the desktop and WhatsApp that keep the index fed from where you already write.

The self-hosted stack is more than one process. The official compose file runs a Django server, Postgres with the pgvector extension for embeddings, a SearXNG instance for web search and a Terrarium sandbox so the model can run Python during analysis. Agents with their own persona, knowledge, chat model and tools, and scheduled automations that rerun research on a timetable, only work if something is awake when they fire. That is the case for a server: the index, the schedules and the sync endpoints stay reachable from every device you use.

Khoj hosting plans

PlusRecommended minimum
4 GB RAM · 2 vCPU · NVMe
$11.99/mo
Order
ProTeams & heavier load
8 GB RAM · 3 vCPU · NVMe
$18.99/mo
Order

Every plan is a dedicated single-tenant instance with NVMe storage, nightly off-box backups, TLS and managed patching. Quarterly, semi-annual and annual terms are discounted at checkout. Model tokens are not included — you bring your own key.

How we run Khoj

The Khoj server container (ghcr.io/khoj-ai/khoj) on port 42110, Postgres with pgvector for accounts, chats and embeddings, SearXNG for web search and the Terrarium sandbox for code execution, following the project’s docker-compose.

What we configure for you

  • KHOJ_DJANGO_SECRET_KEY and the admin account (KHOJ_ADMIN_EMAIL, KHOJ_ADMIN_PASSWORD) generated per instance
  • KHOJ_DOMAIN and KHOJ_ALLOWED_DOMAIN set to your hostname so the admin panel’s CSRF checks pass behind TLS
  • KHOJ_SEARXNG_URL and KHOJ_TERRARIUM_URL wired to the bundled search and sandbox containers
  • Your model keys: OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, or OPENAI_BASE_URL for a self-hosted runtime
  • Multi-user login through Google OAuth or Resend magic links when you want more than one account
  • KHOJ_TELEMETRY_DISABLE if you prefer no usage telemetry

Which plan fits

Plus (4 GB, 2 vCPU) suits one to a few people with a few thousand notes and PDFs, with chat going to a hosted model. Pro (8 GB, 3 vCPU) is the plan for a shared team instance or large libraries, because document embedding runs on the server’s CPU during indexing while SearXNG and the sandbox share the same memory.

The honest limits

Khoj’s own docs suggest 8 GB RAM and a GPU for running chat models inside Khoj; on our CPU plans chat goes to your API key or a separate Ollama plan instead. Indexing a large PDF library for the first time is slow on CPU, and the upstream computer-use operator container is not part of the stack.

What people run Khoj for

Second brain

Ask questions across years of notes instead of grepping folders.

Research assistant

Point it at a document library and interrogate it in plain language.

Obsidian and Emacs sync

Keep the index current from the editor you already write in, on every device.

Scheduled research

Have an automation rerun a query on a timetable and collect what changed.

Licence and version policy

ItemPosition
Upstream licenceAGPL-3.0
Sold as managed hostingYes
TenancySingle tenant — one customer per instance
PatchingCritical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Reading on Khoj

Works well with Khoj

More AI assistants & chat we host

Questions

Khoj hosting: your questions

Which chat models can Khoj use?
Khoj talks to OpenAI, Anthropic and Gemini with your own API key, and to any OpenAI-compatible server through OPENAI_BASE_URL, so an Ollama, llama.cpp or vLLM endpoint works as well. Chat models are added in the admin panel, and each agent you create can use a different one.
How do my documents get into a hosted Khoj?
Drag and drop files into the web app, connect a Notion workspace, or install the Obsidian plugin, Emacs package or desktop app and point them at your folders so changes upload to your instance. Supported formats include PDF, Markdown, org-mode, Word, plain text and images.
Can several people log in to one Khoj instance?
Yes. In multi-user mode Khoj supports Google sign-in (GOOGLE_CLIENT_ID and GOOGLE_CLIENT_SECRET) and email magic links sent through Resend (RESEND_API_KEY, RESEND_EMAIL). Each person gets their own chats and indexed files, while the admin account created at install manages models and settings at /server/admin.
What does a Khoj backup actually contain?
The state that matters sits in Postgres: accounts, chat history, agents, automations and the pgvector embeddings of your indexed documents. Nightly off-box backups capture that database along with Khoj’s config volume, so a restore brings the index back without re-uploading and re-embedding every file.
Can I move from my own Khoj Docker install?
Yes. Send us a pg_dump of your Khoj database and we restore it onto your instance, so users, conversations and indexed entries arrive intact. Keep the same search and embedding model settings you used before; if the model changes, existing vectors no longer match new queries and you will need to re-index.

Run Khoj without running a server

A dedicated Khoj instance, deployed and maintained by us. Month to month, cancel whenever.