Can Khoj use a 100% local LLM without external API calls?

Yes. Install Ollama on the same VPS or a machine on your network, then configure Khoj to point to your Ollama endpoint. No data leaves your infrastructure.

How does first access work?

ServOrbit creates an initial admin account with email admin@khoj.local and a random password visible in your client area. Log in to configure your LLM, then create your own user account if you need multiple users.

What file types can Khoj index?

Khoj indexes Markdown/Obsidian notes, PDFs, plain text files, web pages and emails (IMAP). It also extracts content from images embedded in your documents if you have configured a vision model.

What is the difference from AnythingLLM or PrivateGPT?

AnythingLLM and PrivateGPT focus on pure RAG (question → document). Khoj goes further with autonomous agents: they can browse the web, chain multiple steps and aggregate external sources in addition to your local knowledge base. Native Obsidian integration is also a key differentiator.

Build, host and run AI solutions.

Khoj

Khoj — self-hosted personal AI assistant: semantic search on notes and docs, autonomous agents, Obsidian and browser extension support.

4 GB minimum (8 GB recommended) RAM 2 vCPUs minimum Port 42110 In validation

In validation

Automatic installation for this solution is ready and currently going through our tests on a real server. Ordering will open as soon as validation is complete.

Tech stack

Docker ComposePythonPostgreSQL 15 + pgvectorFastAPI

Minimum RAM4 GB minimum (8 GB recommended)

Minimum CPU2 vCPUs minimum

Default port42110

Compatible OSChoice of Linux distributions

Khoj is an open-source AI assistant (AGPL-3.0) you self-host on your own VPS. It automatically indexes your Obsidian notes, PDF documents and text files, letting you ask natural language questions and trigger autonomous agents that can search the web. Access it from the web interface, Chrome and Firefox browser extensions, or the Obsidian plugin.

Key features

Semantic search on your notes, PDFs and documents — no cloud involved

Autonomous agents that browse the web and aggregate sources

Native Obsidian integration, Chrome and Firefox extensions, web interface

Works with local LLMs via Ollama or OpenAI / Anthropic APIs

PostgreSQL + pgvector backend: fast indexing and vector search

Multimodal: ask questions about images embedded in your notes

When to use this solution?

1

Technical knowledge base

Query your Obsidian vault of technical documentation without leaving your IDE. Khoj finds architecture decisions, useful commands and meeting notes by meaning, not by keyword.

2

Legal or HR document research

Load your contracts, internal procedures and regulatory texts. Ask precise questions about your data without exposing any document to a third-party service.

3

News monitoring and synthesis

Khoj agents can query the web, aggregate multiple sources and deliver a structured summary directly from the interface or browser extension.

Deploy Khoj on your VPS

Guide optimized for ServOrbit Cloud VPS.

01

Order a ServOrbit VPS

Plan for at least 4 GB of RAM and 2 vCPUs. Khoj loads embedding models at startup: an undersized server causes significant indexing delays.

02

Point your domain to the VPS

Khoj requires a FQDN for its authentication URLs and task callbacks. Add an A record at your registrar before deploying.

03

Deploy Khoj from the ServOrbit catalog

Choose the Khoj app in the AI category. ServOrbit generates a random admin password, installs the Docker Compose stack (PostgreSQL + pgvector + Khoj server) and configures nginx with your domain.

04

Log in and configure a LLM

Access the web interface with email [email protected] and the password shown in your client area. Go to Settings → LLM to connect Ollama (local) or your OpenAI key.

05

Connect the Obsidian plugin

Install the official Khoj plugin from the Obsidian library (Settings → Community plugins → Khoj), enter your instance URL and API key. Vault indexing starts immediately.

Frequently asked questions

No. Khoj runs on CPU only: it delegates LLM inference to your configured LLM (local Ollama, OpenAI API, etc.) and does not embed any generation model internally. The only local load is document embedding, which remains modest.

Complementary solutions

Artificial Intelligence

Khoj isn't orderable yet.