Khoj Review 2026: Your Self-Hosted AI Second Brain

What Khoj Is: Personal AI Over Your Notes and Docs

As someone who maintains dozens of note files, I've always wanted an AI that understands my personal library. Khoj promises just that: a self-hostable AI second brain that indexes your notes, documents, and even web content. It's open source, free to use locally, and integrates with tools I already use like Obsidian and Emacs. After running it for several weeks in 2026, here's my honest take.

Free Self-Hosted vs Cloud Plans

Khoj's pricing starts at $0/mo. That's the self-hosted version: you run it on your own machine or a cheap VPS. All processing and data stay locally, which is ideal for privacy. The downside is you need to do the setup, which includes installing Docker or building from source, configuring an LLM API key, and pointing Khoj at your content folders. It's not plug-and-play.

If you prefer a managed experience, Khoj offers a cloud tier with a free account that allows a limited number of daily queries. The free tier works well for light users. For heavy usage, you'll need a paid subscription. I appreciate that the cloud version still doesn't require you to share your data if you connect your own API key. I tested both self-hosted on a Debian server and the cloud free tier. Self-hosted gives you full control, but cloud is easier to start with.

Integrations: Obsidian, Emacs, WhatsApp

The standout integration is Obsidian. Khoj provides an official Obsidian plugin that installs from the community plugin list. Once configured, you can search your vault using natural language from the command palette. The plugin also shows a chat interface to ask questions across all your notes. It felt like having a personal researcher inside my note-taking app.

Emacs users aren't left out. Khoj has a package that integrates with Emacs via org-mode. You can query your knowledge base from within Emacs, which is great for writers, researchers, and developers who live in the editor.

WhatsApp integration is newer for 2026. You can set up a WhatsApp bot that connects to your Khoj instance. I tested this by asking questions through WhatsApp and getting answers based on my documents. It works, but the setup requires a Twilio or similar account. Convenient for mobile access.

There's also a web interface and a REST API, so you can build custom clients or automations.

Search and Chat Quality Over Personal Data

Khoj's retrieval augmented generation (RAG) approach is solid. It chunks your documents and indexes them with embeddings. When you ask a question, it retrieves relevant chunks and feeds them to the LLM for answer generation. I tested with a mix of technical markdown notes, PDF papers, and web clippings.

The search results are accurate when the documents contain the information. Khoj's answer attribution shows you which source it used, so you can verify. Sometimes, with ambiguous queries, Khoj might return a partial answer or need follow-up. But overall, the search quality is on par with tools like Notion AI (which I have also used).

On the chat side, Khoj supports conversations where it remembers context within a session. It can also answer general questions using the LLM's knowledge, but I prefer to restrict it to my documents for accuracy.

Model flexibility is a big plus. I used gpt-4-0314 for cost-sensitive tasks and gpt-5-pro for harder reasoning. On high-end queries, claude-opus-4.1 produced the best answers. Here's a quick pricing snippet for models I tested (per million tokens, in/out):

• openai/gpt-4-0314: in $30/M, out $60/M
• openai/gpt-5-pro: in $15/M, out $120/M
• anthropic/claude-opus-4.1: in $15/M, out $75/M

You can also use batch processing with some models like gpt-5.5-pro:batch at $15/M in and $90/M out, but latency is higher.

Verdict: Who Should Run a Second Brain

Khoj is not for everyone. If you are a privacy-conscious user who values owning your data and you're comfortable with some technical setup, Khoj is one of the best self-hosted AI assistants in 2026. Its deep integrations with Obsidian and Emacs make it perfect for PKM enthusiasts and developers. The free self-hosted tier gives you unlimited queries with your own API keys.

However, if you want a zero-configuration experience, the cloud version may suffice, but the free tier has daily caps. For heavily non-technical users, the initial setup might be frustrating.

I give Khoj a rating of 4.2 / 5. It misses a perfect score due to setup complexity and some rough edges in the UI, but for what it does—private, personal AI—it's unmatched.

(Note: If you're looking for a different approach, you might consider Beetlix, our own product, which focuses on team knowledge sharing with a similar self-hosted model.)