Khoj indexes your own material and lets you ask questions of it. The pitch is a personal assistant that knows what you have written rather than what the internet has written.

What it connects to

Local files including Markdown, PDF, and plain text, plus integrations with Obsidian, Notion, and GitHub. There are Obsidian and Emacs plugins so the search happens where you already work.

That last point matters more than the feature list. A knowledge tool you have to switch applications to use is a tool you stop using.

Semantic search over your own material

Conventional search finds the words you typed. Semantic search finds the meaning, so a question about “that thing I wrote about container networking” surfaces the note even if it never used those words.

For a large notes collection accumulated over years, this is the difference between an archive and something you can actually retrieve from.

Chat with context

Beyond search, Khoj can answer questions using your documents as context, in the same way a RAG application does, and it can search the web through SearXNG when the answer is not in your material.

Local or hosted models

It works with local models through Ollama or with hosted APIs. For a tool reading your personal notes and email, local inference is the configuration that makes the premise coherent.

Practical notes

Indexing quality determines everything. Well-structured Markdown indexes cleanly; a folder of scanned PDFs without OCR does not.

Initial indexing of a large collection takes a while and generates a lot of embedding work, which is faster with a GPU.

Where it fits

Khoj for searching material you already have, spread across several places.

AnythingLLM for asking questions of a defined document set you upload deliberately.

Onyx for the same thing across a team’s business systems.

License

Khoj is released under the AGPL-3.0 License.

Links

• Website
• Repository