REVIEW · OPEN-SOURCE AI SECOND BRAIN
Verdict: Khoj is a thoughtful AI second brain: its search finds notes and documents by meaning, not just keywords, and it works on the web, in Obsidian, in Emacs and on the desktop. Whether it is safe depends on where you run it: the desktop app syncs your files to the Khoj server you connect to, so self-host it if your documents must stay with you.
10 November 2026 · By Ishtiaque Hossain · 4 minute read
Looking for a file you can describe but not name? Try Shlexball. The first 30 searches are free and need no account.
macOS 15 or later. Apple silicon. 16 GB memory. On a phone? Email yourself the link.
This review is based on Khoj's own pages, read on 14 October 2026. Every claim links to the page it came from. We are the makers of Shlexball, so we say plainly where Khoj does better.
| What it is | An AI second brain: search, chat, agents and automations across your documents and the web |
| Where it runs | Web app, Obsidian, Emacs, a desktop companion app, or a server you host |
| Search | Natural-language search by meaning, with query filters |
| Where your files go | Synced to the Khoj server the app connects to; self-hosting keeps them on your server |
| Made by | Khoj, a small applied AI company that builds in the open |
| Best for | Note-takers who want to search and chat with a personal knowledge base |
Khoj calls itself your AI second brain: build agents, schedule automations and research across your documents and the web, using any AI model as your personal assistant.
It comes from a small applied AI company that says it builds in the open and aims for transparent tools you can understand. The same team also makes Open Paper, a research workbench, and Pipali, an AI co-worker in beta that runs on your computer.
You can use Khoj in a web browser, in Obsidian through its search icon or command palette, in Emacs, or through a desktop companion app.
Khoj search finds relevant notes and documents from your knowledge base in natural language, and you can add query filters to limit which entries it searches.
Under the hood, a bi-encoder model turns each chunk of your documents into a meaning vector, a list of numbers that captures what the text is about. Your query becomes a meaning vector too, and Khoj compares them to find the closest chunks.
A slower but more accurate cross-encoder model then re-ranks those results for your query. That is why Khoj can match a question to a note that uses different words.
Khoj works on the files you give it. In the desktop companion app you choose files and folders in Settings, click Save, and Khoj keeps them in sync so it can answer when you search or chat.
You can also install the Khoj web app as a desktop app from Chrome, which Khoj calls a progressive web app.
If you change the search model later, you have to re-index all your documents.
The desktop app uploads your knowledge base to the Khoj server it is connected to and keeps it in sync. That server is Khoj's service unless you point the app elsewhere.
Self-hosting changes that. You set the Khoj URL in the desktop app to your own server, by default an address on your own machine, and generate an API key from your web app, or run the server in anonymous mode without one.
When self-hosting, Khoj sets up a local search model by default. You can instead use models from Hugging Face, the OpenAI API, Azure OpenAI or OpenAI-compatible servers such as Ollama, so where the processing happens is your choice.
The pages we read do not list prices.
Khoj suits people who already keep notes, especially in Obsidian or Emacs, and want to search and chat with them by meaning.
If privacy matters, plan to self-host from the start; Khoj's documentation walks through the server settings.
Shlexball also searches by what you remember, but it does it on your Mac across your folders: describe the file in plain English and it searches names and the text inside files, including PDFs and Word documents.
Nothing to sync and no server to run: its language model runs on your Mac and files are never uploaded. The first 30 searches are free.
One app, one job, nothing to set up: Shlexball, 30 free searches.
▶ DOWNLOAD SHLEXBALLThe desktop app syncs your files to the Khoj server it connects to. To keep documents on your own machine, self-host Khoj and point the app at your server.
A bi-encoder model turns document chunks and your query into meaning vectors and finds the closest matches, and a cross-encoder model re-ranks them.
Yes. A desktop companion app keeps chosen files and folders in sync with Khoj, and the web app can also be installed from Chrome.
When self-hosting, Khoj sets up a local search model by default, and supports OpenAI-compatible servers such as Ollama.