How to set up a knowledge base with Osaurus

Ask a standard AI model for your office Wi-Fi password and it won't know. Ask it how long your company takes to process a refund - same answer. Models can't see your documents unless you connect them.

Osaurus Knowledge fixes that. You point it at a local folder - markdown, PDFs, code, whatever you've got - and your agent can read it. Indexing happens on your Mac. Nothing is uploaded, nothing is copied to a server, and your original files are never altered or moved.

Here's the whole setup. It takes about two minutes.

Video Walkthrough

Check out the video below for a quick walkthrough.

Why bother connecting your own files?

Because most of the questions people actually want to ask an AI are about their own work.

Not "write me a poem about autumn". More like: what's our refund window, what did we agree in the client brief, which script handles the nightly export, what's the guest Wi-Fi password. Every one of those lives in a document somewhere on your machine, and a general-purpose model has no idea it exists.

The usual answer is to upload everything to a cloud service and hope for the best. That's fine for a recipe. It's less fine for HR policies, client contracts, unreleased product specs, or anything covered by an NDA.

Osaurus takes the other route: the files stay where they are, the indexing happens locally, and the model runs on your hardware. Your data does not leave your Mac.

What you need before you start

Not much:

• Osaurus installed on a Mac running macOS 15.5 or later, Apple Silicon
• A folder on your machine with the documents you want the agent to read

That's it. No account, no API key, no subscription. If you haven't installed it yet, grab it from the download page.

Getting your files ready

For this walkthrough we're using a folder called Company Policies containing two markdown files - one for refund guidelines, one for guest Wi-Fi access. Small, boring, exactly the kind of thing that's annoying to look up and perfect for a knowledge base.

Osaurus indexes files in place. Your originals aren't altered, moved, duplicated or sent anywhere. If you edit a document later, it's still your document in your folder - Osaurus is just reading it.

You don't need to reorganise anything first. A folder that already makes sense to you will make sense to the agent.

How do you add a knowledge collection in Osaurus?

Four steps.

1. Open Settings. From the top menu in Osaurus.
2. Go to Knowledge. It's in the left sidebar, under Agents and Automation.
3. Click Add Collection. Then fill in two fields:
   • Title - what this collection is. Ours is Company Policies.
   • Summary - a brief description of what lives in here.
4. Click Choose, select your folder, and hit Add. Osaurus indexes the files locally, straight away.

Why the summary field matters more than you'd think

It's tempting to skip it or type something vague. Don't.

The summary is how your agent decides when to consult this collection. It's not a label for your benefit - it's the instruction the model reads when it's working out whether the answer to a question might live here.

A vague summary means the agent either ignores a collection that had the answer, or trawls one that didn't. A specific one means it goes straight to the right place.

Compare:

• Weak: "Company stuff"
• Better: "Internal company policies — refund and returns guidelines, guest Wi-Fi access, expense limits and travel booking rules"

The second one tells the agent exactly which questions belong here. Name the topics. Use the words people would actually search for.

If you're setting up several collections, this is the field that decides whether the whole thing feels sharp or sluggish. Worth thirty seconds.

Testing that it worked

Open a fresh chat and ask something that can only be answered from your documents. In the walkthrough:

What is the guest Wi-Fi password and how many days do refunds take to process?

Two questions, two different source files, one prompt.

Osaurus searches the knowledge collection, pulls the password and the refund policy directly from the local markdown files, and answers in seconds. You can see it searching - it's not guessing, and it's not making anything up from training data. It's reading your files.

That's the test worth running on any knowledge base: ask something the model could not possibly know otherwise. If it answers correctly, your collection is wired up properly.

What to put in a knowledge base

Once it's working, the obvious question is what else to point it at. Things that work well:

• Team SOPs and internal policies - the questions people ask in Slack every week
• Project documentation and technical specs
• Client guides, onboarding packs, style guides
• A codebase, so the agent can answer questions about how something actually works
• Meeting notes and research, if you keep them as files

The pattern is: anything you currently find by searching, opening and scrolling. If you'd have to go and look it up, it belongs in a collection.

Separate collections work better than one enormous one. A collection per domain, each with a clear summary, gives the agent a much better shot at routing the question correctly.

The privacy bit, plainly

Everything above happens on your machine.

• Files are indexed in place - not copied, not moved, not modified
• Nothing is uploaded to a cloud service
• No account is required to use any of it
• The model runs locally on Apple Silicon

This is the part that makes a knowledge base genuinely useful at work rather than a compliance problem. You can point it at contracts, HR documents, unreleased plans and client material without anyone needing to sign off on a data-processing agreement, because there's no processing happening anywhere but your laptop.

Osaurus is free and MIT licensed. The source is on GitHub if you want to check any of the above rather than take our word for it.

Your agent, with a memory of your work

A knowledge base turns Osaurus from a model that knows general things into an assistant that knows your things. It reads your files, executes your tools, and remembers your work - all offline.

Set up one collection today. Start with the folder you open most often.

Download Osaurus — free, MIT, macOS 15.5+.