Skip to main content

Quick Start

Five minutes from a fresh Mac to a working AI you own. Here's the whole flow.

1. Install​

Grab the DMG from osaurus.ai, drag it into your Applications folder, and launch it from Spotlight (⌘ Space → "Osaurus"). That's it.

If macOS shows a security dialog on first open, right-click Osaurus.app and choose Open — you only do this once. Full guide: Installation.

2. Walk through onboarding​

A short wizard greets you on first launch and walks through five quick steps.

Welcome​

The hero screen. Click Get Started.

Create your agent​

Pick a name and a starter template (you can change everything later). An agent is a saved configuration — a system prompt, theme, default model, memory of its own. You can always come back and create more. Agents →

If you want to skip and use the default Osaurus agent, click Skip for now.

Configure your AI​

Pick how you want to run models:

OptionBest forSetup
Apple Foundation (macOS 26+)Zero setup, fast on Apple Neural EngineBuilt in
Local model (MLX)Privacy-first, runs offlineDownload a model from the picker — Gemma 4 E2B it 4bit is the recommended starter (~1.5 GB)
Cloud providerFrontier-class modelsPaste an API key for OpenAI, Anthropic, xAI, or OpenRouter

You can always add more later from Management → Models or Management → Providers.

Set up identity​

Osaurus creates a key for you (and your future agents) the first time it runs. It lives in your iCloud Keychain, gated by Face ID / Touch ID. A one-time recovery code is shown — write this down somewhere safe, it's the only thing that can recover your identity if you lose access to your Mac.

This step is skippable but recommended — it unlocks access keys and Public Links later. Identity →

How it works​

A short carousel covers what you can do once you're in: tasks, the Sandbox, personalization, and privacy. After you finish, the chat overlay opens and you're ready to go.

3. Pick your Core Model​

This is the one knob most people miss on day one. Open Settings (⌘ ,) → General → Core Model and pick a model.

Why it matters​

The Core Model is a small, lightweight model Osaurus uses for background work:

  • Remembering things — distills your conversations into compact facts the AI can recall later. Without a Core Model set, this never runs and memory pauses.
  • Picking the right tools — surfaces the relevant tools and skills for each message. Falls back to your chat model when Core Model is unset.

What to pick​

If foundation is available, that's the right answer 99% of the time — it's free, fast, and never leaves your Mac. Otherwise:

You havePick
macOS 26+foundation (Apple's on-device model)
macOS 15.5+ with a local modelThe smallest fast model you've downloaded (e.g. gemma-4-e2b-it-4bit)
Cloud-only setupAny cheap, fast remote model (e.g. anthropic/claude-haiku-4-5)
tip

Choosing Use chat model (default) in this picker leaves Core Model unset. That's fine for ad-hoc usage, but memory won't update. Pick an explicit model if you want memory and auto-selected tools working in the background.

4. Try your first chat​

Press ⌘; from anywhere on your Mac. The chat overlay appears.

Type something:

Hi! Tell me a fun fact about dinosaurs.

Press Enter. You'll see the response stream in real-time. Press ⌘; again to dismiss.

Have it actually do something​

Try giving the agent a real task — it'll write a plan, use tools, and bring back the result:

  1. Press ⌘; to open chat
  2. Click the folder icon next to the input bar and pick a folder you don't mind it touching
  3. Ask: "Summarize what's in this folder and add a README.md describing it."

You'll see a live to-do list appear and tick off as the agent reads files, drafts the README, and writes it. The new file shows up as an artifact card right in the chat.

On macOS 26+, toggle the Sandbox for shell access in an isolated Linux VM — handy for running scripts, scraping URLs, or trying out a package. Tasks →

Try voice​

Click the microphone in the input bar and speak. Transcription happens entirely on-device via Apple's Neural Engine — your voice never leaves your Mac. Or set up the global Transcription Mode hotkey to dictate into any app on your Mac. Voice →

What's next​

Now that you're set up:

For everyday use:

  • Chat — overlay, multi-window, sessions, shortcuts
  • Agents — create specialized assistants for different tasks
  • Memory — what your AI remembers and how
  • Skills — reusable expertise, automatically loaded
  • Voice — dictate, wake-words, global transcription
  • Themes — make the chat overlay yours

For developers:

Care about privacy? The whole story is on the Privacy & Trust page.

Need help? Join the Discord or open a GitHub issue.