You already know Dottie is a Mac AI agent, not another chat window. This page is the first 10 minutes on an Apple Silicon Mac: download, first launch, one model path, one permission, one real task.

If you wanted a category ranking instead, start at best AI assistants for Mac. If you wanted voice depth, jump to the voice control guide after minute 10.

TL;DR

Minutes	Do this	Done when
0–2	Download the zip from the landing page	Dottie.app is in Applications (or wherever you dropped it)
2–4	Open it. Let first-launch on-device models finish	Chat window opens; you can type
4–6	Pick a model path: local, BYOK, or Dottie Pro	One path replies to a hello
6–8	Enable one tool category you will actually use	That permission is on; the rest stay off
8–10	Run one Mac task (calendar, a file find, or a folder tidy)	Something on the Mac changed that you did not click

Before you start

Requirements are strict. Dottie will not install on Intel.

• Apple Silicon (M1 or later)
• macOS 15.4 or newer
• Disk space for the first-launch on-device models (about 2 GB)
• Optional: an API key if you want BYOK (xAI, OpenAI, Anthropic, or Cerebras)

On Intel, use a chat client or a local runner instead. Dottie is not the product for that hardware.

The app is free. Optional Dottie Pro is $20/month for managed models. It does not unlock the 134 tools. Cumulative downloads are past 1,000 (api.dottie.ai, 2026-08-17).

Minutes 0–2: download

1. Open the Dottie landing page. The demo, requirements, and Pro link live there.
2. Download the zip. Direct asset: Dottie.zip.
3. Unzip and move Dottie.app to Applications.
4. Open it. If Gatekeeper asks, that is a normal first-open of a GitHub-signed desktop app.

There is no Mac App Store listing. There is no account required for local or BYOK.

Minutes 2–4: first launch

First launch downloads the default on-device models for speech and local chat (about 2 GB). Stay on a network until that finishes. After that, the local path can run offline.

You should see a chat surface and access to Settings. Type hello before you grant any tools. A reply means the model path is alive. A hang usually means the download is still running, or you picked a cloud path with no key.

Three local servers sit on localhost (nothing is exposed by default):

Server	Role
Gateway	Chat, tool calls, streaming
Audio	Speech recognition and TTS
LLM	On-device models when you choose local

Longer architecture: What Is Dottie Desktop.

Minutes 4–6: pick one model path

You do not need all three today.

Path	Use it when	Cost
Local (MLX / Ollama)	Privacy or offline	Hardware time only
BYOK	You already pay xAI, OpenAI, Anthropic, or Cerebras	Your existing API bill
Dottie Pro ($20/mo)	You want managed models and no key plumbing	$20/month, same email as in-app register

A small local model is enough to move files and read a calendar. Save frontier cloud models for hard reasoning. Local setup beyond the default download: run local AI models on Mac.

Prefix /think later if you want slower multi-step reasoning before tool calls. Skip it in minute five.

Minutes 6–8: grant one permission

Every tool category starts off. That is the product, not a missing setup step.

Pick the job you will run in minute 10 and enable only that category:

First task	Enable
”What’s on my calendar today?”	Calendar
”Find PDFs I downloaded this week”	Files
”Open Safari and go to GitHub”	App control (and Safari if asked)
“Draft an email to Alex saying I’ll be late”	Mail (sending still follows your confirm rules)

Leave Mail, Messages, screen, and Terminal off until you have a reason. Destructive steps still ask. Dottie is not a silent root shell.

Minutes 8–10: one real Mac task

Type or speak one of these. Voice is optional in the first 10 minutes.

Safe first tasks

• “What’s on my calendar today?”
• “Find PDFs I downloaded this week”
• “How much free disk space do I have?”

One-step automation

• “Create a folder on my Desktop called Inbox and move today’s screenshots into it”
• “Open Notes and make a note titled First Dottie task”

You are done when the Mac changed. A paragraph of advice with no file move, no event, and no app launch is a chat client. That is not this product. The split is spelled out in AI agent for Mac.

Voice, if you want it now

Three modes. All can stay on-device.

1. Push-to-talk. Hold the hotkey (check Settings for the current binding; landing copy documents Option+Space for dictation), speak, release.
2. Chat mic. Mic in the Dottie window. The utterance is an agent message with whatever tools you enabled.
3. Wake word. “Hey Dottie” from anywhere. On-device phrase detection; a floating overlay offers Send, Edit, or Dismiss.

Full walkthrough: voice control guide. You can finish the first 10 minutes on text only.

What not to do in minute one

• Do not enable every permission “so it works.” You will not remember what you granted.
• Do not start with “clean up my whole disk.” Start with one folder you can undo.
• Do not judge local models on a long essay. Judge them on a file move.
• Do not skip the landing page and grab a random GitHub tree. Releases live at dottie-desktop-releases.

After minute 10

Next job	Read
What to ask next	50 things to ask Dottie on your Mac
Hands-free voice	Control your Mac with voice
Siri vs a Mac agent	Siri alternative for Mac
What this replaces (and what it does not)	I Replaced Spotlight, Alfred, and Siri with One Mac App
Coming from Alfred workflows	Best Alfred workflows for Mac
Offline / Ollama / MLX	Run local AI models on Mac
Ollama vs LM Studio on Apple Silicon	Ollama vs LM Studio
Multi-model chat client vs Mac agent	BoltAI vs Dottie
More than one-shot commands	Automate your Mac with AI
How Dottie sits next to ChatGPT and Raycast	Dottie vs other Mac AI tools
Product definition	What Is Dottie Desktop

Frequently Asked Questions

What Mac do I need to get started with Dottie?

Apple Silicon (M1 or later) and macOS 15.4 or newer. Intel Macs are not supported. First launch downloads about 2 GB of default on-device models for speech and local chat.

How long does Dottie take to set up on a Mac?

About 10 minutes if the first-launch model download is already cached or on a fast network: install, pick a model path, enable one permission, run one task. The model download is the only step that can stretch past that.

Do I need an account or Dottie Pro to start?

No. Local and BYOK work with no Dottie account. Optional Dottie Pro at $20 per month is managed models only. It does not unlock the 134 tools.

Which model should I pick in the first 10 minutes?

Local if you want privacy and a zero-config hello. BYOK if you already pay OpenAI, Anthropic, xAI, or Cerebras and want a stronger model on day one. Pro if you want managed models and no key plumbing.

Why are all Dottie permissions off after install?

Deny-by-default is the security model. Enable the one category you need for the first task. Mail, Messages, screen, and Terminal can wait.

Can I start with voice instead of typing?

Yes. Push-to-talk, the in-window mic, and the Hey Dottie wake word are available immediately. Text is enough to prove the agent works. See the voice control guide for the three modes.

Where do I download Dottie for Mac?

From the dottie.ai landing page, or the latest Dottie.zip on the dottie-desktop-releases GitHub repo. There is no Mac App Store listing.

Related Posts

• 50 Things to Ask Dottie on Your Mac: catalog after minute 10
• What Is Dottie Desktop?: product definition
• Voice Control Guide: wake word, push-to-talk, TTS
• Run Local AI Models on Mac: Ollama / MLX after the default download
• Ollama vs LM Studio: which local runner on Apple Silicon
• BoltAI vs Dottie: multi-model chat client vs Mac agent
• Automate Your Mac with AI: command list after you are installed
• Dottie vs ChatGPT, Raycast AI & Apple Intelligence: multi-tool map
• Best AI Assistants for Mac: category ranking
• Introducing Dottie Desktop: launch post