The One Mac App Every AI Power User Installs First

People ask me some version of this question constantly: “New Mac, I care about AI — what do I install first?” And after setting up three Macs of my own and walking maybe a dozen friends through theirs, I have a confident answer: Raycast. Not an AI app in the obvious sense — not a chatbot, not a model runner — and that’s precisely why it wins. But “trust me” is not an argument, so let me show my reasoning: what the first-install slot actually demands, why Raycast fills it, and honestly, which people should pick something else.

What makes a “first app” — three criteria

The first app you install shapes how every later tool gets used, so the bar is specific:

1. Universal access point. AI’s value on a desktop is proportional to how little friction sits between any moment of work and the model. An app you must switch to loses to an app summoned over whatever you’re doing. The first install should put AI one keystroke from everywhere.
2. Local + cloud model support. Power users route by task: local models for private or trivial things, frontier cloud models for hard things. A first app that locks you to one provider — or one privacy posture — is a future regret.
3. Extensibility. Your AI workflows in six months don’t exist yet. The first app should be a platform that grows new capabilities without you waiting for a vendor roadmap.

Score the obvious candidates against this and the field thins fast. ChatGPT’s Mac app: strong model, but a destination, not a layer — fails 1 and 2. LM Studio: superb local runner, but it’s where models live, not where work happens — fails 1. Shortcuts: genuinely universal and free, but building each workflow is a project — weak on day-one value. Raycast is the one tool that scores on all three. Here’s what that looks like in practice.

The Raycast deep-dive

Raycast replaces Spotlight (⌘-Space) and turns the launcher into a programmable command surface. The AI-relevant pillars:

AI Commands over any selection. Select text in any app, hit a hotkey, and run a saved prompt against it — fix grammar, translate, summarize, “make this email 50% shorter,” “explain this error.” You write the prompts once; they become system-wide verbs. My Czech↔English translation command gets used 10+ times a day across Mail, Slack, and the browser, and the fact that it works in place — no copy, switch, paste, copy, switch, paste — is the entire difference between an AI you use constantly and one you use when you remember to. Quick AI is the complement: tap ⌘-Space, type a question, get an answer inline without any app switch.

Real model routing, including Ollama. Raycast AI (Pro plan, ~$8/month) provides hosted access to GPT-class, Claude-class, and Gemini-class models, selectable per command. And — this is the power-user unlock — community extensions connect Raycast to a local Ollama server, so the same interface drives localhost:11434. My setup: translation and grammar commands run on local Qwen (private, free, fast enough), while “review this architecture sketch” goes to a frontier cloud model. One keystroke, two privacy tiers. You can also skip the Pro plan entirely and run Raycast free with only Ollama behind it — slower top-end intelligence, zero subscription, nothing leaves the Mac.

Clipboard history. Sounds mundane; isn’t. Searchable history of everything you’ve copied is the connective tissue of AI work — you’re perpetually shuttling prompts, outputs, code, and quotes between apps, and ⌘-Shift-V into a searchable archive means nothing is ever lost to an overwrite. Of everything in this post, this is the feature people later tell me they can’t live without.

The extensions store. Thousands of community extensions: GitHub, Linear, Notion, Spotify, window management (built-in, actually — halves and thirds via hotkeys, replacing a paid app), and a steady stream of AI integrations. Criterion 3, satisfied — when some new model server or AI service matters next year, odds are someone’s already shipped the extension.

The runners-up — and who should genuinely pick them instead

A clear winner doesn’t mean one-size-fits-all. Three apps regularly deserve the first slot for specific people:

LM Studio — for model tinkerers. If your actual hobby is the models — comparing quantizations, watching tokens-per-second, testing every new release the week it drops — LM Studio is your home screen, not Raycast. Visual model browser, MLX-optimized inference on Apple Silicon, per-model parameter tuning, and a local API server that other tools (including Raycast) can sit on top of. Tinkerers install LM Studio first and Raycast second; everyone else, the reverse.

MacWhisper — for the meeting-heavy. If your week is 15+ hours of calls, the highest-value AI on your Mac isn’t a launcher — it’s local transcription. MacWhisper runs Whisper models on-device: drop in a recording, get an accurate transcript plus AI summary, nothing uploaded anywhere. For consultants, researchers doing interviews, and anyone whose job is mostly conversations, this delivers more value per week than everything else in this post combined.

Shortcuts — for the no-new-apps minimalist. Already on your Mac, free, and as the previous posts in this series show, capable of building briefings, note pipelines, and AI automations against a local model. The honest tradeoff: you’re trading Raycast’s day-one polish for hours of assembly. Pick this if your constraint is “no new software” (locked-down work machines, principled minimalism) rather than “no effort.”

If you recognized yourself in none of those three profiles, the answer is Raycast.

The 30-minute first-day setup

Installing the app is not the win; configuring it is. brew install --cask raycast, then these five configurations, in value order — 30 minutes total:

1. Take over ⌘-Space (3 min). System Settings → Keyboard Shortcuts → Spotlight → disable; set Raycast to ⌘-Space in its settings. This step is psychological as much as functional: the new muscle memory only forms if the old keystroke leads to the new tool.
2. Enable clipboard history with a hotkey (2 min). Raycast Settings → Extensions → Clipboard History → assign ⌘-Shift-V. Done. It now silently archives everything.
3. Create your first three AI Commands (10 min). Don’t import a pack of fifty — write the three you’ll actually use: Fix Grammar (output corrected text only), Summarize (5 bullets), and one personal to your work (mine: translate to/from Czech; yours might be “explain this code” or “make this client-friendly”). Assign the AI Command hotkey so they’re reachable over any selected text.
4. Connect your models (10 min). Either start the Pro trial and set per-command models, or brew install ollama, ollama pull qwen2.5:14b (8B on a 16 GB machine), add the Ollama extension, and point your three commands at it. Decide your privacy line now: which commands are allowed to leave the machine.
5. Set window-management hotkeys (5 min). Left half, right half, maximize — ⌃⌥←, ⌃⌥→, ⌃⌥↩. Not AI, but you’ll use them 50 times a day, and they’re why Raycast displaces a paid app on day one.

End of day one, you have: AI over every text field on the system, two-tier model routing, a searchable clipboard, and window management — from one install. That’s the case in full. The first app for an AI power user isn’t the smartest model or the best chat window; it’s the layer that makes every model one keystroke away from the work you’re already doing. Raycast is that layer. Install it first, give it the 30 minutes, and let everything else you add this year plug into it.