Raycast AI Review: The Launcher That Quietly Became My Favorite AI App
It lives one keystroke away, routes to whatever model fits the task, and finally makes "AI everywhere on your Mac" feel less like a slogan. The catch is the price stack and the Mac-only lock-in.
Raycast AI is the rare AI feature that earns its keep by getting out of the way. It's a fast keyboard launcher with frontier models wired into the same command bar you already hit to open apps and resize windows, and that shortcut-everywhere context is more useful day-to-day than any standalone chat tab. Pro at $8/month covers the everyday models; the $8 Advanced AI add-on opens up GPT-5, Claude, Gemini and friends if you actually need them. It misses Editors' Choice by a hair on Mac-only lock-in and a weak file search, but if you live on macOS and want one keystroke between you and an AI that already knows your selection, this is the one to beat.
Raycast started life as a Spotlight replacement, and for a couple of years that's all I used it for. I kept ignoring the AI layer because I already pay for ChatGPT and Claude, and I didn't see the point of a third chat box. I was wrong. After a month of forcing myself to use Raycast AI instead of tabbing over to a browser, I'm not going back.
The pitch is small but powerful. Hit your hotkey, type or invoke an AI Command, and you get an answer where you already are: over a selection in any app, against your clipboard, or as a quick chat that disappears the moment you're done. No browser tab, no model picker if you don't want one, no copying answers back and forth. I put it through a normal week: drafting emails, rewriting half-baked Slack messages, summarizing long PDFs, translating product copy, fact-checking a research doc, and a couple of times asking it to explain a chunk of SQL I inherited. It handled almost all of it without me leaving the app I was in.
This review covers the Pro tier with the Advanced AI add-on (so $16/month all-in), tested on an M-series MacBook Pro running macOS Tahoe. I also poked at the Windows beta long enough to confirm it's still a beta.
Pros
- The "AI lives in your command bar" model is genuinely better than tabbing to a chat app. Select text anywhere, hit the hotkey, run an AI Command, paste the result. It's a 3-second loop instead of a 30-second one
- Auto Model picks a sensible model for each task so you stop fiddling with pickers. When I did want to steer, swapping between GPT-5, Claude, and Gemini mid-conversation is one keystroke
- AI Commands are the killer feature. "Fix grammar," "summarize," "explain this code," plus any custom prompt you write, all bindable to a hotkey or a typed alias. It turns prompts you'd otherwise rewrite ten times a week into muscle memory
- Bring-your-own-key support lets you plug in your existing OpenAI, Anthropic, Google, or OpenRouter API keys and skip the AI tier entirely if you already pay for those. Rare flexibility in this category
- Ollama integration runs 100+ open-source models locally with no cloud round-trip, which is the right answer for anything sensitive
- The launcher itself is still the fastest one on Mac. The window opens effectively instantly on Apple Silicon, and clipboard history, window management, and snippets are all included in the base app
Cons
- The pricing stack is confusing. Pro is $8/month, but the frontier models most people actually want (GPT-5, Claude Sonnet, Gemini 2.5 Pro, o3) are locked behind a separate $8 Advanced AI add-on, making the real "I want the good models" price $16/month
- Mac-first to the point of Mac-only in practice. There's an iOS companion and a Windows public beta, but the full experience is macOS, full stop. If you switch machines between OSes during the day, this isn't your tool
- File search is still the weakest pillar compared to Alfred. If finding a specific file fast is your #1 launcher need, you'll notice
- Some of the smartest features (Auto Model, custom AI providers, the Advanced AI catalog) are buried in settings menus a new user won't find on day one
What it actually does that nothing else does
The thing Raycast AI gets right, and what every standalone chat app gets wrong, is where the AI lives. Every other tool I pay for ($20 ChatGPT, $20 Claude, $20-ish for whatever browser-AI of the month) is a destination. You stop what you’re doing, switch apps or tabs, paste context, get a result, paste it back. Raycast AI is the opposite. It’s already there. Hit your hotkey, the bar drops down, you ask, you get an answer, the bar disappears. Total time: a few seconds. Total context switches: zero.
The AI Commands feature is where that pays off hardest. You can define a prompt once (“rewrite this Slack message to be 30% shorter and less formal,” say, or “explain this regex line by line”) and then summon it on any selected text in any app via a keyboard shortcut. After a week of building a handful of these, I realized I’d stopped opening ChatGPT for the small stuff entirely. The small stuff is most of the stuff.
Auto Model, added in v1.102, is the other quietly excellent feature. Instead of staring at a model picker every time you start a chat, Raycast routes your prompt to a sensible model based on the task. Quick rewrite? A fast cheap model. Hard reasoning question? It reaches for o3 or Claude Sonnet. You can override per-chat with one keystroke if you disagree. This is what every AI app should have done eighteen months ago.
The model lineup is genuinely good
With Pro + Advanced AI, the lineup of frontier models you can route to is the actual selling point: GPT-5, Claude 3.7 Sonnet, o3 and o3-mini, Gemini 2.5 Pro, plus whatever Raycast adds through the year. Unlimited use of all of them, no per-message metering, no “you’ve used 80% of your fast requests” warnings. For $16/month total, that’s a lot of frontier AI for less than the price of a single direct subscription to any one of those providers.
If you’d rather not pay Raycast for AI at all, bring-your-own-key works. Plug in your existing OpenAI, Anthropic, Google, or OpenRouter API keys and Raycast routes through those instead. You skip the Pro AI tier entirely and just pay for the underlying API usage. This is rare flexibility in the category (most launcher AIs lock you into their own billing) and it’s exactly the option a power user wants.
For the privacy-sensitive, the Ollama integration since v1.99 runs more than 100 open-source models locally on your machine. No cloud, no API keys, no telemetry. It’s slower and the models aren’t frontier-class, but for code or documents you can’t send out, it’s the right tool.
Where it lets you down
The pricing stack is the first place new users get confused, and pretending otherwise wouldn’t be honest. Pro is $8/month on the annual plan ($10 month-to-month), and that includes unlimited chats, but only with the standard models (GPT-4o-mini, Claude Haiku, Llama). The frontier models, which is what almost everyone reading this actually wants, are gated behind a separate Advanced AI add-on at another $8/month on top of Pro. So the real “I want the good models” price is $16/month, and Raycast’s pricing page doesn’t communicate that as cleanly as it should. It’s still a fair deal, cheaper than any standalone Pro plan, but the layered structure means a lot of users get to checkout and feel like they were upsold.
The platform story is the second wall. Raycast is a Mac-first product to the point of being effectively Mac-only. There’s an iOS companion app with the core chat, notes, snippets, and quick links, and a Windows public beta exists, but if you split your day between a Mac and a Windows laptop or a Linux box, you’re not getting the full experience on the other side. That’s a deal-breaker for some people. Know it going in.
File search is the third weak spot. Compared to Alfred, Raycast’s file search is the pillar that hasn’t fully caught up, and if your top reason for using a launcher is “find that one PDF I downloaded last Tuesday,” you’ll feel it. For everyone else, it doesn’t really matter; you’re using the launcher for apps, snippets, and AI, not as a finder replacement.
Finally, some of the smartest features are buried. Auto Model, custom AI providers, the Advanced AI catalog, Ollama setup. These all live in nested settings panes a brand-new user won’t stumble across in their first hour. The product would be better if the first-run experience walked you through them.
Who it’s for
If you’re on a Mac all day, hit ⌘+Space dozens of times an hour, and have started to feel like “open a browser tab, find the ChatGPT pin, wait for the model picker, paste, copy back” is the slowest part of your workflow, this is the tool. The $8 Pro tier is worth it for the launcher alone. The $8 Advanced AI add-on is worth it the moment you realize you’ve stopped opening other AI apps for quick tasks.
If you’re a heavy chat user who lives inside long, exploratory ChatGPT or Claude conversations, you won’t replace those with Raycast AI. The chat UX inside a command bar is great for short turns and a couple of follow-ups; it’s not where you want to write a 20-message thread debugging an idea. Keep the standalone app for that and use Raycast AI for everything else.
And if you’re cross-platform, wait. The Windows beta exists and is improving, but a Mac-first product is going to be a year or two behind on Windows for the foreseeable future. Don’t pay for what you can’t fully use.
The bottom line
Raycast AI doesn’t beat ChatGPT or Claude at being a chat app. It beats them at being available. One keystroke, model already picked, context already grabbed from your selection, answer already where you needed it. That ambient availability is the actual product, and once you’ve felt it, opening a browser tab for AI feels like fax-machine behavior. The pricing layers are clumsy, the Mac-only reality is real, and file search needs work, so it lands at 89 and just misses our Editors’ Choice. But if you’re on a Mac and you’ve been paying for a chat app you keep tabbing back to, this is the upgrade. It’s the better daily driver for AI on macOS, and it earns its keep the first week.
Sources
FAQ
What did Raycast AI score?
An 89 out of 100. That's just under our 90 threshold for Editors' Choice. It loses those last points on the two-layer pricing (Pro plus the Advanced AI add-on to reach the frontier models), the Mac-only reality, and a file search that still trails Alfred.
Is Raycast Pro worth $8 a month?
If you're on a Mac and live at the keyboard, yes, and it's worth it before the AI even enters the picture. Cloud sync, unlimited clipboard history, and the polish on top of the free launcher pay for it. Pro also includes unlimited chats with standard models (GPT-4o-mini, Claude Haiku, Llama). Add the $8 Advanced AI add-on only if you actually want to route everything through frontier models.
Do I still need ChatGPT Plus or Claude Pro if I have Raycast AI?
Probably not, if your usage is mostly "quick AI tasks throughout the day." Raycast Pro plus Advanced AI gives you unlimited use of GPT-5, Claude, Gemini, and o3 inside the launcher for $16/month, less than either standalone subscription. You lose the deep chat-history UX and features like ChatGPT's voice mode, so if you do long, exploratory conversations you may still want the standalone app.
Can I use my own API keys?
Yes. Raycast lets you plug in your own keys from OpenAI, Anthropic, Google, or OpenRouter and skip the Pro AI tier entirely. If you already pay for API access, this is the cheapest way in.
Does Raycast AI work offline?
Sort of. Since v1.99 Raycast integrates with Ollama, which lets you run more than 100 open-source models locally with no cloud round-trip. It's the right pick for sensitive code or documents. Cloud models obviously still need a connection.