Raycast Pro Review: The Keyboard-First AI Layer That Made My Command Bar Smarter Than My Browser

What it’s actually good at

Quick AI is the feature that quietly changed how I work. Hit your global hotkey, type a question, press Tab, and get a citation-backed answer inside the same little bar you use to launch apps. That’s it. No browser tab, no waiting on a chat UI to render, no context switch. For the class of question that makes up most of my AI use, quick lookups, definitions, translations, “what’s the flag for silent install of X,” “regex for a US phone number,” “rewrite this Slack message so it doesn’t sound passive-aggressive,” it’s genuinely the fastest AI tool I own. And AI Extensions like @file-explorer, @selected-text, @terminal, @raycast-focus, and third-party ones like @slack, @notion, and @linear now plug into Quick AI and AI Chat , so the same command bar can pull context from what you’re actually working on.

AI Chat is where the money goes when one answer isn’t enough. Ask a question, follow up, and keep the thread going, and with AI Extensions and MCP servers, AI Chat can pull in live context and take actions across your apps and tools, like creating a Linear issue or searching your codebase, without leaving the conversation. That last part is what separates Raycast’s AI from “a chat window on a hotkey.” When I @-mention Notion and ask it to summarize a page, it summarizes the actual page. When I @-mention Linear and ask it to file a bug, it files the bug. That’s the small, boring version of “agents” that actually earns its keep: task completion inside your existing tools, not a spinning demo.

The model story is the part I underestimated. All Raycast AI features are powered by different AI providers: OpenAI, Anthropic, Perplexity, Groq, Baseten, Mistral AI, Google (Gemini), xAI and Replicate. Pro gets you unlimited use of the standard lineup, and the Auto Model feature added in v1.102 picks the best model for each task automatically rather than forcing you to choose , which is the right default. If you want to reach for frontier reasoning, reasoning models including GPT-5 Reasoning, Claude Sonnet 4 Reasoning and Grok-4.20 Reasoning are there via the Advanced AI add-on. And if you want to skip Raycast’s model pricing entirely, Bring Your Own Key is supported for OpenAI, Anthropic, and Google . Plug in the API key you already have and route around the AI tier. Very few tools in this category let you do that without gymnastics.

The privacy story is also unusually good for a tool that lives this close to your keystrokes. Raycast says it never collects sensitive information, none of your inputs are recorded or used to train models, features are powered by external providers, and if Cloud Sync is enabled, AI Chats are saved encrypted at rest and in transit so they can be accessed from any of your Macs. If you want to go further, since v1.99 Raycast integrates with Ollama to run more than 100 open-source models locally, no cloud, no API keys, complete privacy. That’s the setup I’d use in any regulated shop.

And I keep forgetting to mention the non-AI stuff because it’s just… there. Unlimited clipboard history that’s searchable across text, images, and links. Snippets with dynamic placeholders. Quicklinks with arguments. Window management that’s better than Rectangle. Notes that sync. A calculator that speaks plain English. The extensions store with over 2,000 options covering GitHub, Notion, Linear, Jira, Slack, Docker and more . This was already a great $8 before the AI got good.

Where it lets you down

File search is the one place Alfred still has an edge. File search remains the weakest pillar compared with Alfred , and even after Raycast 2.0’s rewrite, a custom Rust-based file indexer to make this work on both platforms without relying on Spotlight on Mac or the Windows Search index, both of which have limitations for this use case , heavy filesystem users will still notice. If your workflow is 90% “find me this file buried six folders deep and act on it,” Alfred is still the sharper knife. For everyone else, the v2 unified search is a real step up.

The Windows story is the other rough edge, and it’s worth being honest about. Raycast launched a public beta for Windows in late 2025, and by mid-2026 the Windows version is still in beta, supporting Windows 10 version 1903 and later, as well as Windows 11. It works. It’s not a joke port. But AI Chat, Notes, and more features are coming to Windows in the coming months , read: they aren’t there yet. Some Mac-native integrations like Apple Shortcuts don’t have Windows equivalents, and full feature parity with the Mac version is targeted for Q4 . If you’re a Windows-only shop planning to standardize on Raycast today, wait a quarter. If you’re a Mac-first user who occasionally uses a Windows laptop, install the beta and forgive the gaps.

The pricing add-on is the smallest but most annoying gripe. Pro at $8/month is a straightforward yes. The Advanced AI add-on that doubles the bill to $16/month for frontier reasoning models is a much narrower “yes,” and the way it’s marketed makes it sound like the “real” AI plan. It isn’t. Most people should skip it unless they know specifically why they want GPT-5 Reasoning or Claude Sonnet 4 Reasoning inside their launcher.

Should you pay for it?

If you’re on a Mac and you use AI more than once a day, yes. Buy Pro. Raycast Pro pricing in 2026 is $8/user per month billed annually ($96/year) or $10/user per month billed monthly, the annual plan saves you $24 per year, roughly 20% versus paying month-to-month. One seat covers every Mac on your account. There are no metered credits, no surprise overages, no rolling usage pool to babysit. The AI alone replaces a ChatGPT Plus subscription for the everyday quick-lookup workflow at less than half the price, and you still get the ChatGPT app if you want it. This just becomes the layer you actually reach for first.

If you already pay for OpenAI or Anthropic API access, the play is even better: stay on the free tier for AI, use Bring Your Own Key, and pay $8 for Pro anyway to get unlimited clipboard history, cloud sync, custom themes, unlimited Notes, and the extension polish. The AI in this configuration costs you exactly what your existing API bill costs you. That’s the setup I’d recommend to any developer who already has an Anthropic key sitting in a .env somewhere.

Skip the Advanced AI add-on unless you have a specific reason to want frontier reasoning models one hotkey away. And students should note that Raycast offers a Pro plan 50% discount to verified students by completing a form, the student discount only applies to the Raycast Pro plan, and not the Advanced AI add-on .

For teams, the $12/user/mo Teams plan adds shared extensions, snippets, and quicklinks across the org, a fair upgrade if you want to standardize a “how we do things” layer on top of everyone’s launchers, but individual Pro seats are cheaper if you don’t need the shared library.

The bottom line

Raycast Pro is the tool I’d most miss if it disappeared tomorrow. Quick AI made my browser tabs quieter. AI Chat with MCP made “have the AI do it” actually feel like a workflow instead of a demo. Bring Your Own Key is the kind of pricing flexibility that should be standard and almost never is. It loses a few points to Alfred on pure file search and to itself on the still-cooking Windows build, but on the Mac in 2026 this is the most useful $8/month in software. If you’re still paying $20 for ChatGPT Plus and using it mostly for quick lookups, cancel it, spend eight bucks here, and thank me next week.