Summary: Raycast AI is most useful when you stop treating it as another chat window and turn your repeated writing chores into keyboard commands. This playbook builds a small daily kit: a one-hotkey spelling fix, three custom AI Commands for replies, summaries and translation, a screen-aware "what am I looking at?" shortcut, and a weekday morning Automation. It also shows how to keep the monthly credit allowance from running out by moving routine jobs to a local model or your own API key.
Who this playbook is for
Raycast is a launcher for Mac and Windows (with an iOS app) that already handles app search, clipboard history, snippets and window management on its free plan. The AI features sit on top of that, and that is the reason to use them: you can run a prompt on the text you have selected in Slack, Mail, a browser or your code editor without copying anything into a separate tab.
The scenario here is a typical knowledge worker who writes a lot of short messages in English, often in a hurry, and sometimes needs to reply in another language. Everything below works on the cheapest paid tier. One thing to know up front: Raycast's own pricing FAQ says AI is not part of the Free plan, so a $0 account gets the launcher but none of the steps in this guide. New accounts get a 7-day Pro trial with 500 AI credits, which is enough to build and test the whole kit before paying.
Step 1: Turn on Quick Fix for typos everywhere
Quick Fix is the lowest-effort win. It runs the built-in Fix Spelling and Grammar AI Command on whatever field you are typing in and replaces the text in place.
- Type a message in any app, for example a Slack reply.
- Double-press the right Shift key, which is the default Quick Fix hotkey.
- On Mac, approve the Accessibility permission the first time Raycast asks, so it can read and replace text in the active app.
- Open Settings → AI → Quick Fix and decide on Prefer Selection: on (the default) it fixes only highlighted text, off it fixes the whole focused field.
On Mac there is also an experimental Preserve Formatting option for apps like Slack and Notion, which merges the corrections back so bold text, links and custom emoji survive. If your messages are mostly formatted, turn it on and check a few results before trusting it.
Step 2: Build three custom AI Commands
The built-in commands (Improve Writing, Change Tone to Professional, Summarize Webpage and others) are a good start, but the real value comes from commands written for your own habits. Search Create AI Command in Raycast. The only required field is the prompt; the rest control the name, model, creativity and where the answer goes.
The setting that matters most is Output Behavior. Replace Selection writes the result back over your highlighted text, which is ideal for rewrites. Open in Raycast shows the answer in a window where you can read it, ask a follow-up or paste it, which is safer for anything you want to check first. Turning on Highlight Editing Changes for rewrite commands shows exactly which words were changed.
Command A: Reply in my tone
Prompt: Draft a short reply to this message in a friendly, direct tone. Keep it under 80 words and do not promise dates: {selection}. Set Output Behavior to Open in Raycast, because a reply should always be read before it is sent. {selection} is a Dynamic Placeholder that inserts the text selected in the frontmost app.
Command B: Three-bullet TL;DR
Prompt: Summarize {selection} into three bullet points, then list any action items with owners if they are named. Use a low creativity setting so the summary sticks to the source. For long articles in a browser, the {browser-tab} placeholder inserts the content of the focused tab instead, but it needs the Raycast Browser Companion extension.
Command C: Translate with a language prompt
Prompt: Translate {selection} into {argument name="Language"}. Keep names, numbers and code unchanged. The {argument} placeholder makes Raycast ask for the language each time you run it, so one command covers every language you need. The manual notes you can use up to three different arguments in one command.
To run any of them: select text, open Raycast, type the command name and press Enter. If you would rather start from a built-in command, open Search AI Commands, select it and press ⌘ D (Ctrl D on Windows) to duplicate it, then edit the prompt.
Step 3: Ask about whatever is on screen
Some questions are hard to phrase without pasting a lot of context: an error dialog, a dense spreadsheet, a settings page in a language you do not read. Screen Awareness handles this.
- Run Send Focused Window to AI from Raycast. The first run opens Set Up Screen Awareness.
- On Mac, grant Screen Recording and Accessibility.
- Accept the suggested hotkey (a double tap of the right
⌘/Ctrlkey) or record your own.
After that, pressing the hotkey opens AI Chat with a capture attached: the app name and window title, readable text, your selection, the focused field and a screenshot. Click the attachment card to see what was actually included before you ask. There are narrower commands too, such as Send Selected Text to AI or Send Screen Area to AI, which are better when the window contains things you would rather not send.
Step 4: Schedule a weekday morning brief
Automations run a prompt on a schedule and drop the result into AI Chat. A sensible first one is a morning plan built from your calendar:
- Open Settings → AI → Automations, click Create Automation and choose Create Manually.
- Name it, pick a model, and choose an existing chat as the destination so each run can see the previous days' notes. Pick New Chat instead if you want separate daily reports.
- Write the prompt, mentioning tools with
@, for example@calendar list today's meetings and suggest two focus blocks between them. - Choose Daily, set a time and tick Monday to Friday under Days.
- Save, then use Run Automation Now from the actions menu to test it immediately.
Two limits from the manual shape how you use this. Runs happen only while your computer is awake and Raycast is open; a missed run fires once at the next launch. And automations do not sync, are stored on the device where you created them, and each device holds up to 20. Create them on the machine that is usually on at the scheduled time.
Step 5: Keep credits for the jobs that need them
Raycast AI runs on credits, and how many a message uses depends on the model and how much text it reads and writes. Long chats, attachments, tool calls and regenerated answers all add up, and the model picker shows a Cost meter next to Speed and Intelligence so you can compare before choosing. A practical split for this kit:
- Quick Fix: always uses Raycast AI credits, even if you set AI Commands to use another provider, so it is your fixed baseline cost.
- Summaries and translations: point these commands at a local model. In Settings → AI → Models & Providers → Local Models, install the Ollama app (there is a button for it), type a model name and click Install. Local conversations do not use Raycast credits and stay on your machine. If Ollama misbehaves, our Ollama troubleshooting guide covers the usual causes.
- Hard reasoning or long documents: add your own key under Models & Providers → API Keys for Anthropic, Google, OpenAI or OpenRouter, click Verify, then Save. Those requests are billed by the provider instead. An OpenRouter key also unlocks models Raycast has not added yet; see OpenRouter free models, limits and fees before you rely on its free models.
Check usage with Settings → Account → Show details, or type /usage in AI Chat. Bringing your own key, subscription or local model still requires a paid Raycast plan.
Raycast plans compared
Figures below are from Raycast's pricing page and manual on the verification date, in US dollars for an individual account.
| Plan | Price | AI credits per month | AI models | Notes |
|---|---|---|---|---|
| Free | $0 | None (no AI) | Not included | Launcher, extensions, Notes, 3 months of clipboard history; no BYOK or Cloud Sync |
| Pro | $10/month or $96/year | 500 | Fast and powerful models | 7-day trial with 500 credits for new accounts; BYOK, local models, dictation, Cloud Sync; 50% student discount |
| Pro+ (shown as Plus) | $20/month or $192/year | 3,000 | All models, including top-tier | Top-tier models are not part of the trial |
| Max | $50/month or $480/year | 7,500 | All models, including top-tier | Same features as Pro+ with a larger allowance |
Unused monthly credits roll over for one extra billing cycle (each grant expires after 60 days). Top-ups come in packs of 500, 2,500, 5,000 and 10,000 credits, expire 12 months after purchase, and are not available during the trial. Team plans are priced per user and are not covered here.
Editor's note
Who it's for: people who already live on the keyboard and write many short messages, emails or code reviews every day. The savings come from skipping copy and paste, not from better answers, so the more often you repeat a task, the more this setup pays off.
Who it's not for: anyone who wants free AI. The Free plan has no AI at all. If you mainly dictate rather than type, compare dedicated tools first in our Wispr Flow vs Aqua Voice vs Willow comparison.
Gotchas and fixes:
- Quick Fix does nothing on Mac. It needs Accessibility access: System Settings → Privacy & Security → Accessibility (renamed Device Control and Data Access on macOS 27).
- Requests still use Raycast credits after adding a key. Make sure the key is toggled on in API Keys, then re-pick the model in the command or chat. A small key icon next to the model name confirms it is using your key.
- A downloaded Ollama model is missing from the picker. Click Sync Models, and run
ollama listto confirm Ollama sees it. Note that Raycast caps the context sent to local models at 64k tokens. - An automation shows as Overdue or never ran. Raycast was closed or the target chat was busy or archived. Unarchive the chat, or switch the destination to New Chat, then use Run Automation Now.
- "No Credits Left" mid-answer. Top up or switch to a local or BYOK model, then choose Try Again. A provider rate-limit error is different: buying credits does not remove it.
FAQ
Q: Can I use Raycast AI without paying?
A: Only during the 7-day Pro trial for new accounts, which includes 500 credits and the Pro model list. After that, AI requires a paid plan; even using your own API key or a local model needs one.
Q: Does Raycast see my prompts if I use a local model?
A: According to the manual, local model requests go straight from Raycast to Ollama on your machine. The exception is tools Raycast runs on its own servers, such as image generation, which use a Raycast AI model and credits for that step.
Q: Do my AI Commands and Automations sync between computers?
A: Automations do not; they live and run on the device where you created them. For commands, you can generate a share link or import them from a JSON file with Import AI Commands.
Start with Quick Fix and one custom command this week, and add the rest only once you notice yourself reaching for them. The official Raycast screenshots are at the end of this post.
Images: Raycast pricing page captured by lightmind24; Raycast AI Command, Create AI Command and Bring Your Own Key screenshots from the official Raycast Manual.
Last verified: 2026-11-04
Sources: Raycast pricing, Raycast billing, AI Commands, Dynamic Placeholders, Screen Awareness, Automations, AI Credits, Local Models, Bring Your Own Key.
댓글
댓글 쓰기