Raycast AI | AI-Lab

What It Is

Four custom AI commands built inside Raycast (Summarize Page, Promptify, Jargon, and Summarize YouTube Transcript) to embed AI wherever I'm working on macOS. Raycast is a Spotlight-style launcher that provides file search, window management, clipboard history, snippets, and a long catalog of community extensions.

What I Learned

Custom shortcuts matter even with desktop apps in the picture — LLM desktop apps and browser plugins have lowered the friction, but a custom command on a Raycast hotkey lets me write the prompt once, lock in the format and voice I want, and recall it instantly without re-explaining myself. I also generally prefer more functionality inside tools I already use rather than a dozen hyperspecialized ones each accessed a different way.
Choosing the right model for a task comes down to evaluating on real inputs — For testing the YouTube command, I ran the transcripts for 2 videos I'd already watched through Claude Haiku 4.5 and Gemini 3.1 Flash-Lite to see which output matched my preferences. Gemini won here on speed and staying truer to the creator's phrasing.
The best prompt rules came from watching the model do something wrong — I added Promptify's new:original length ratio rule after early outputs were too verbose, and "Preserve Author's Voice" went into both summarizers because the model tended to drift into meta-commentary and lose the speaker's POV.

How It Works

Each command is a small config inside Raycast: a system prompt, a model selection, and a {clipboard} or {browser-tab} placeholder for the input
Summarize Page runs on the active browser tab and returns a thesis-first summary that preserves the author's tone and POV
Promptify takes whatever's in the clipboard and rewrites it as a precise prompt for an AI model, with rules on length, formatting, and not inventing details that weren't in the original request
Jargon is the smallest of the four: one instruction to strip acronyms and jargon from the clipboard and return only the rewritten text, no headers or follow-ups
Summarize YouTube Transcript takes a transcript pasted into the clipboard and returns a TLDR, core ideas, notable insights, 2-3 verbatim quotes, and any frameworks the creator walked through

Built With

Raycast hosts the four custom commands I can launch from anywhere on macOS. The web summarizer, prompt rewriter, and de-jargon command run on Claude 4.5 Haiku, and Gemini 3.1 Flash-Lite summarizes YouTube transcripts. ChatGPT Images 2.0 generated hero image.

Loading content...

What It Is

What I Learned

Custom shortcuts matter even with desktop apps in the picture — LLM desktop apps and browser plugins have lowered the friction, but a custom command on a Raycast hotkey lets me write the prompt once, lock in the format and voice I want, and recall it instantly without re-explaining myself. I also generally prefer more functionality inside tools I already use rather than a dozen hyperspecialized ones each accessed a different way.

Choosing the right model for a task comes down to evaluating on real inputs — For testing the YouTube command, I ran the transcripts for 2 videos I'd already watched through Claude Haiku 4.5 and Gemini 3.1 Flash-Lite to see which output matched my preferences. Gemini won here on speed and staying truer to the creator's phrasing.

The best prompt rules came from watching the model do something wrong — I added Promptify's new:original length ratio rule after early outputs were too verbose, and "Preserve Author's Voice" went into both summarizers because the model tended to drift into meta-commentary and lose the speaker's POV.

How It Works

Each command is a small config inside Raycast: a system prompt, a model selection, and a {clipboard} or {browser-tab} placeholder for the input

Summarize Page runs on the active browser tab and returns a thesis-first summary that preserves the author's tone and POV

Promptify takes whatever's in the clipboard and rewrites it as a precise prompt for an AI model, with rules on length, formatting, and not inventing details that weren't in the original request

Jargon is the smallest of the four: one instruction to strip acronyms and jargon from the clipboard and return only the rewritten text, no headers or follow-ups

Summarize YouTube Transcript takes a transcript pasted into the clipboard and returns a TLDR, core ideas, notable insights, 2-3 verbatim quotes, and any frameworks the creator walked through