AI Actions
Three defaults (Basic, Polish, Translate) plus your own. Auto on streaming, on-demand on recordings, standalone on any selected text.
AI Actions are how Quilin turns what you said into what you wanted to say. Three ship with the app. You can write your own. They run in three places: auto-applied on streaming, on-demand on recordings, standalone on any selected text.
The three defaults
| Action | What it does |
|---|---|
| Basic | Cleans the raw transcript — fixes punctuation, capitalization, drops the "um"s. The lightest possible polish. |
| Polish | Full rewrite to a chosen tone — Slack-casual, email-formal, document-precise. The transcript becomes the draft. |
| Translate | Target-language picker, formatting preserved. Same recording, second language, ready to send. |
These three cover what people actually do with a transcript: drop it in cleanly, rewrite it for a specific audience, or send it in another language. We picked the model and the prompt for each one so they Just Work. No knobs.
“Let’s ship the bilingual landing 下周, then start prep for the App Store review.”
Three places they fire
- Auto-applied on streaming. You pick one Action as your default streaming polish. As you talk, the transcript lands at your cursor with that Action already applied. One Action, live, every time. Most people leave it on Basic.
- On-demand on recordings. Hit any Action button on a saved recording. Re-run a different one tomorrow. Many Actions, same recording, different views.
- Standalone on selected text. Select any text anywhere on macOS, hit the AI Action shortcut, and the selection is replaced with the transformed version. No recording needed — works on what's already on screen.
Re-runnable, on purpose
Most dictation tools treat transcription as one-shot: speak, get text, done. Quilin keeps the audio and lets you re-run any Action against it. So:
- You ran Polish on a meeting for a Slack-casual draft. You actually need it for an email. Re-run Polish with the email-formal tone — the old draft is in the history, not overwritten.
- You ran Basic during streaming and want a fuller rewrite. Open the recording and hit Polish — same audio, different output.
- A week later you decide you want the recap in Chinese. Re-run Translate on the recording, or run it on the Polish output. Either works.
The point isn't that each Action is perfect on the first try — it's that the audio stays and you can keep working with it.
Write your own
Name it, write the system prompt, pick the model (or stay on the default). Custom Actions are stored on your Mac; sync is opt-in. Use them the same way as the defaults — auto on streaming, on-demand on recordings, standalone on selected text.
Examples people set up:
- Summarize with the headings and bullet style your team uses.
- Action items that extract todos with owners.
- Rewrite as email with your signature and salutation pattern.
- Code-comment cleanup for dictated code reviews.
If you write one that more than a few people might want, send it over — we collect community Actions and ship the popular ones as starters.
Local first. Cloud opt-in.
The three default Actions run on-device by default — your transcript doesn't leave the Mac. Custom Actions can opt into cloud providers (bring your own key, or use the paid tier). You choose, per Action, which way each one runs.
See transcription backends for how the same local-vs-cloud trade-off shows up on the dictation side.
Why no BYOK on the defaults?
A common ask is "let me wire my own API key into the built-in Actions". We're not doing that on the three defaults, on purpose.
The product is the choice. For each default Action we picked the model, the prompt, the temperature, and the output format. Letting users bring their own keys would either degrade the experience silently (cheaper model, same UI, worse output — bad outcome) or push the burden of model selection onto someone trying to write an email faster (also bad outcome).
Custom Actions are where BYOK lives. If you want raw provider access and full control over the prompt, write a custom Action. That's the whole point of the custom path — defaults are managed for "Just Works", customs are yours.
See pricing for the current tiers and what's metered.
What about the model behind each default Action?
We don't publicize the exact model per Action because we'll swap it as better options ship. The pattern is: pick the smallest model that gets the output reliably right, until the next size up is meaningfully better and worth the latency.
If you notice a default Action consistently producing bad output for your use case, tell us — we re-prompt and re-bench these regularly, and "Polish feels worse than last month" is a real bug to us.