AI Prompts

AI Prompts let Tuna run reusable instructions on text you already have in the launcher.

A prompt is just a Markdown file. Tuna shows those files as normal launcher items, so you can search for a prompt, send text to it, and keep working with the result.

What Prompts Are For

Use prompts for small text transformations you repeat often:

• summarize notes
• rewrite selected text
• turn rough ideas into a cleaner draft
• extract action items
• translate text
• make a message shorter or friendlier
• clean up a Talk Mode transcript

The result is plain text. That means it can continue through Tuna like any other Text Mode subject: paste it, copy it, send it somewhere else, or chain another action.

Running A Prompt

There are a few ways to run prompts:

• choose some text, then use Send to Prompt…
• choose a prompt first, then use Run with Input…
• use Send to Custom Prompt… for a one-off instruction you do not want to save

For longer runs, use the Shelf variants:

• Send to Prompt on Shelf…
• Run with Input on Shelf…
• Send to Custom Prompt on Shelf…

Inline runs put the generated text back into the current command flow. Shelf runs let the work finish separately and put the result on the Shelf.

Saving Your Own Prompts

Prompts live here:

~/Library/Application Support/Tuna/Prompts/

Each prompt is a Markdown file. The file name becomes the stable prompt identity.

A simple prompt can be only body text:

Summarize this in five bullets.

You can also add simple frontmatter:

---
title: Friendly Rewrite
description: Make text warmer and clearer without changing the meaning.
---

Rewrite the input in a friendly, direct tone.

Tuna uses title and description when present. Unknown frontmatter is ignored.

AI Provider Settings

Prompts run through the AI provider you choose in the Prompts extension settings.

Supported provider paths include:

• Apple/on-device models when available
• OpenAI-compatible endpoints
• localhost endpoints for models you run on your Mac
• custom remote endpoints

API keys and tokens are stored in Keychain.

Privacy

Prompt input is the text you choose to send to the configured provider.

If you use an on-device provider, processing stays on that provider path. If you use a localhost provider, Tuna talks to the service running on your Mac. If you use a remote endpoint, that endpoint receives the prompt input.

See Privacy and Local Processing for the broader privacy model.

Dictation Post-Processing

Prompts can also be used after dictation.

For example, Talk Mode can transcribe your speech and then run a selected prompt to clean up the transcript before the text continues through Tuna.

That makes prompts useful for turning rough spoken notes into paste-ready text.