Promptly Download

Talk to any app on your Mac.

Free, and it works anywhere there’s a text cursor: Slack, Gmail, your code editor, the terminal.

On an Intel Mac? Download the Intel version. Not sure? Apple menu › About This Mac says Chip on Apple Silicon, Processor on Intel.

Words per minute, for most people

Speak at 150.
Type at 40.

So talk instead. Everything floating here was spoken, not typed. Your words land where your cursor is, or come back from Claude ready to paste into whatever you’re working in.

Hold the button. Talk. Let go.

Same words, different destination. Pick where your line goes, then hold the button (or Space while it’s focused). In the app it’s a double-tap of ⌃, from anywhere on your Mac.

so for the support dashboard, um, I want it to pull tickets from zendesk every five minutes, group them by product area, and flag anything that’s been waiting more than like four hours, and the support lead wants a summary email at 9

Structured prompt

ready to paste into any AI 58 86
Goal
Build a small internal dashboard that pulls support tickets from Zendesk every five minutes, groups them by product area, and highlights anything that has been waiting more than four hours, so the support lead can see at a glance where customers are stuck.
Context
The support team lead wants a daily summary email at 9am.
Steps
  1. Write the sync job and check it against a sample Zendesk response.
  2. Build the grouped table and the four-hour flag.
  3. Add the 9am summary email.
Output format
A short plan, then the code for each step with its tests.
Success criteria
A ticket waiting 4 hours and 1 minute is flagged; one at 3 hours 59 is not. No email addresses end up in the database.
Assumptions (check)
Zendesk API token is available as an environment variable.

promptstask briefsn8n workflowsVeo shotsharnesses
emailsdesign briefsimage promptsclean textyour exact words

Ten destinations. Each gets its own format.

Dictation types what you said. The nine Craft modes send it to Claude and bring back what that destination actually takes.

  1. Dictation

    Your words typed where your cursor is. Nothing rewritten, no AI call.

    You said: “can you move standup to ten tomorrow, I’ve got the dentist”

    Slack#team

    can you move standup to ten tomorrow, I’ve got the dentist

  2. Prompt

    A rough idea becomes a detailed prompt: goal, context, requirements, constraints, steps, output format, success criteria, assumptions. Written for Claude; one pick rewrites it for Gemini, ChatGPT or Grok.

    You said: “so for the support dashboard, um, I want it to pull tickets from zendesk every five minutes…”

    Goal
    Build a small internal dashboard that pulls support tickets from Zendesk every five minutes, groups them by product area, and highlights anything waiting more than four hours.
    Success criteria
    A ticket waiting 4 hours and 1 minute is flagged; one at 3 hours 59 is not. No email addresses end up in the database.
  3. Code

    A task brief Claude Code can finish and verify, or rewritten for Gemini CLI, Codex or Grok.

    Example. You said: “add the four hour flag to the ticket table and make sure it’s tested”

    Task
    Flag tickets waiting more than four hours in the grouped ticket table.
    Done when
    npm test passes with a case at 4h 01m (flagged) and 3h 59m (not flagged).
  4. Design

    A new web page, or changes to one you’ve built, as a brief for working HTML and CSS.

    Example. You said: “the pricing cards should stack on phones and match the new header”

    Change
    Pricing cards use the header’s type sizes and spacing.
    On phones
    Cards stack in one column below 640 px.
    Keep
    Current copy and prices. Output working HTML and CSS.
  5. Polish

    Paste rough text, speak the change, get it back clean.

    Example. Pasted: “hey so deploy is gonna be late cuz qa found stuff” · You said: “make it calm and clear”

    The deploy is moving: QA found a few issues and we’re fixing them first. I’ll share the new time today.

  6. Email

    Speak the situation, get a ready-to-send email in the tone you pick.

    Example. You said: “tell Priya the invoice will be a week late, keep it warm”

    Subject Invoice for March, a week later than planned

    Hi Priya, a quick heads-up: the March invoice will reach you a week later than we said. Nothing is wrong on our side, it’s just taking longer to close the month. Thanks for bearing with us.

  7. Image

    An image prompt for Nano Banana and ChatGPT, with optional Midjourney flags.

    Example. You said: “coffee on a rainy café window, morning, moody greens”

    A cup of black coffee on a rain-streaked café window ledge, soft morning light, shallow depth of field, muted greens.

    --ar 3:2 --style raw Midjourney flags, optional

  8. Video

    A video prompt for Google Veo 3.1: camera, style, sound and dialogue.

    You said: “a boat leaving the harbour at dawn, slow push in, warm light, gulls”

    A slow push-in on a fishing boat leaving a quiet harbour at dawn, medium shot, golden-hour light with a teal and orange grade, 16:9 at 1080p, gentle ambient sound of gulls and water.

  9. Workflow

    Plain English becomes importable n8n workflow JSON.

    You said: “when someone fills in the typeform add a row to google sheets and post in slack”

    1. Typeform Trigger
    2. Append to Google Sheets
    3. Post to Slack
  10. Harness

    Describe a job, get a loop Claude Code runs until it’s done, with checks, stop conditions and an optional schedule on your Mac.

    You said: “every night fix the failing tests one at a time and never touch the migrations”

    daily 02:00  Pick › Fix › Check › Record

    Stops when all tests pass, a test still fails after 3 tries, or after 25 passes or 2 hours. Never edits db/migrations/.

  • Iterate. Speak a change to any result, “make it shorter” or “add a test”, and it’s revised.
  • Your words. Names and terms it should spell right (N10 becomes n8n). Correct a word twice and it’s added for you.
  • It writes like you. Notes about how you write shape Email and Polish, and grow as you edit, each change with Undo.
  • History. Everything you’ve said and made, searchable.

One prompt. Any AI.

Promptly writes for Claude first. Pick Gemini, ChatGPT or Grok and it rewrites the same prompt the way that AI reads best, with every detail kept. Standard is one version that works on all of them.

You don’t need a key for any of them: the prompt is only reformatted, by the AI Promptly already uses.

Support dashboard

written for Claude
Goal:
Build a small internal dashboard that pulls support tickets from Zendesk every five minutes, groups them by product area, and highlights anything waiting more than four hours.

Context:
The support lead wants a summary email at 9am.

Requirements:
- Sync every five minutes; a failed sync retries on the next run.
- Group by product area, with a count per group.
- Flag tickets waiting over four hours at the top of each group.

Success criteria:
- A ticket waiting 4 hours and 1 minute is flagged; one at 3 hours 59 is not.
- No email addresses end up in the database.

Twenty-four seconds, start to finish.

From a double-tap of Control to a scored prompt, five outputs and nothing uploaded but text.

Scored against what you actually said.

Any prompt can be scored next to your original words, so you see what improved and what’s still missing. This is the support dashboard line from the top of the page.

5886: what you said scored 58, the prompt scored 86.

Clarity60 › 88
Specificity45 › 85
Context55 › 80
Actionability40 › 90

What you said The prompt

Still missing Neither says where the Zendesk API token comes from.

Promptly's window in light mode: the finished prompt, and below it the score panel showing 58 to 86 with bars for Clarity, Specificity, Context and Actionability.
The same result in the app.

Your voice never leaves the Mac.

Speech is transcribed on your Mac by the built-in whisper.cpp, in 90+ languages. Only the text goes to the AI you chose: Claude through your own Claude Code sign-in, or OpenAI, Gemini or Grok with your own key. Dictation doesn’t call an AI at all.

No extra bill

Promptly runs Claude through the Claude Code plan you already pay for, so there’s no second AI subscription. No Claude plan? Add your own OpenAI, Gemini or Grok key in Settings instead. If Claude Code isn’t set up, Promptly installs it or signs you in from inside the app.

Audio stays inside your Mac and is turned into text there. Only the text travels to your AI, on your own account. Promptly has no servers. Your Mac microphone whisper.cpp text Your AIyour account Promptly serversthere aren’t any
Audio sent
0 B
Text
to your own AI account only
Promptly servers
none
AI bill from Promptly
none, uses your Claude plan or your own key

Set up in three steps.

  1. Install it

    Paste one line into Terminal and it opens straight away, or download the app and allow it once. Both ways are below.

  2. Allow the mic, connect Claude Code

    Give it the microphone, and Accessibility so it can type into other apps. Setup installs Claude Code or signs you in, right there.

  3. Double-tap Control and talk

    A small pill shows while you talk: recording, thinking, done. Prefer another key? Option+Space, hold Right Option, or hold Fn/Globe.

Two ways to install.

Promptly is signed by its maker but not notarised by Apple, so a browser download needs one extra click the first time. The Terminal line skips it.

One line in Terminal

Open Terminal, paste this and press Return. It downloads the latest release, checks it’s signed by Promptly, puts it in Applications and opens it. It also updates an older copy.

curl -fsSL https://aakashdhar.me/promptly/install.sh | bash

Prefer to look first? Read the script. It only touches Promptly.

Download the app

  1. Download the disk image and open it.
  2. Drag Promptly into Applications.
  3. Open Promptly. When macOS says it can’t check it, choose Done, not Move to Bin.
  4. Go to System Settings › Privacy & Security and click Open Anyway. You only do this once.
Download for Mac

On an Intel Mac? Download the Intel version. Not sure? Apple menu › About This Mac says Chip on Apple Silicon, Processor on Intel.

Built with the vibe skills.

Promptly was planned, built, reviewed and shipped with vibe-*, an open set of Claude Code skills I wrote. Every feature starts as a written spec and a task list, and nothing moves to the next phase until a review gate passes.

  1. brainstorm:

    Brief

    Who it’s for, what the first version does, and what it deliberately doesn’t.

  2. architect:

    Rules

    The structure and rules every later change is checked against.

  3. feature:

    Spec first

    A spec with acceptance criteria, a plan and a task list, before any code.

  4. next

    Build

    One task at a time. Tests after each one; code and docs committed separately.

  5. review:

    Gate

    No phase moves on with a serious finding still open.

  6. release

    Ship

    A signed build, a GitHub release, and this page’s download updated.

Commits since 18 April 2026
861
Feature specs written first
39
Spec reviews
22
Code review gates
28
Bug fixes with a written diagnosis
13
Automated tests
394
Releases
49

Before you download.

The honest parts, up front.

Does it work on Windows or Linux?

Windows, yes: Windows 10 (22H2 or later) and Windows 11 on x64 PCs. Linux isn’t supported.

Do I need a Claude subscription?

Not necessarily. Craft works best with Claude Code on a Pro or Max plan, and uses that plan, so there’s no extra bill. Without one, add your own OpenAI, Gemini or Grok key in Settings › AI; that company bills you for what you use. Dictation needs neither.

Why does macOS warn me the first time?

The app is personally signed rather than notarised by Apple. Open System Settings › Privacy & Security and choose Open Anyway. You only do this once. Or install it with the one line in Terminal, which doesn’t need that step.

What permissions does it ask for?

The microphone, to hear you. Accessibility, to type the result into the app you’re using. Nothing else.

Does my audio get uploaded anywhere?

No. Transcription runs on your Mac with the built-in whisper.cpp. Only the text of a Craft request goes to the AI you chose: Claude through your own Claude Code sign-in, or your own OpenAI, Gemini or Grok key. Promptly has no servers of its own.

Can I use the prompts in ChatGPT or Gemini?

Yes. Prompts are written for Claude first. Pick Gemini, ChatGPT or Grok in the “For” menu next to Copy and Promptly rewrites it the way that AI works best, keeping every detail. Standard is one version that works on all of them. You don’t need a key for that AI; it’s only reformatted.

Which languages can I speak?

Over 90. There’s an optional “Best accuracy” model you can download in Settings if you want fewer mistakes.