How to use Parlando

Getting started

Parlando is dictation for the Mac. Hold a shortcut, speak, let go: the text appears where your cursor is, in any app.

1. Install. Open the downloaded Parlando.dmg and drag Parlando to Applications. Parlando needs macOS 14 Sonoma or later.
2. Open it. A short setup walks you through the rest: the microphone, the permission to type for you, and an engine.
3. Pick an engine. Download Parakeet Ultra (595 MB, once) to transcribe on your Mac, or paste a Gemini or OpenAI key to use the cloud.
4. Practise. The last step lets you try the real shortcut, ⌘⇧Space, inside the setup window.

Your first dictation

1. Click into any text field: a Mail draft, Slack, Notes, a browser.
2. Hold ⌘⇧Space and speak. The indicator appears with a timer, and its line moves with your voice.
3. Let go. A moment later the text is typed where your cursor was.

Prefer not to hold keys? Tap the shortcut once to start and again to stop.

Permissions

Parlando asks for two things. Both are standard macOS permissions you can change any time in System Settings › Privacy & Security.

Don't want to give Parlando the typing permission?

Choose Settings › Recording › After you stop › Copy it, I'll paste. Parlando copies the text and the indicator tells you to press ⌘V. No permission needed.

What Parlando never does

It doesn't log your keystrokes, read other apps' windows, or record in the background. Global shortcuts use the standard macOS hotkey API, which needs no permission.

Engines and languages

The engine turns your voice into words. Choose it in Settings › Engine.

Which language does Parlando listen for?

In this order, the first that's set wins:

1. the language you picked in the indicator for this one take,
2. the language of the Voice that's running,
3. the language you last set for this engine on the Engine page.

Parlando never translates unless a Voice or a Command Mode instruction asks for it.

Shortcuts for engine and language pairs

In Settings › Shortcuts you can bind a key to a pair, for example “Whisper in Czech” or “Parakeet in English”. Pressing it dictates with that pair once, without changing your defaults.

Dictating

Tap or hold

The dictation shortcut works both ways: hold it while you speak and let go to finish, or tap to start and tap again to stop.

Pause, cancel, auto-stop

• Pause with the pause shortcut. Paused time is dropped, so the transcript has no gap.
• Cancel with the × in the indicator or the cancel shortcut. Nothing is transcribed or saved.
• Sounds when recording starts and stops: Settings › Recording › Sound. Off by default.
• Auto-stop on silence ends the take after a pause of the length you choose. Off by default; turn it on in Settings › Recording.

Transcribe only

A second shortcut, transcribe only, skips the AI step: the words as you said them, with on-device cleanup.

The Record page

Parlando's main window opens on Record: the line and what you said. Press Space to start and stop, ↩ to finish, Esc to cancel. The result is copied and saved to History; it isn't typed into another app. Pick the Voice, engine and language from the menu at the top right.

Audio files

Drop an audio or video file anywhere on the window, or choose File › Transcribe Audio File… (⌘O). The transcript lands in History with a File badge.

Undo

Set an undo last dictation shortcut in Settings › Shortcuts. It sends ⌘Z to the app Parlando typed into, within 60 seconds and while that app is still in front.

Live text

With Gemini Live or OpenAI Live as your engine, words appear on a small glass card next to the indicator while you speak. The final text is typed once, when you stop, so the app you're in never sees half-finished words.

• The card shows the last three lines. Its darkness is in Settings › Appearance › Live text.
• If the stream fails, Parlando transcribes the recorded audio the normal way. You don't lose the take.
• Gemini has two modes: Verbatim (Parlando's own cleanup runs afterwards) and Smart (Gemini removes fillers and fixes punctuation itself, replacing Parlando's cleanup).

Live text needs your own Gemini or OpenAI key (Settings › API keys).

The indicator

The indicator floats over the app you're in while you dictate, and for a few seconds after. Pick a style in Settings › Appearance:

Position: top or bottom, left, centre or right. It appears on the screen your pointer is on. Change it in Settings › Appearance, or from the menu bar icon's right-click menu.

Per-take changes: click the language or Voice chip to change them for this dictation only. Your defaults stay as they are.

After the paste

The indicator says where the text went (“Pasted into Mail”) and offers your other Voices for a few seconds. Pick one and Parlando rewrites the pasted text in place.

The line

It follows your voice while Parlando listens, a highlight runs along it while Parlando transcribes and cleans up, and it lies flat when it’s done. A gap in the middle means no microphone. With Reduce Motion on, it doesn’t drift.

Shortcuts

Change any of them in Settings › Shortcuts. Only the first has a default.

In the main window

Voices and cleanup

A Voice is how Parlando writes for you. Each one combines:

• a cleanup level (below),
• an optional style prompt in plain words, like “Write it as a short, friendly email”,
• optionally its own AI provider and model, language and engine,
• optionally the apps it switches on in.

Everything a Voice does is compiled into a single AI call, and the editor shows you exactly what will be sent. A Voice with no cleanup and no style is Raw: no AI at all.

Cleanup levels

The wording of each level is editable at the bottom of Settings › Voices. On-device filler removal (no AI) is there too, and knows Czech hesitation sounds as well as English ones.

Voices per app

Add apps to a Voice and it switches on by itself when you dictate into them: an email Voice in Mail, a short one in Slack. The first matching Voice in your list wins, so drag to reorder.

Changing the Voice for one take

Click the Voice chip in the indicator. It applies to that dictation only. Already pasted? The indicator offers the other Voices for a few seconds after the paste; pick one and Parlando rewrites the text in place.

Try it

The Voice editor has a Try it box: it runs the Voice on a sample sentence, or your own, before you keep it.

Rewrite later

In History, Rewrite as runs any transcript through another Voice.

Command Mode

Rewrite text you already have, by saying what to change.

1. Set a Command Mode shortcut in Settings › Shortcuts.
2. Select text in any app.
3. Hold the shortcut and say an instruction: “make it shorter”, “more formal”, “turn this into bullet points”, “translate to Czech”.
4. Let go. Parlando replaces the selection with the result.

The indicator shows “Rewrite:” and the start of your selection while it listens. The undo shortcut, or the app's own ⌘Z, brings the original back.

Parlando reads the selection directly when the app allows it, and copies it (⌘C) otherwise. The result is pasted over the selection, so the app’s own undo works.

Spoken edits and punctuation

Change your mind

Say “scratch that” to drop the sentence you just said, or “delete last word” to take one word back. “Delete that” works when you say it on its own, between pauses. In Czech: “smaž to” and “smaž poslední slovo”. Self-corrections like “Thursday, no wait, Friday” are tidied by the Standard and Aggressive cleanup levels, keeping the version you meant.

Say the punctuation

Spoken punctuation is turned into symbols:

Dictionary and learning

Teach Parlando the words it gets wrong. Settings › Dictionary has three lists:

• Words Parlando should know: names and jargon, sent as hints to the engine.
• Always capitalize: “netsuite” becomes NetSuite, every time.
• Replacements: “cooper netties” becomes Kubernetes.

Learning from your corrections

When you fix a misheard word in a History transcript and click away, Parlando offers to remember it: Remember “NetSuite” for “net sweet”? Choose Remember and it goes into the Dictionary. Parlando only offers this for short, word-level fixes, not rewrites, and never adds anything without asking.

Snippets

Store blocks of text you type often and say a trigger to insert them. Add them in Settings › Snippets.

Say “insert address” (or “insert signature”, whatever you named it) in the middle of a dictation. The stored text is dropped in exactly as written; AI Voices work around it without changing it.

History and Ask

Every dictation is saved on your Mac with its audio. Open the main window › History.

• Search with ⌘F, filter by source (dictation, file, CLI).
• Read a transcript as a page: date, length, engine, language and Voice on top, the audio player, then the text. Edit the text in place.
• Replay the audio with a scrubbable waveform.
• Copy, Rewrite as another Voice, or Export as text, Markdown or audio.
• Select several to export or delete them together. Deleting always asks first.

Ask your history

Press ⌘K (or Ask your history at the top right) and ask a question: “What did I promise Maya?”, “What did I say about the budget this week?”. The answer comes only from your transcripts, with links to the takes it used. “today”, “yesterday” and “this week” narrow the search.

Keeping audio

Keep recordings forever, or let Parlando delete the audio after 14, 30 or 90 days. Transcripts are kept either way. Set it in Settings › General.

The menu bar

Click Parlando's menu bar icon for your last four dictations: copy one, or paste it into the app you were last in.

Insights

The Insights page counts your dictation, from your own History, on your Mac:

• words this week and the time saved over typing,
• a two-week bar chart and your best day,
• your streak, speaking pace and all-time total,
• how many fillers and false starts were tidied away,
• where you dictate most, and which Voices you use.

Only dictations count, not imported files or CLI runs. Time saved compares your pace with typing at about 40 words a minute.

CLI and MCP

The parlando command gives your terminal and your AI agents the same engine, Dictionary, Voices and History as the app. It prints only its result to stdout, so it's safe in pipes.

The command is inside the app. Install it once to put it on your PATH.

Install

"/Applications/Parlando.app/Contents/Helpers/parlando" install-symlink
# or into a folder you own:
"/Applications/Parlando.app/Contents/Helpers/parlando" install-symlink --target ~/bin/parlando

Settings › General › Command line shows whether it's installed and copies the command for you.

Updating from Aria, the app's earlier name? The aria command keeps working as an alias for parlando.

Commands

Ready-made ways to use it are in Recipes, below.

Useful flags: --voice NAME (case-insensitive, a unique start is enough), --raw, --language cs, --seconds 30, --until-silence, --no-save; for history --today, --since 2026-10-01, --search, --limit, --last, --json. Every command has --help.

If nothing was said, parlando dictate prints nothing and exits with status 1, so git commit -m "$(parlando dictate)" fails instead of committing an empty message.

Microphone in the terminal: macOS asks on behalf of your terminal app (Terminal, iTerm, Ghostty, VS Code), not Parlando. Allow it once in System Settings › Privacy & Security › Microphone.

MCP server

parlando mcp is a Model Context Protocol server over stdio with four tools: search_history, recent_transcripts, transcribe_file and rewrite_text.

# Claude Code
claude mcp add --scope user parlando -- parlando mcp

# Cursor: ~/.cursor/mcp.json
# Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json
{ "mcpServers": { "parlando": { "command": "/usr/local/bin/parlando", "args": ["mcp"] } } }

Then ask your agent things like “What did I promise Maya on Tuesday?” or “Summarise the recording in ~/Recordings/kickoff.m4a”.

The server hands your transcripts to whichever agent you connect it to, so add it only to agents you'd show your History.

Use the full path in apps started from the Dock; they don't always see your shell's PATH. History is re-read on every call, so new dictations show up right away.

Recipes

Small, real things to do with parlando and the MCP server. Each one works as written once the command is installed.

Transcribe a meeting and get the action items

With the MCP server connected, ask your agent in plain words: “Transcribe ~/Recordings/acme-kickoff.m4a and list the decisions and owners.” It calls transcribe_file; the audio is transcribed on your Mac. Without an agent:

parlando transcribe ~/Recordings/acme-kickoff.m4a --no-save \
  | claude -p "List the decisions, the owners and the dates"

Voice memos become notes

Every recording dropped into ~/Memos/Inbox becomes a Markdown note in ~/Notes, and the audio moves to ~/Memos/Done. launchd runs the script whenever the folder changes.

#!/bin/bash
# ~/bin/memos.sh
PATH=/usr/local/bin:$PATH
shopt -s nullglob
for f in ~/Memos/Inbox/*.m4a; do
  parlando transcribe "$f" --no-save \
    -o ~/Notes/"$(basename "$f" .m4a)".md && mv "$f" ~/Memos/Done/
done

<!-- ~/Library/LaunchAgents/com.example.memos.plist -->
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0"><dict>
  <key>Label</key><string>com.example.memos</string>
  <key>ProgramArguments</key>
  <array><string>/bin/bash</string><string>/Users/you/bin/memos.sh</string></array>
  <key>WatchPaths</key>
  <array><string>/Users/you/Memos/Inbox</string></array>
</dict></plist>

mkdir -p ~/Memos/Inbox ~/Memos/Done ~/Notes
launchctl load ~/Library/LaunchAgents/com.example.memos.plist

launchd doesn't expand ~ in a plist, so write the full paths there. Add --voice NAME to tidy each memo with one of your Voices.

Commit by voice

git commit -m "$(parlando dictate --voice 'Vibe Coding')"

If you say nothing, parlando dictate exits with status 1 and the commit doesn't happen.

Speak a prompt to Claude Code

parlando dictate | claude -p

Ask what you said

With the MCP server connected, ask “What did I promise Maya on Tuesday?” The agent calls search_history and answers from your takes. From a shell:

parlando history --search "Maya" --limit 5 --json \
  | claude -p "What did I promise Maya?"

Speak an email onto the clipboard

parlando dictate --until-silence | parlando rewrite --voice email | pbcopy

# keep it as a one-word alias
alias mailit='parlando dictate --until-silence | parlando rewrite --voice email | pbcopy'

Rewrite the clipboard in a Voice

pbpaste | parlando rewrite --voice email | pbcopy

Dictate straight to the clipboard

parlando dictate --until-silence | pbcopy

Today's standup, as action items

parlando history --today --search standup --json | claude -p "Turn these into action items"

A journal that writes itself

Every evening, cron hands the day's dictations to Claude and files the entry.

#!/bin/bash
# ~/bin/journal.sh
/usr/local/bin/parlando history --today --limit 100 --json \
  | claude -p "Write tonight's journal entry from what I said today" \
  >> ~/Journal/$(date +%F).md

$ crontab -l
30 21 * * * ~/bin/journal.sh

Cron runs with a short PATH, so use full paths in the script (/usr/local/bin/parlando, and wherever claude lives). In a crontab line itself, a % must be escaped as \%; calling a script avoids that.

Settings map

Open Settings with ⌘, or the Settings row at the bottom of the sidebar. Each row shows its current value.

Backup and moving Macs

Settings › General › Backup exports your Voices, Dictionary, Snippets and shortcuts to one .ariaconfig file. Import it on another Mac. API keys are deliberately left out; add them again there.

Privacy

• No account, no analytics, no Parlando server. We never receive your audio, text or usage.
• On-device by default. Parakeet and Whisper run on your Mac.
• Cloud only with your key. Cloud engines, live text, AI Voices, Command Mode and Ask send data directly from your Mac to the provider you chose.
• Local storage. History, audio and settings live in your user Library. Keys live in the macOS Keychain.

The full policy, with each provider listed: Privacy.

Troubleshooting

The indicator shows, but no text appears in my app

Parlando can't type for you yet. Open System Settings › Privacy & Security › Accessibility and switch Parlando on. If it's already on (for example after an update), switch it off and on again. Or use Settings › Recording › After you stop › Copy it, I'll paste.

Nothing happens when I press the shortcut

• Another app may own the same keys. Pick a different shortcut in Settings › Shortcuts.
• Check that an engine is ready: Settings › Engine. Without one, the indicator says so and links there.
• After the trial, recording waits for a licence. The indicator and the Record page say so.

The first word gets cut off with AirPods

Bluetooth microphones take a moment to wake. Start speaking half a second after the indicator appears, or choose the built-in microphone in Settings › Recording. Parlando follows you when you switch between AirPods and the built-in mic.

A Keychain prompt asks about “com.openclaw.aria”

That's Parlando reading your API keys. Click Always Allow so it doesn't ask again.

Live text doesn't appear

Live text needs Gemini Live or OpenAI Live as your engine and a working key in Settings › API keys. If the stream drops, Parlando still transcribes the take; you just don't see words while you speak.

An AI Voice returns the text unchanged

The Voice has no AI model it can run. The Voice editor shows an amber note when that's the case; set one in Settings › AI model and add the key.

Czech filler words survive

On-device cleanup removes hesitation sounds (ehm, eee, hm) in Czech but keeps words like no, jako or prostě, which can carry meaning. The Light level and above remove them when they're only fillers.

The model download stalls

Check your connection and free disk space (about 600 MB per model), then start the download again from Settings › Engine.

FAQ

Does Parlando work offline?

Yes, with Parakeet or Whisper and on-device cleanup. AI Voices, live text, Command Mode and Ask need a connection to your provider.

What does the AI cost me?

Whatever your provider charges for the calls you make, billed to you directly. A cleaned-up dictation is a short request; most cost a fraction of a cent.

Is it a subscription?

No. 14 days free, then one licence: $49, or $39 at launch. Every 1.x update is included, and you have 14 days to ask for a refund.

How many Macs can I use it on?

Up to three Macs you use yourself at a time. See the licence terms.

How do updates arrive?

Parlando checks for updates itself and asks before installing. Updates are signed; the check sends only the app version.

Does it work on Intel Macs?

Parlando needs macOS 14 Sonoma or later. Local models run best on Apple silicon; on older Macs pick a smaller Whisper model or a cloud engine.

Can I use it in several languages?

Yes. Set a language per engine or per Voice, bind engine-and-language pairs to shortcuts, or switch in the indicator for one take.

Where is my data?

In your user Library on your Mac, under Application Support/Aria (the folder keeps the app's earlier name). We have no copy.