A little setup.
A lot less typing.

What’s new

Hlas 1.10.0 adds Bark 3.0, a local English speech model (macOS 15+), as a third option beside Whisper and Parakeet. Speech models settings now list every downloaded model with its size and a way to delete it, and show how much memory Hlas and its loaded models use.

Hlas 1.10.1 adds Bark to the setup walkthrough, with a Plus/Pro choice and a warning if your setup language is not English.

Hlas 1.9.0 added live meeting notes and final summaries. Text output now has Cleanup, Dictionary, and Translation sections; Meetings has Recording, Notes, and Reminders. Each category remembers its last section, and search opens the matching control.

Recent releases also added local speaker labels, a clearer dictation menu, five-step setup with practice, an indicator that follows your focused display, Undo Delete in History, and optional automatic update checks.

Read the 1.10.1 release notes. Already using Hlas? Right-click its microphone and choose Check for Updates.

1. Set up Hlas

You’ll need an Apple Silicon Mac (M1 or newer) with macOS 14 Sonoma or later. Meeting recording requires macOS 14.4+; Apple Translation requires macOS 15+.

To check your Mac, click the Apple menu in the top-left corner of your screen, then About This Mac. Look for an Apple M-series chip and the macOS version.

Step 1. Download and install

1. Download the app. Click Download Hlas. Wait for your browser to finish downloading a file named Hlas.dmg. A DMG is a Mac installation file.
2. Open the downloaded file. In Finder, choose Downloads in the left sidebar, then double-click Hlas.dmg. This opens the installer.
3. Copy Hlas into Applications. Drag the Hlas app onto the Applications folder in the installer. If you don’t see an Applications shortcut, open Applications in a second Finder window and drag Hlas there. Wait for copying to finish.
4. Open your installed copy. In Finder, click Applications and double-click Hlas. If macOS asks whether you want to open an app downloaded from the internet, check that it is Hlas and choose Open.

Step 2. Find Hlas in the menu bar

The menu bar is the thin strip across the top of your Mac’s screen, near the clock and Wi-Fi icon. Hlas appears there as a small microphone. It does not have an icon in the Dock along the bottom of the screen.

Click the microphone once to open Hlas’s small control panel. To open the app menu, right-click the microphone. On a trackpad, click with two fingers; you can also hold the Control key while clicking. Choose Settings in that menu whenever this guide refers to Hlas Settings.

Step 3. Complete the setup assistant

The five-step assistant saves your progress, so you can close it and return later. Use Continue and Back to move between steps.

1. Language: choose the language you speak. This helps Hlas recommend a speech model; it does not turn on translation.
2. Microphone: allow Hlas to record your voice when macOS asks.
3. Speech model: prepare the recommended Parakeet model, or choose Whisper or Bark (English only, macOS 15+) under Other speech models. All three run on your Mac. Download it once and wait for it to be ready; you can choose your text destination while it prepares.
4. Text destination: choose Paste into the app I’m using for automatic paste, or I’ll copy and paste myself. Automatic paste needs Accessibility access; manual copying works without it. The assistant offers the relevant permission steps.
5. Try it: record a short practice sentence with the shortcut shown. The default is ⌘ Shift Space. Practice stays in setup: it does not paste into another app, save to History, translate, or apply AI cleanup. Finish once the practice succeeds. You can change the shortcut and recording mode here.

What the permissions are for

Manage access in System Settings → Privacy & Security. To repeat setup, open Hlas Settings → Advanced → Show Setup Assistant.

2. Dictate your first text

Try this in a blank note first. You’ll know the basics are working before using Hlas for an email or message.

1. Open Notes and create a new note. Click in the note’s writing area. You should see a blinking vertical line—this is your cursor, and it marks where the words will appear.
2. Start recording. Hold down the Command key (⌘) and Shift, then press the Space bar. In Toggle mode, release all three keys and start talking. In Push-to-talk mode, keep holding the shortcut while you talk.
3. Say a short sentence. For example: “Remember to buy milk on the way home.” Speak at a normal pace. You don’t need to read out each letter.
4. Stop recording. In Toggle mode, press ⌘ Shift Space again. In Push-to-talk mode, release the keys.
5. Wait for your words. Hlas processes the recording after you stop. Your text should appear in the note when automatic paste is on; it does not need to appear word by word as you speak.
6. Read and correct the result. Click into the text and edit it just as you would something you typed. Voice recognition can make mistakes, especially with names.

Choose how you record

• Push-to-talk: hold the shortcut while speaking, then release to finish.
• Toggle: press once to start and again to stop.
• Cancel: press Esc during dictation to discard the take.

Change the shortcut and recording mode in Settings → Dictation. Click the menu bar icon for the Dictate button, recording mode, Paste automatically, and Fast/Polished text controls; right-click it for Settings, History, updates, and Quit.

Choose whether text is pasted automatically

1. Right-click the Hlas microphone and choose Settings.
2. Click Dictation on the left.
3. Find Text destination and turn on Paste into the app I’m using.
4. Leave Press Return after pasting off while learning. This lets you review a message before sending it.
5. Click back into your note before starting another dictation.

Make it fit your workflow

Paste into the app I’m using requests a paste into the app focused when you started recording and restores the previous clipboard afterward. Turn it off to copy the result yourself. Press Return after pasting is a separate, optional setting: it can send a message immediately in chat apps.

For feedback without opening the popover, enable On-screen indicator or Sound cues in Settings → Dictation → Feedback. Choose Pill or Orb and drag it to a convenient spot. Follow focused display is on by default: it follows the active window’s screen, or the pointer’s screen when focus is unavailable. Turn it off to keep the saved position. Reset indicator position restores the default placement.

3. Models & languages

You can leave your first model selected if it works well. Change it when you need a different language or want to try another balance of speed and accuracy.

1. Right-click the microphone, choose Settings, then click Speech models.
2. Under Processing location, select This Mac, then choose Parakeet, Whisper, or Bark (English only, macOS 15+). This keeps audio processing on your Mac.
3. With WhisperKit, choose a model and click Download if it is not already downloaded, or Select if it is. Base is a small starting option. With local Parakeet, let its model finish downloading and preparing. With Bark, choose Plus (about 200 MB) or Pro (about 1.2 GB), then let it download and prepare; the first preparation can take a few minutes.
4. Wait for loading to finish, then try the same short sentence in Notes again.

A language setting describes the language you are speaking; it does not translate your words. With WhisperKit, use the Language controls in Speech models settings to choose it. Automatic detection means Hlas tries to work it out for each recording. Local Parakeet detects its supported languages automatically. Bark understands only English and has no language selection; other languages come out as wrong text, not as an error, so Hlas warns you when your language is not English.

Speech models settings also list every downloaded model with its size and a Delete option under Downloaded models, and show a live memory meter for Hlas and the models currently loaded. Model figures are estimates; the total is measured.

Bark credit: “Underdog Bark 3.0” by Conway Research, licensed under CC BY 4.0. It derives from NVIDIA’s parakeet-tdt-0.6b-v3 and Moondream’s parakeet-ultra, also CC BY 4.0. Conway Research post-trained and compressed it; no endorsement is implied.

Larger Whisper models use more disk space and take longer to load. Models are cached, so you don’t download them for each session.

4. Clean up your text

AI cleanup is an extra step that rewrites spoken wording into cleaner text. Basic dictation works without it. Start with this feature only after your first dictation works.

1. Right-click the microphone, open Settings, and click Text output → Cleanup.
2. Choose Polished · English. Original keeps AI polishing off.
3. Allow its separate model download to finish (about 350 MB). If you see a Download or Retry button, use it and wait until the model is ready.
4. Open Polish options and choose a tone. Casual suits informal wording; formal suits more formal writing. You can leave the default tone while trying it. Keep the list and email options off for your first test.
5. Return to Notes and dictate the example below. Compare the result with what you said.

“S1-mini” by “Superwhisper” runs on your Mac. It removes fillers, resolves spoken corrections, adds punctuation, and formats spoken numbers and addresses.

• Tone: casual, semi-casual, semi-formal, or formal.
• Allow bullet lists: format three or more listed items as Markdown bullets.
• Format as email: arrange a greeting, body, and sign-off.

AI cleanup supports English text and is not applied to meeting recordings. If the cleanup model is unavailable or cannot produce a usable result, Hlas keeps the transcript without the AI rewrite.

Teach Hlas a name it keeps getting wrong

1. In Settings → Text output → Dictionary, find Personal Dictionary and turn on Enable personal dictionary.
2. Click the + below the list to add an entry.
3. Enter the word Hlas wrote incorrectly and the spelling you want it to use instead. For example, replace “glass” with “Hlas” only if that is a correction you want applied.
4. Try a new dictation containing that name. Dictionary entries affect later processing; they don’t rewrite earlier saved transcripts.

Simple text rules without AI cleanup

Dictionary replacements and any translation run before AI cleanup.

Without AI cleanup, you can enable Remove filler words for English fillers, and Smart punctuation for basic capitalization and a final period. Those two rules are disabled while AI cleanup is active.

5. Translation

Translation lets you speak in one language and receive written text in another. The target language is the language you want to write in.

1. Open Hlas Settings and click Text output → Translation.
2. Turn translation on and choose your target language—for example, English if you want English text.
3. If asked to download a language pack, let it finish while connected to the internet. This file lets the translation work on your Mac.
4. Return to your note and dictate one sentence in your spoken language. After you stop, check that the result is in your chosen target language.
5. Turn translation off when you want to write in the language you are speaking again.

• WhisperKit to English: uses Whisper’s built-in English translation.
• Apple Translation: supports translation with either local engine on macOS 15+. Download any requested language packs before using it offline.

If automatic language detection gives unexpected results, choose your spoken language in Speech models settings. Available translation languages depend on Apple’s installed language support. If translation fails, Hlas keeps the original text and skips English polishing for that result.

6. Record meetings

Hlas records a call without a meeting bot and transcribes the audio on your Mac. On macOS 14.4+, you can capture microphone audio (your voice), system audio (sound playing on your Mac), or both. Make a short test recording before your first full meeting.

1. Open Settings → Meetings → Recording. Choose Capture system audio, Capture microphone, or both. You can assign a separate Meeting shortcut here.
2. Join the call, click Hlas’s menu bar microphone, then choose Record a meeting. Allow System Audio Recording access if prompted. The meeting window shows elapsed time and activity meters for the selected inputs. Quiet meters can mean no one is speaking; they do not prove permission is missing.
3. Choose Finish meeting in the menu, or Stop & transcribe in the meeting window. Leave Hlas open while it combines and transcribes the tracks.
4. Read the result in the meeting window’s Transcript tab. To save it for later, enable Save new transcripts to History in Settings → Privacy & permissions before recording.

Meetings are never pasted into another app automatically. You can dictate while a meeting records; dictation pauses during final audio processing. With live notes enabled, brief draft transcription also shares the speech engine. If final transcription fails, use Retry transcription in the meeting window or Retry in the menu. The audio stays on your Mac for retry and is removed after successful transcription.

Add speaker labels

1. Before recording, open Settings → Meetings → Recording and enable Identify speakers. It is off by default.
2. Wait for the separate speaker model to download (about 200 MB). Retry the download here if needed.
3. Finish a meeting as usual. After transcription, Hlas shows Identifying speakers… and adds numbered labels to the final transcript.

Labels distinguish up to eight voices as Speaker 1, Speaker 2, and so on. When microphone and system-audio tracks clearly distinguish your voice, it can appear as Me. Hlas does not learn people’s names. If identification or word alignment fails, or only one speaker is detected, it keeps the complete plain transcript. Speaker processing is local and does not affect dictation.

To remove the speaker model, turn Identify speakers off and choose Delete Model in the same section after meeting processing finishes.

Set up calendar reminders

1. Open Settings → Meetings → Reminders and turn on Enable calendar reminders.
2. Allow Calendar access and notifications when prompted. Choose the calendars to watch; with none selected, no events match.
3. For a first test, leave title filters empty and the video-link requirement off. Later, you can filter by title keywords or require a Zoom, Google Meet, Teams, or Webex link.
4. Keep Hlas running. Use the reminder when a matching event begins.

Start meeting recording automatically adds a visible, cancelable 10-second countdown. Choose Cancel automatic recording in the meeting window to stop that countdown.

7. Live meeting notes

New in 1.9.0, optional notes turn transcripts into Summary, Decisions, Actions, and Open questions. They are off by default. Each item includes a source excerpt you can check against what was said.

Connect a notes model

1. Have an OpenAI-compatible server running with a model loaded, or obtain your provider’s endpoint and model ID. An endpoint is the address Hlas sends text to; a model ID names the model at that address. Hlas does not supply a default notes server.
2. Before starting a meeting, open Settings → Meetings → Notes.
3. Enter the Base URL. For LM Studio running on this Mac, a typical example is http://localhost:1234/v1. Use the address your server actually provides, including /v1 when required.
4. Enter an API key if your provider requires one. Local servers may not need it. Hlas stores the key in macOS Keychain.
5. Choose Test & list models, then choose an available model or enter its exact Model ID. Click Save connection.
6. Turn on Generate live and final meeting notes, then record a short test meeting. Connection settings are disabled while a meeting is active; saved changes apply to the next meeting.

Read, check, and export

The meeting window has Notes and Transcript tabs. With notes enabled, drafts refresh about every 30 seconds when the speech engine is available. Drafts can change or miss audio while the engine is busy; finishing uses the complete final transcript and includes speaker labels when enabled.

Expand Source excerpt to review a note’s evidence. Action owners or deadlines can be Unspecified when the transcript does not establish them. Source excerpts help you check a note; they do not guarantee that its interpretation is correct.

Use Copy notes or Export notes… for Markdown. Final notes are saved with the meeting when History saving is on. In History, select the meeting and use the Notes/Transcript control to read either version.

If a notes request fails, Hlas keeps the transcript and last successful draft. Use Retry notes after correcting the connection; Refresh notes requests a new version when available. Review generated notes before using them as minutes.

8. History & privacy

History is the list of transcripts Hlas has saved on this Mac. It helps you find a previous dictation or read a meeting after it finishes.

1. Right-click the Hlas microphone and choose History.
2. Click a transcript to select it. Use the search field to look for a word, or choose the Meeting filter to narrow the list to meetings.
3. Right-click an entry to open its menu. Choose Copy to put its text on the clipboard, then click in your destination document and press ⌘ V to paste.
4. Choose Export if you want a separate file. Pick a name and location in the save window.

If History is empty, open Settings → Privacy & permissions and check Save new transcripts to History. Turning it on saves future transcripts; it won’t recreate earlier unsaved recordings.

In Settings → Privacy & permissions, choose whether to save new transcripts and how long to keep them. Turning saving off leaves existing entries intact. Individual deletions offer Undo Delete during the current app session. Clear all and retention removal are permanent. Shortening retention asks for confirmation; it also applies to existing entries when saving is off.

What stays local

Local speech recognition, AI cleanup, and transcript history stay on your Mac. Dictation audio files are temporary. Meeting audio is retained if transcription fails so you can retry; it is removed after successful transcription.

Internet access is used for model downloads and update checks. Optional remote dictation sends audio to your configured server. Optional meeting notes send transcript text to the endpoint you choose. Hlas has no telemetry or analytics. Once text is inserted into another app, that app’s own storage and syncing settings apply.

9. Settings at a glance

Automatically check for updates in Settings → App is on by default. Hlas checks at startup and hourly while running. Failed automatic checks stay silent; available updates have a menu-bar indicator and versioned right-click item. Turn automatic checks off if you prefer, and use Check for Updates manually at any time.

10. Troubleshooting