Skip to main content
To use Utter offline, download a local voice model, select it for your recording mode, and use on-device AI processing if that mode rewrites your text. Transcription and AI rewriting are separate stages. Selecting a local voice model does not make a cloud AI model work offline. Download the models while connected, then test the complete mode without a network connection before relying on it for private or offline work. Local model availability depends on your device and app version. Choose from the local models actually offered in your picker; not every model supports every language or speaker-label feature.

​
Overview

Use Utter locally in two stages to avoid paid transcription or AI usage:
  1. Local transcription: Choose an on-device speech model (Parakeet) for voice-to-text.
  2. Free AI processing: Choose Apple Intelligence as your default AI model.
What you get: On-device transcription without a per-minute provider charge. Local processing still uses your device’s storage, memory, and battery. Cloud providers and BYOK are separate, internet-connected options.

​
Stage 1: Set Up Offline Transcription

On Mac, select an available local transcription model, such as Parakeet, in the default model settings. Check the mode you actually use as well: a mode-specific model can differ from the default.

​
Mac

1

Open Advanced

In Utter, go to Settings > Advanced.
2

Change the Default Transcription Model

At the top of Default Models, select a Parakeet model for Transcription. Built-in models are marked with a MacBook icon in the model list.
Select a Parakeet local voice model in Utter advanced settings
3

Let the Model Download

Selecting a local model starts its download. Wait until it is ready before disconnecting. Transcription using that model runs on your device.
What to expect: New recordings use the Parakeet model, continue working without internet for transcription, and do not use paid transcription minutes.
A local voice model covers speech-to-text only. Cloud AI rewriting, remote servers, and syncing may still need a connection.

​
iPhone

1

Open the Mode Editor

Create or edit the mode you want to use for offline dictation in Utter.
2

Choose a Local Voice Model

Open Advanced Settings, then Model Selection > Voice Model. Choose a local model available on your device and allow its download to finish.
3

Check the Voice Language

In Advanced Settings > Voice Language, choose a language supported by that model. Save the mode.
What to expect: A new recording made with this mode uses the selected local voice model. If no compatible local model appears, update Utter and check your device’s available model options; do not assume a cloud model will work in Airplane Mode.

​
Stage 2: Use Apple Intelligence for AI Processing

Apple Intelligence is the recommended local AI model for most users. It works well for simple cleanup, short summaries, and lightweight writing help.
Utter’s Apple Intelligence model integration requires macOS 26 or iOS 26 or later, compatible Apple Intelligence hardware and language settings, and a downloaded system model. Enable Apple Intelligence & Siri in your device’s settings. The picker reports when the model is unavailable or still preparing.
1

Open Advanced

On Mac, go to Settings > Advanced. On iPhone, edit your mode and open Advanced Settings > Model Selection.
2

Choose Apple Intelligence

Open the AI Model picker and select Apple Intelligence. Check any mode-specific model selection too.
Apple Intelligence selected in the Utter AI model picker
3

Test One Short Recording

Record a short sentence, then use a mode that applies AI processing, such as cleanup, summary, or writing help.
What to expect: Utter’s Apple Intelligence integration uses the on-device system language model without a provider API key. Long transcripts or unsupported requests may exceed its capabilities; test with a short recording first. If that did not work: Check the operating-system requirement above, enable Apple Intelligence, and wait for the system model to finish preparing. Restart Utter and select the model again. A cloud-model fallback needs internet and is not an offline substitute.

​
Verify That Your Whole Mode Works Offline

1

Finish Downloads While Online

Confirm both your voice model and any local AI model are ready.
2

Disconnect Your Device

On Mac, disconnect Wi-Fi and any wired network. On iPhone, enable Airplane Mode and ensure Wi-Fi is also off.
3

Make a Short Test Recording

Use your chosen mode to dictate a sentence with a name or phrase you can check.
4

Check the Final Output

Confirm transcription finishes and, if enabled, AI rewriting finishes too. Reconnect when the test is complete.
If transcription succeeds but rewriting fails, recheck AI Model rather than changing microphone permissions. If the recording is silent, use the permissions guide. If a particular language fails, check model and language selection.
Apple Intelligence is the fastest setup because it does not require installing LM Studio, Ollama, or a separate local server.

​
Advanced Fallback: Use LM Studio or Ollama

Use this path only if Apple Intelligence is not available on your Mac, or if you already run a local OpenAI-compatible server.

​
LM Studio

1

Open the Developer Tab

Open LM Studio, enable Power User Mode, then open the Developer tab.
LM Studio power user mode and developer tab
2

Load a Model and Start the Server

Make sure a model is loaded, enable the server, then copy the Model ID and Base URL from the Developer tab.
LM Studio server enabled with model ID and base URL
3

Open Utter Custom Models

In Utter, go to Settings > Advanced > Custom Models, then choose Custom OpenAI Compatible as the provider.
Utter custom model provider set to OpenAI compatible
4

Add the Local Model

Click Add Custom Model. Paste the Model ID and Base URL, then click Test to verify the connection and Save.
Utter add custom model form with model ID and base URL
5

Set the Local Model as Default

In Default Models for AI models, select the newly created model (it has a key icon). This makes it the default for all AI post-processing.
Set the LM Studio model as the default AI model in Utter
What to expect: AI output uses the server you configured. For an entirely on-device Mac workflow, that server must run on the same Mac; a remote or LAN server is not on-device processing. Keep the server and its loaded model running. If that did not work: Click Test in the custom model form again, verify the local server URL and model ID, then save and reselect the model.

​
Ollama

If you already use Ollama, configure it as an OpenAI-compatible provider and follow the same Custom Models steps above. The key requirement is a local server URL and a model ID exposed by your Ollama setup.

Bring Your Own Keys

Connect your own model providers and API keys.

Use Utter for Free

Set up free AI models for Utter.

​
Still Need Help?

If setup still fails, contact support with your device model, OS and Utter versions, selected voice and AI models, and whether transcription or rewriting failed. Do not send API keys or private recordings.