On-Device AI

Last updated: 2026-04-16

Run AI entirely on your phone — no internet required, no data sent to servers. On-Device AI gives you the full power of ListAIse's AI features with complete privacy and instant responses after a one-time setup.

100% Private

No data leaves device

Works Offline

No internet needed

Instant Results

After first warm-up

Voice Offline

Transcription on-device

What is On-Device AI?

ListAIse offers two AI modes: Cloud AI (default) and On-Device AI. Cloud AI sends your requests to remote servers for processing. On-Device AI runs the full AI model locally on your phone using Gemma 4 — Google's state-of-the-art on-device language model.

Cloud AI

No storage needed
Always up-to-date model
Requires internet
Data processed remotely

On-Device AI

100% private
Works fully offline
Instant after warm-up
Requires 3–4 GB storage

Which AI am I using right now?

Open any list, add an item by voice, and look at the small label below the item while it processes. It shows either "On-device (Gemma)" or "Cloud" to tell you exactly which AI is active.

How to Enable On-Device AI

Switching to On-Device AI takes less than a minute. The model downloads once and then stays on your device.

Open App Settings

Tap the profile icon in the top-right corner of the home screen, then select Settings.

Navigate to AI Provider

Scroll to the AI section and tap AI Provider. You'll see the current selection: Cloud (default) or On-Device.

Select On-Device AI

Tap On-Device (Gemma). ListAIse will prompt you to download the model if it's not already on your device. Confirm to begin the download (~3–4 GB).

Wait for Download

The model downloads in the background. You can continue using the app normally during this time. A progress indicator shows download status.

You're ready

Once downloaded, On-Device AI activates automatically. All subsequent AI requests process entirely on your device.

Wi-Fi recommended

The Gemma model is 3–4 GB. Download it over Wi-Fi to avoid using mobile data.

First-Run Warm-Up

The first time you use On-Device AI after launching the app, the model needs to load into memory. This is called the warm-up. It only happens once per app session.

First request may take up to 60 seconds

The warm-up is a one-time cost per session. After the model is loaded, every subsequent request is instant — faster than cloud AI because there's no network round-trip.

What you'll see during warm-up

Loading sheet appears

A modal sheet slides up showing "Preparing on-device AI…" with an indeterminate progress indicator. ListAIse is loading the Gemma model into memory.

Model ready

Once loaded, the sheet closes and your request processes immediately. The model stays warm for the rest of the session.

Proactive warm-up

ListAIse tries to warm up the model in the background when you open a list, before you even tap the microphone. This makes the first request feel instant in most cases.

On-Device Voice Transcription

On-Device AI includes a separate model for voice transcription: Whisper, running locally via a compact on-device implementation. This means your voice never leaves your phone — not even for transcription.

Whisper On-Device

The Whisper model handles speech-to-text entirely on your device. Once installed, voice input works with no internet connection — perfect for grocery stores with poor signal.

Setting up on-device voice

Tap the microphone icon

In any list, tap the microphone button to start voice input.

Install the voice model

If Whisper isn't installed yet, a sheet appears showing the model size and an Install button. Tap it to download.

Download completes

A progress bar shows the download. Once complete, voice input activates immediately. Future sessions use the already-installed model.

Model sizes

Whisper is available in multiple sizes (approximately 600 MB to 1.5 GB). Larger models offer better accuracy across accents and noisy environments. Choose based on your available storage and accuracy needs.

Privacy & Advantages

On-Device AI isn't just a privacy feature — it's a fundamentally better experience in the right context.

Complete Privacy

Your grocery lists, voice recordings, and recipe content never leave your device. No third-party servers process your data — ever.

Fully Offline

AI features work in supermarket basements, rural areas, parking garages, and anywhere with poor or no connectivity.

Zero Network Latency

After the initial warm-up, requests are processed instantly — no round-trip to a server, no waiting for a response over the network.

You're in Control

You decide which AI handles your data. Switch between Cloud and On-Device anytime from Settings. Your preference persists across sessions.

Device Requirements

On-Device AI runs large neural network models, so it requires available storage space and a modern device processor.

Model	Purpose	Size
Gemma 4	Recipe analysis, AI features	3.1 GB – 4.4 GB
Whisper	Voice transcription	~600 MB – 1.5 GB

Storage tip

You don't need to install both models. Download only what you use — Gemma for AI features, Whisper for voice input. Each can be enabled or disabled independently.

Fallback Behavior

If On-Device AI encounters an error (model loading failed, not enough memory, etc.), ListAIse never silently falls back to cloud AI. You always stay in control.

When local AI fails, you choose

Use cloud this time — process this single request via Cloud AI. Your AI provider setting doesn't change.
Cancel — abort the request entirely. No data is sent anywhere.

No silent data leaks

ListAIse will never automatically send your data to cloud AI without asking. The fallback sheet always appears, giving you the final say.

Best Practices

Get the most out of On-Device AI with these recommendations.

Use On-Device AI when…

• You're shopping in areas with poor connectivity
• You're privacy-conscious and don't want any data leaving your device
• You use voice input frequently and want offline transcription
• You want instant responses without network delays

Use Cloud AI when…

• Your device has limited storage (under 5 GB free)
• You're on a newer device that hasn't warmed up the model yet
• You want to conserve device RAM during heavy multitasking

First session tip

The model warm-up on first use can take up to 60 seconds. If you're in a hurry, open ListAIse a minute before you start shopping so the model loads in the background.