Private AI translation, anywhere

Speak freely with an offline AI interpreter in your pocket.

Salut is an Android app for local translation, voice conversations, mirror mode, and AI chat. It runs inference on your device, so your words do not need to leave the phone.

On-device AI No cloud inference Speech + text
Offline AI
You

Where is the nearest train station?

Salut

Dove si trova la stazione ferroviaria piu vicina?

CPU/GPU aware Streaming output
183 language catalog entries
4 core modes: Translate, Mirror, Chat, Settings
0 cloud calls required for inference

What Salut can do

Built for real conversations, not demo prompts.

From quick text translation to face-to-face speech, Salut keeps the interaction fast, visual, and grounded in local AI.

Aa

Offline text translation

Type a phrase, choose languages, and translate without sending the text to a remote inference API.

Mic

Speech to translation

Record a voice message, transcribe it, detect language context, and turn it into translated text.

180

Mirror mode

A face-to-face translator view rotates the output for the person sitting across from you.

Chat

Local AI chat

Ask questions, continue a session, stream answers, and render markdown while staying on device.

Img

Multimodal input

Attach images, audio, documents, or capture a photo directly from chat when you need more context.

Tune

Session controls

Toggle thinking mode, adjust temperature and sampling behavior, and monitor model performance.

Privacy by architecture

Your conversations stay close.

Salut is designed around local inference. Cloudflare R2 is used for model distribution, but the translation and chat work happens on the Android device after a model is downloaded.

Local model runtime Background model downloads Crash-isolated model process Real-time system stats

AI engine

Powered by Gemma 4-class local models.

Salut uses Gemma 4-family LiteRT language models for on-device translation and chat experiences. The app focuses on practical private assistance: translate speech, reason about attachments, and keep conversations responsive on supported Android hardware.

Gemma 4 LiteRT On-device

Adaptive backend selection can use CPU, GPU, or compatible NPU paths depending on the model and device.

How it feels

Fast enough for travel. Flexible enough for work.

01

Pick languages

Use your favorites or detect the source language from audio.

02

Speak or type

Capture everyday phrases, longer questions, or chat context.

03

Share the answer

Read, show mirror output, or continue in the local AI chat.

Ready for Cloudflare Pages

A simple static landing page.

This site has no server-side runtime and no package install step. Point Cloudflare Pages at the web directory and publish it as static assets.

Build command: none Output directory: web Framework preset: none/static

Android first

Carry a private AI interpreter wherever the network gets thin.

Salut is built for Android devices capable of running local AI models. Download a model once, then translate and chat without relying on cloud inference.

Back to top