Open-source · On-device · Yours

The assistant that never phones home.

Ari hears you, understands you, and answers — right on your device. Open-source, extensible, and it speaks more than one language. No account. No data broker. No catch.

Pre-release · Android first, desktop & more to come
Runs offline
No telemetry
100% open-source
See it in action

Watch Ari do its thing.

A quick look at Ari listening, thinking, and answering — all on the phone, no cloud round-trip. Yes, it really is this quick.

Ari for Android
Looping demo
Video coming soon
Ari for Linux

A proper Linux desktop app is on the workbench — libadwaita, built for GNOME. Same engine, same skills, same "stays on your machine" promise.

Coming soon
Privacy

Your data stays yours. Here's how.

Most assistants ship your voice to a data centre and ask you to trust the policy. Ari mostly doesn't need to — and when something does use the internet, it says so.

On your device

The thinking happens here

The listening, the transcribing, and the working-out-what-you-meant all happen on your phone. The internet is optional.

Opt-in cloud

You decide what leaves

Want to plug in ChatGPT or Claude? Go for it — but that's your call, and every skill is upfront about when it uses the network.

Open source

Don't just take our word

Every line is out in the open. Read the code, check the claims, or fork the lot.

Read the whole privacy story →

Features

Everything a good assistant should be. Nothing it shouldn't.

“Hey Ari” wake word

Say the magic words and Ari wakes up — no button, no fuss. The listening runs on-device, so there's no always-on mic piping your kitchen chatter off to the cloud.

Skills & assistants

Weather, timers, reminders, smart home, music, directions — all installable from an open, signed registry. Want a big cloud brain too? Add one. Or don't.

Speaks your language

English and Italian today, and adding a language is translation work, not a rewrite. Ari was built multilingual from day one, not bolted on later.

A brain that works offline

There's a small language model living on your device for the open-ended questions. No signal? Ari still has answers. Prefer a cloud model? Plug it in — your call.

Talk or type

Voice-first, but never voice-only. Say it, dictate it, or just type it — and answers come back as cards you can tap or talk to.

Yours to extend

Fancy building a skill? The simple kind takes about fifteen minutes and zero build tools. Need real logic? Ship a sandboxed WASM one instead.

Get Ari

Not out yet — but close.

Ari's in active pre-release. Star the repo to follow along, and you'll be first to know the moment the Android build lands.

Star on GitHubF-Droid & Google Play — coming soon