On-device + cloud · Android beta

Cloud or on-device.
Your call.

A real AI assistant for Android — flip one toggle to switch between blazing cloud voice and full on-device privacy. Same conversation, same memory, your choice every time.

Download APK See how it works
Free, no signup Android 8.0+ latest
RAI AI assistant illustration
v1.0
Latest build
~150ms
Cloud response
13
Languages
100%
On-device option
Features

One assistant. Two modes.
Both fully yours.

RAI is built around a single idea: you should be able to choose where your AI runs, and switch any time without losing your conversation.

Private Memory toggle

One toggle. Cloud or on-device.

Flip Private Memory ON and everything runs locally — the LLM, transcription, your memory. Flip it OFF and you get LiveKit + a real-time cloud model for sub-second voice. Your call, every session.

  • On-device LLM running entirely on your phone
  • Cloud mode: ~150ms real-time WebRTC voice
  • Switch mid-conversation — memory carries over
Private Memory
Choose where your AI runs
On-device

Full privacy. Local LLM + RAG.

Local ~2s first token
Cloud

Real-time WebRTC voice.

Streaming ~150ms latency
Real-time voice

Sub-second voice that actually flows.

In cloud mode, RAI streams audio over WebRTC for ~150ms response time. You can interrupt mid-sentence and the AI rolls with it — it feels like talking to a human, not a chatbot.

  • Full-duplex audio with interrupt support
  • Echo cancellation, noise suppression on-device
  • Auto speakerphone routing — hands-free anywhere
You What's the weather looking like tomorrow?
RAI · Tomorrow's looking partly cloudy, high of 72. You'll want a light jacket in the morning.
You Actually wait — what about Sunday?
Responded in 152ms
On-device memory

Remembers what matters. Forgets nothing.

A local vector database with 384-dimensional embeddings lets RAI recall relevant context from past conversations — semantically, not just by keyword. The memory lives on your phone and travels with you via encrypted backup.

  • 384-D vector DB (SQLite-vec via native C++)
  • Semantic search across every past conversation
  • AES-256 portable backup — phone-to-phone
Recalled from memory 4 results
0.92
Maya's flight lands at SFO Saturday 3:40 PM
last Tuesday · conversation
0.86
She mentioned wanting to try the Mission place
3 weeks ago · text recap
0.81
"Don't forget she's vegetarian"
last month · note
0.74
Walked through restaurant options together
2 weeks ago · summary
Context-aware

Reads the room. Reads your phone.

RAI knows whether you're walking, driving, or sitting still. It sees your battery, your network state, your storage. It adapts: shorter answers when you're moving, ultra-concise below 5% battery, graceful degradation on bad signal.

  • Activity Recognition: walking · driving · stationary
  • Battery, network, storage all influence responses
  • Auto-switches to local mode when signal drops
4% battery
Walking
WiFi
RAI · adapted Keeping it brief — low battery. Maya lands 3:40 Saturday. Mission place is reserved.
Tools

An AI that actually does things.

RAI doesn't just answer — it reaches into your phone. Search contacts with fuzzy matching, pull SMS history, check Gmail, query system info. Real actions, with your permission, from one conversation.

  • Contact search with fuzzy typo tolerance
  • SMS retrieval, Gmail integration, system queries
  • Each tool runs with explicit Android permissions
Find Maya in contacts
Matched: Maya Hong (0.94 score)
Done
Search SMS · "flight"
3 messages · Maya, Aug 22 · Aug 19
Done
Check Gmail · flight itinerary
Querying inbox...
Running
Privacy

Your assistant. Your phone. Your data.

Every conversation stays on your device. The database is encrypted at rest. We don't ship telemetry, don't log audio, don't sell data. The only thing that ever leaves your phone is what you explicitly send to a cloud model.

  • SQLCipher database encryption at rest
  • Audio never written permanently — recordings discarded after each turn
  • Zero telemetry. AES-256 encrypted optional backups.
Conversations Local only
Database SQLCipher
Voice recordings Never stored
Telemetry Zero
Platforms

Built for Android.
More on the way.

RAI runs on Android because that's where local AI got viable first. iOS and a web companion are next.

Android

Beta available now. Android 8.0 and up. Direct APK install, no Play Store required.

iOS

SwiftUI build in progress. Same cloud-or-on-device story, same memory format.

Web

Browser companion in progress. Pick up conversations from your laptop, sync with your phone.

FAQ

Questions, answered.

Cloud, local, privacy — here's what people ask most.

What's the difference between cloud and on-device mode?+
Cloud mode streams audio over WebRTC to a real-time model — you get ~150ms voice response and can interrupt mid-sentence. On-device mode runs everything locally: a tuned foundation model for the LLM, on-device transcription, and a local vector database for memory. The toggle in Settings is called "Private Memory" and you can flip it any time.
Does the on-device LLM actually work on my phone?+
Yes — on modern Android devices the local LLM handles full conversations, summarization, and tool routing without a network connection. First-token latency is around 2 seconds, then tokens stream from there. The on-device AI space is moving fast and we continuously evaluate and ship the best local model available.
What tools can RAI actually use on my phone?+
Contacts (fuzzy search with typo tolerance), SMS (read-only retrieval with fuzzy contact matching), Gmail integration, and Android system info (battery, network, motion state, storage). Each requires explicit Android permission and you can revoke any of them in Settings.
What happens to my voice recordings?+
Nothing — they're never stored permanently. In on-device mode, audio is transcribed and immediately discarded. In cloud mode, audio streams directly to the realtime model over an encrypted WebRTC channel and is never written to disk on either end of the connection.
Can I move my memory to a new phone?+
Yes. The memory database can be exported as an AES-256 encrypted backup with a password you choose. Restore on the new phone, enter the password, and your conversation history and vector memory are back exactly as they were.
Is RAI free?+
The app is free during beta. On-device mode is free forever — no subscription, no usage caps, no internet required. Cloud mode uses paid model APIs; we may introduce a Pro tier for heavy cloud usage in the future.
Changelog

Shipped recently.

We push updates often. Here's what's new.

v…
Loading changelog…

Cloud or on-device.
Your assistant. Your choice.

Free during beta, no signup, no telemetry. Install the APK and you're three taps from a real AI assistant.

Download for Android Also: try RAI Inbox