A fully local, privacy-first AI assistant for Android. No cloud, no subscriptions, no data leaving the device — and when one phone isn't enough, your phones team up automatically.
Every mainstream AI assistant — ChatGPT, Gemini, Copilot — sends your conversation to a server, processes it there, and sends a response back. Your words leave your device on every single message. Muse runs a full language model on the phone itself, so nothing ever travels over the internet.
Conversations stay on the device. No API keys, no accounts, no telemetry on your chats. Works in airplane mode.
Streaming token generation powered by llama.cpp, optimised with ARM NEON / fp16 / dotprod — same engine as desktop LLM runners, on your phone.
No model loaded? Another phone on the same Wi-Fi can answer for you — automatically discovered, automatically routed. No setup required.
Muse remembers facts you tell it across sessions. It extracts, stores, and injects context automatically — like a note-taking app that's also the AI.
Speech-to-text (Android SpeechRecognizer) and text-to-speech (Piper TTS) both run on-device. Voice chats work with no connection.
Gemma 4 vision support built in. Send a photo and ask a question — image encoding and token generation both run locally on the phone.
Muse can read incoming notifications and auto-reply — entirely on-device. The AI drafts the reply, you approve it.
"Remind me in 2 hours" or "add milk to my shopping list" — Muse parses intent and creates alarms, timers, and tasks without any third-party app or internet.
Full UI localisation in English, French, Arabic (RTL), Russian, and Hindi. Language is switchable at runtime from settings.
This is the feature that has no equivalent in any other mobile AI app. When multiple phones are on the same Wi-Fi network, they discover each other automatically and share inference work — no pairing, no QR codes, no configuration.
/capability endpoint. The Orchestrator picks the fastest matching peer.
Any phone actively chatting. Always on. Discovers peers, dispatches jobs, handles fallback. No toggle needed — it's automatic.
A phone that opts in to share its GPU/CPU. One toggle in Settings. Registers on the LAN, accepts inference jobs from any Orchestrator peer.
Exposes the full Muse chat interface over TCP port 8765 — for browser access or remote Lucy-to-Lucy chat. Separate from the mesh.
There are excellent desktop LLM runners and cloud AI apps. There is nothing else that combines all of these capabilities on Android.
| Capability | Muse | ChatGPT / Gemini | Ollama (desktop) | Jan (desktop) | PocketPal AI |
|---|---|---|---|---|---|
| Runs on Android phone | ✓ Yes | ✓ Yes (cloud) | ✗ Desktop only | ✗ Desktop only | ✓ Yes |
| Fully offline inference | ✓ Yes | ✗ Requires internet | ✓ Yes | ✓ Yes | ✓ Yes |
| Zero data sent to cloud | ✓ Yes | ✗ All data sent | ✓ Yes | ✓ Yes | ✓ Yes |
| Cross-device mesh inference | ✓ Yes (auto) | ✗ No | ✗ No | ✗ No | ✗ No |
| Persistent memory across sessions | ✓ Yes | ~ Paid tier only | ✗ No | ✗ No | ✗ No |
| Offline voice mode (STT + TTS) | ✓ Yes | ~ Cloud TTS | ✗ No | ✗ No | ✗ No |
| Vision / image understanding | ✓ Yes (on-device) | ✓ Yes (cloud) | ✓ Yes (desktop) | ✓ Yes (desktop) | ✗ No |
| On-device reminders / tasks | ✓ Yes | ~ Cloud-dependent | ✗ No | ✗ No | ✗ No |
| Notification AI auto-reply | ✓ Yes | ✗ No | ✗ No | ✗ No | ✗ No |
| No subscription / no account needed | ✓ Yes | ✗ Account required | ✓ Yes | ✓ Yes | ✓ Yes |
| Model download + management in-app | ✓ Yes | N/A | ✓ Yes | ✓ Yes | ✓ Yes |
Muse doesn't reinvent inference. It wraps the same engine that powers desktop LLM runners — with Android-native integration built on top.
These are not marketing claims — they are architectural facts. There is no server to send data to because there is no server.