A native Wear OS client for chatting with Mistral AI models through the Mistral API. Built with Kotlin and Compose for Wear OS, designed for round displays !
Chat from your wrist with keyboard or on-device voice input, keep a persistent chat history, and get Markdown-formatted answers all in a UI built for the small round screen.
This is not an official Mistral project and it's not affiliated to them in any way. It does not wrap chat.mistral.ai; it talks directly to
api.mistral.ai using an API key that you create yourself at console.mistral.ai/api-keys.
Later, I will try to add support for people that don't want to use their API or that pay for the pro sub like me and would benefit from that.
| Home | Chat | Side panel |
|---|---|---|
 |
 |
 |
| Settings | Voice input | Information |
 |
 |
 |
- Chat: conversations with Mistral models, replies rendered with light Markdown (bold, italic, strikethrough, inline code, bullet lists), a thinking indicator while waiting, and haptics-friendly round-screen layout.
- Voice input: long-press the send button to dictate, tap again to stop. Transcription runs fully on-device with Vosk models (~30-50 MB) which are downloaded from within the app and never send your audio anywhere. The transcript lands in the message field for review
- Chat history: conversations persist on the watch across restarts. Reopen them from the side panel or the full history screen, long-press a chat to pin it or delete it, and wipe everything from Settings if needed.
- Side panel: swipe from the right edge for a new chat, settings, and recent conversations.
- API key storage: the key is stored encrypted on the device and never leaves it (except obviously to talk to the Mistral servers).
- BYOK: create a free API key at console.mistral.ai/api-keys and paste it once and... you're set ! :D
Early development, but very very usable! :)
- Wear OS + Compose scaffold -> OK
- API key set up screen -> OK
- Safe local API storage -> OK
- Single conversation with non-streaming replies -> OK
- Conversation history -> OK
- Voice input -> OK
- Possibility to pick it as main assistant
- SSE streaming responses with haptic feedback
- Model picker
- Add support for other AI providers via API key (Anthropic, Google, OpenAI, Mammouth AI, OpenRouter,...)
- Add possibility to use a very small local LLM, that will run on-device exclusively
./gradlew assembleDebugRequires the Android SDK (API 33+, target API 36) and JDK 17+. Install it on a watch or emulator:
adb install app/build/outputs/apk/debug/app-debug.apkObviously GPL-3.0 :P