yxq953 / MuseLite

# MuseLite
MuseLite is a Python-first AI workspace for Android. It combines an OpenAI-compatible chat client, persistent conversations, an execution sandbox, browser tools, scheduled prompts, and optional phone automation in one native Android application.
## Features
- Chat with any OpenAI-compatible HTTPS provider.
- Persistent conversations and model settings stored in SQLite.
- API keys encrypted with the Android Keystore.
- Scheduled prompts: type a prompt, choose a time, and let MuseLite send it in the selected conversation.
- Agent tool loops for multi-step work, with progress and stop controls.
- Reusable Skills with on-demand instructions, scripts, references, and templates; a built-in Skill Creator creates new skills through chat.
- A persistent Alpine Linux sandbox powered by PRoot for shell commands and workspace files.
- Browser tools for fetching pages, managing tabs, and saving useful results into the workspace.
- Optional Android phone control through the accessibility service and a foreground task service.
- Safe Markdown rendering for assistant messages.
- Android 11+ arm64 support, with a desktop Toga preview for development.
## Requirements
- Windows x86-64 for the provided build script.
- Python 3.13.
- Internet access on the first build to download Briefcase, the Android SDK, Gradle, and Maven dependencies.
- An Android 11+ arm64 device or emulator for the packaged application.
WSL, Linux, and administrator privileges are not required by `build_windows.ps1`.
## Build the APK on Windows
From the repository root, run:
 [code] 
The script creates a local `.venv`, downloads build tools into `cache`, stages the Android project, and writes a debug APK to `dist`. On the first run, accept the Android SDK licenses when prompted.
Install the newest APK on a connected device with:
 [code] 
Debug APKs are intended for development and testing. They are not release-signed packages for an app store.
## Configure a model
Open **Settings** after launching MuseLite and enter:
1. An OpenAI-compatible HTTPS base URL.
2. A model ID.
3. An API key.
For example, DeepSeek can use `https://api.deepseek.com` with the model ID `deepseek-flash`. The provider endpoint is validated before settings are saved. The API key is kept in the app's private storage and encrypted with Android Keystore.
## Schedule a prompt
1. Tap **Start New Conversation**.
2. Type the prompt you want MuseLite to send later.
3. Tap **Schedule** beside the input field.
4. Choose the hour and minute, then tap **Confirm**.
MuseLite returns to the main screen after confirmation. At the selected time, Android opens the saved conversation and sends the prompt that was in the input field when the task was created. An empty draft is not saved as a recent conversation.
## Use Agent phone control
Phone control is disabled by default. To enable it:
1. Open **Settings > Phone Control**.
2. Enable the MuseLite accessibility service in Android system settings.
3. Grant notification permission w