hsuehly/xiaoyu
Media & DesignAn open-source Flutter app for building and talking to customizable digital companions via real-time voice and video.
What it does
Xiaoyu is a Flutter app for iOS and Android that lets you create a digital companion and talk to it in real time. You pick or upload an avatar image, define a persona and a voice, and then start a full-screen voice/video call. The app connects to Vidu's S2-Avatar real-time service and supports voices from Qwen and Doubao, with quick-action buttons during a call.
The core flow is simple: create a character, wait for the avatar to be ready, then begin a session with billing confirmation and microphone access. Chat history, characters and messages are stored locally, and you can review call records and reconcile platform end states from the history view. Text chat is available with Qwen, while Qwen and Doubao both support categorized quick actions.
Getting started
You need Flutter 3.50.0 / Dart 3.14.0, plus a target device on Android 8.0+ or iOS 15+. Platform SDK and signing details are in the development guide.
git clone https://github.com/hsuehly/xiaoyu.git
cd xiaoyu
flutter pub get
flutter devices
flutter run -d DEVICE_ID
Replace DEVICE_ID with an ID from flutter devices. After launch, configure and check your Vidu API Key under Mine → Settings (top right) → Account Settings. Then create a character (choose an official avatar or upload a photo, fill in the persona, pick a voice) and wait until the avatar is ready. Tapping the character in "My Digital Human" starts a call after you confirm billing and allow microphone access.
Constraints
Real calls require Vidu real-time access permission and balance; charges and end states follow Vidu platform records, and a "pending reconciliation" status should be checked in the history. The Doubao domestic version supports voice and quick actions but not text chat or voice switching mid-call. The app must stay in the foreground during a call. The current 0.1.0+1 build is in internal testing; component mode, knowledge-base import, memory management and in-app voice cloning are not yet available.
The original code is licensed AGPL-3.0-only, so modification and distribution carry source-availability obligations. Brand assets and third-party materials are excluded from that license, and external dependencies follow their own terms.
HotRepo's take
The repository appeared on HotRepo's new-project list on 2026-11-20 with 12 stars gained over a roughly one-hour window (113 stars, 23 forks), which reflects a short burst of early attention rather than a sustained trend. There is no GitHub daily-rank record yet. What stands out is the concrete scope: private scripted characters that can be called, with explicit Vidu billing and foreground-only call limits rather than a vague "AI companion" pitch.
Written by AI from the project's README, updated . The project's own docs are the source of truth.