Set Up VAssist
The setup wizard runs on first launch and starts with a welcome screen, followed by Live Assistant mode, AI provider, text-to-speech, speech-to-text, and extra tools. Android and desktop also have a background step after Live Assistant. Local engine and model downloads can take additional time.
Android wallpaper
If you're on Android and plan to use the live wallpaper, do setup in the main app first. The wallpaper reads its settings from the app.
Overview
| Step | What you configure |
|---|---|
| Welcome | Intro and docs link. Chrome builds show Chrome AI guidance here. |
| Live Assistant | Avatar on/off, display mode, position (browser/extension installs) |
| Background | Android and desktop only. Enable a bundled wallpaper and choose which ones to include. |
| AI Configuration | Your main AI provider |
| Text-to-Speech | Choose and configure spoken replies |
| Speech-to-Text | Choose and configure microphone input |
| AI+ Features | Optional tools: translate, summarize, rewrite, write |
1. Welcome
Just an intro screen. Links to the docs and, on Chrome, shows any Chrome AI setup guidance that applies to your browser. Hit Next.
2. Live Assistant
Pick whether you want the 3D Live Assistant or a chat-first layout.
- Avatar on: choose Standard Mode (full body) or Portrait Mode (upper body crop). Browser and extension installs also let you pick a starting screen position.
- Avatar off: VAssist runs as a chat window, no 3D rendering.
You can change all of this later in Settings → 3D.
3. Background
Blue Milky Way is selected and enabled by default. Choose other bundled wallpapers to include them in the fade slideshow, or turn the background off. If the avatar is on, you can turn on Stages and choose a stage. A fresh setup starts without a built-in stage selected. Stage options include lighting and local time, season and seasonal effects, environment animation, brightness, parallax, avatar shadows, and avatar alignment where supported. Turning stages off returns to your saved background. See the built-in background previews and stage previews. You can change these later in Settings. Desktop floating mode stays clear. In Android AR, the camera view temporarily hides both the selected background and stage. They return when you exit AR.
The setup step shares the background controls from Settings. Adjust brightness directly or open More image effects to choose a preset and tune contrast, saturation, blur, grayscale, sepia, or hue.
Choose Images, Solid color, or Gradient for the background. Gradient presets include Midnight, Ocean, Aurora, Sunset, and Forest. You can customize both colors and the angle.
On Android, the same selected wallpapers appear in the full app and live wallpaper. You can import your own images later in Settings. On desktop, you can combine the selected wallpapers with image files or a folder. See wallpaper credits for image sources and licensing.
4. AI Configuration
This is the most involved step. Pick one AI provider and finish only the fields it shows you.
Which provider should I start with?
| Platform | Default pick | Switch if |
|---|---|---|
| Android | Android Local | You already have OpenAI or Ollama set up |
| Desktop | Desktop Local | You'd rather use a cloud provider, or Ollama is already running |
| Browser / extension | Chrome AI (if ready) | Chrome AI isn't available yet, or you prefer OpenAI / Ollama |
What each provider asks you to do
| Provider | What the wizard shows |
|---|---|
| Android Local | Model download and selection, compute unit, model parameters, Test Connection, endpoint in Advanced Settings |
| Desktop Local | Engine and runtime setup, model downloads and selection, model parameters, Test Connection |
| Chrome AI | Status, required Chrome flags, Start Model Download, Refresh Status, Chrome internal links |
| OpenAI | API key field, Test Connection |
| Ollama | Endpoint, model name, Test Connection |
Android Local
The default on Android. Runs local GGUF or LiteRT-LM models on-device.
- Leave
Android Localselected. - Download or select a model in the model manager.
- Choose a compute unit if you want to change the default
Autoselection. - Adjust model parameters if needed, then use Test Connection.
- Continue when it responds.
The Advanced Settings endpoint field only matters if you've changed the default local address.
Desktop Local
The default on desktop. Uses llama.cpp through the Electron runtime.
- Leave
Desktop Localselected. - Keep the default
llama-serverengine, or choosenode-llama-cppin Inference Engine. - Install a llama-server runtime pack, or open Runtime Backend for node-llama-cpp and pick your backend:
auto: safest first pick, lets the runtime figure it outcpu: CPU only, slowest, but works everywherecuda: NVIDIA GPUsvulkan: AMD GPUs or other Vulkan-capable hardwaremetal: macOS (Apple Silicon and supported Intel Macs)
- If the runtime needs installing, hit Install and wait for it to finish.
- Download or import a GGUF model in the model manager and select it.
- Adjust model parameters if needed, then hit Test Connection.
- Continue when it responds.
TIP
auto is fine for a first pass. You can change the engine and backend later in Settings → LLM.
Advanced fields (Temperature, Max Tokens, Context Size, GPU Layers, Threads) can all stay at defaults during setup.
Chrome AI
Browser-native Gemini Nano. No API key needed, but it requires a recent Chrome build and specific flags.
- Keep
Chrome AIif the wizard shows it as ready or downloadable. - If the Required Chrome Flags block appears, enable these flags and restart Chrome:
Chrome AI required flags
chrome://flags/#optimization-guide-on-device-model→Enabled BypassPerfRequirementchrome://flags/#prompt-api-for-gemini-nano→Enabledchrome://flags/#prompt-api-for-gemini-nano-multimodal-input→Enabled
- After restarting Chrome, go to
chrome://componentsand let Chrome download Optimization Guide On Device Model if it's still missing. - Hit Start Model Download if the wizard offers it.
- If the download keeps running in the background, check
chrome://on-device-internals/and use Refresh Status in the wizard. - Continue when status shows Chrome AI is ready.
Chrome AI requires Chrome 138 or later. If it's still blocked and you need to move on, switch to OpenAI or Ollama.
OpenAI
- Switch to
OpenAI. - Paste your API key.
- Hit Test Connection.
- Continue when it succeeds.
Model selection and token limits are in Settings → LLM.
Ollama
- Switch to
Ollama. - Leave the endpoint as
http://localhost:11434if Ollama is on the same machine. - Update the model name if needed (replace
llama2with what you actually have installed). - Hit Test Connection.
- Continue when it succeeds.
Change the endpoint only if Ollama is running on a different machine.
5. Text-to-Speech
Choose how the assistant speaks. This is a separate step from Speech-to-Text, so you can configure either feature without the other.
TIP
Disabled is always a valid choice for either side. You can enable and configure voice later in Settings → TTS and Settings → STT.
Quick defaults by platform
| Platform | Default TTS provider |
|---|---|
| Android | Android Local |
| Desktop | Desktop Local |
| Browser / extension | Kokoro |
What each choice asks you to do
| Provider | What the wizard shows |
|---|---|
| Android Local TTS | Voice-pack downloads, active pack and speaker selectors, speed, Test TTS |
| Android Local STT | Model downloads, model, language and compute selectors, Test Recording |
| Desktop Local TTS | Engine picker, GPT-SoVITS install and reference voice form, or Supertonic options |
| Desktop Local STT | Engine picker, faster-whisper setup or whisper.cpp runtime and model downloads, model and language pickers |
| Kokoro TTS | Voice picker, speed, backend, Initialize Model, status, Test Voice |
| Chrome AI STT | Status, required flags, Start Model Download, output language, Refresh Status |
Spoken replies (TTS)
Android Local TTS
The wizard lets you manage Android voice packs directly during setup.
- Leave
Android Localselected to use the default VITS VCTK pack, or download another available pack. - Select the active pack and speaker, then adjust speed if needed.
- Use Test TTS to check speech before continuing.
You can change these choices later in Settings → TTS.
Desktop Local TTS
- Leave
Desktop Localselected. - Choose the TTS engine:
GPT-SoVITSsupports reference voice cloning. Continue with the installation and reference voice steps below.Supertonic 3uses its available voice, language, and quality controls. It does not need a GPT-SoVITS reference voice.
- For GPT-SoVITS, choose a PyTorch backend:
autofor the first passcpu,cuda,rocm,sycl, ormetalif you know your hardware
- Hit Install GPT-SoVITS (or Re-install if it's partially broken) and wait for the install log to finish.
- For GPT-SoVITS, add a reference voice:
- a name for the voice
- the exact words spoken in the audio sample
- the language of that sample
- a clip between 3 and 10 seconds (MP3, WAV, or M4A)
- Save it and keep it selected.
Install time
The GPT-SoVITS install pulls Python, PyTorch, the models, and dependencies. Expect around 5 GB of disk space and 10–30 minutes depending on your connection and whether you're using a GPU backend.
Kokoro TTS
Kokoro runs in the browser with no server needed.
- Leave
Kokoro TTSselected. - Pick a voice.
- Leave backend on
Auto. - Hit Initialize Model and wait for the download.
- Run the voice test.
Backend options: WebGPU is faster (~350 MB download). WASM is smaller (~86 MB) and more stable on weaker hardware. Switch to WASM if WebGPU gives you garbled or broken audio.
OpenAI TTS
- Switch to
OpenAI TTS. - Paste your API key.
- Pick a model and voice.
- Hit the test button.
OpenAI-Compatible TTS
For setups that already have a compatible speech server running.
- Enter the endpoint URL.
- API key if the server needs one.
- Model and voice name.
- Test it.
6. Speech-to-Text
Choose how the assistant transcribes microphone input. This is a separate step from Text-to-Speech. You can disable either one independently.
Quick defaults by platform
| Platform | Default STT provider |
|---|---|
| Android | Android Local |
| Desktop | Desktop Local |
| Browser / extension | Chrome AI, if ready |
What each choice asks you to do
Android Local STT
The wizard provides the Android model downloader and the same model, language, and compute choices available in Settings.
- Leave
Android Localselected to use the downloaded model automatically, or download and select a specific model. - Choose the transcription language and compute provider. Use a multilingual Whisper model or SenseVoice for Chinese or Japanese. English-only Whisper models ignore the language setting.
- Use Test Recording to check transcription before continuing.
NNAPI and XNNPACK may require a custom build. The experimental Snapdragon QNN option requires a compatible device and SenseVoice. You can change these choices later in Settings → STT.
Desktop Local STT
- Leave
Desktop Localselected. - Choose
faster-whisperor whisper.cpp in STT Engine. - For faster-whisper, initialize or verify its runtime. For whisper.cpp, install a runtime pack and download a GGML model.
- Pick a Whisper model:
tiny/tiny.en: faster, smaller, good enough for most usesbase/base.en: more accurate, larger download.envariants are English-only but slightly faster
- Hit Initialize Whisper (or Complete Whisper Setup) when using faster-whisper.
- Wait for the setup log when installing faster-whisper or a whisper.cpp runtime.
- Pick a transcription language (or leave on auto-detect), then use Test Recording (3s).
- Continue when the selected service is ready.
Chrome AI STT
- Keep
Chrome AIif status shows it as ready or downloadable. - Pick an output language.
- If flags are missing, enable them and restart Chrome:
Chrome AI STT required flags
chrome://flags/#optimization-guide-on-device-modelchrome://flags/#prompt-api-for-gemini-nano-multimodal-input
- Use
chrome://on-device-internals/plus Refresh Status if the download stalls in the background.
If Chrome AI is still blocked, switch to Disabled or OpenAI Whisper.
OpenAI Whisper
- Switch to
OpenAI Whisper. - Paste your API key.
- Pick the remote model.
- Continue.
OpenAI-Compatible STT
- Enter the endpoint URL.
- API key if needed.
- Model name.
- Language (or leave on auto-detect).
7. AI+ Features
Five optional tools that sit on top of your main AI provider:
- Translator: translate selected text
- Language Detector: identify what language something is
- Summarizer: shorten longer passages
- Text Rewriter: rephrase, adjust tone, fix grammar
- Content Writer: generate new text from a prompt
All of these can be toggled off for a simpler start and re-enabled later in Settings → AI+.
Done
Finishing setup saves your configuration and opens the main interface. If you ever want to redo these choices, go to Settings → UI → Start Setup Wizard Again.