An on-device intelligent voice assistant designed to boost creation and recording efficiency. Built with Tauri 2 (Rust backend) + Vite (TypeScript frontend), it delivers smooth realtime STT (SenseVoice ONNX local inference) and powerful TTS (Rust libtorch + Python bridge + local/cloud models) while running entirely on your local device to protect privacy.
- Core capabilities: on-device realtime recording and recognition (cursor-follow injection supported), realtime translation, AI proofreading and summarization, text-to-speech and voice cloning
- Main stack: Tauri 2 (Rust), Vite (TypeScript), ONNX Runtime (Rust
ort), Python (PyTorch/Transformers) - Scenarios: meeting/interview notes, bilingual captions, speak-while-you-type (input injection), dubbing and voice cloning
- SenseVoice ONNX local inference with realtime and final results
- Three output modes: cursor-follow injection (recommended) / realtime output / final auto-enter
- Global hotkey
RightAltto start/stop recording anytime - Supported languages:
auto/zh/en/yue/ja/ko/nospeech - Mini Mode: floating realtime captions window at the bottom-right
Download the STT model: SenseVoiceSmall
modelscope download --model iic/SenseVoiceSmall --local_dir ./SenseVoiceSmall
- Built-in translation proxy with Ollama / OpenAI-compatible APIs
- Recommended local translation model:
ZimaBlueAI/HY-MT1.5-1.8(Ollama) - Bilingual captions in the same view with streaming output
- Strategies: translate only final results (saves API quota) or translate realtime results too
- AI proofreading ("Principal" feature): auto-polish recognized text with configurable concurrency (default 2, max 4)
- Smart summaries: scheduled meeting/recording summaries (recommended every 60 seconds), shown as a tree or queue
- Supports OpenAI / Ollama / Volcengine; recommended local model:
qwen3:4b
- Qwen3-TTS with three modes:
Base: high-quality synthesis without reference audioCustomVoice: clone a voice with 5-15 seconds of reference audioVoiceDesign: generate a new voice from text prompts (no reference audio)
- Index-TTS2 emotional vector control + prompt guidance for more expressive cloning
- Cloud API support:
- Volcengine: rich high-quality voices, supports dialects/foreign languages
- OpenAI TTS: alloy / echo / fable / onyx / nova / shimmer
- MiniMax: high-expressiveness TTS models
- TTS runtime can be packaged in the full installer or downloaded on demand via Lite packages
- System tray: close the window to minimize to tray, restore or exit from the tray menu
- Download and install the latest Release package, then double-click to launch
- Open settings and point Model Directory to the SenseVoice model folder (must include
model.onnxandtokens.json) - Press
RightAltto start recording
For model downloads and FAQs, see INSTALL.md.
cd /path/to/boundless-flow
pnpm install
pnpm run tauri:devFull environment setup (Rust/MSVC, Python/TTS, Lite packages, packaging outputs) is in INSTALL.md.
| Setting | Description | Recommendation |
|---|---|---|
| Model Directory | SenseVoice model folder (must include model.onnx, tokens.json) |
Point to the exact model directory |
| Frame Interval (ms) | Audio frame send frequency; lower is more realtime but higher CPU | 20ms |
| Language | auto/zh/en/yue/ja/ko/nospeech | auto |
| TextNorm | Text normalization | auto |
| Output Mode | Cursor-follow injection / realtime output / final auto-enter | Choose by scenario |
Auto-downgrades if a platform does not support a feature (e.g., cursor-follow injection may be unavailable on some platforms).
| Setting | Example |
|---|---|
| Translation API Base URL | http://localhost:11434/v1 (Ollama) / https://api.openai.com/v1 |
| Translation Model | ZimaBlueAI/HY-MT1.5-1.8 / gpt-4o-mini / translategemma |
| Translation API Key | Can be empty for Ollama |
| Streaming Output | Recommended (smoother for local models) |
| Setting | Description | Example |
|---|---|---|
| Enable proofreading & summarization (LLM) | Global toggle | Enabled |
| Summary Provider | OpenAI / Ollama / Volcengine | Ollama |
| Summary API Base URL | Service endpoint | http://localhost:11434/v1 |
| Summary Model | Model ID | qwen3:4b / doubao-seed-1-6 |
| Proofreading Concurrency | 1-4 | 2 |
| Summary Update Interval (s) | Frequency of summaries | 60 |
Qwen3-TTS:
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-Base --local_dir ./Qwen/Qwen3-TTS-12Hz-1.7B-Base
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice --local_dir ./Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign --local_dir ./Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesignIndex-TTS2:
modelscope download --model IndexTeam/IndexTTS-2 --local_dir ./IndexTeam/IndexTTS-2After downloading, set TTS Model Directory to the corresponding model folder in settings.
In settings, choose Volcengine TTS and fill in:
| Field | Description |
|---|---|
| AppId | Volcengine app identifier |
| Token | Access token |
| Cluster | Cluster identifier (e.g., volcano_tts) |
| VoiceType | Voice identifier |
Optional: UID, audio encoding, sample rate, speed/volume/pitch multipliers, emotion, and more.
The Boundless Flow frontend calls backend commands via Tauri invoke, including:
- Config:
get_app_config/set_app_config - Recording:
start_listening/stop_listening - Injection:
inject_text - Translation proxy:
translate_via_backend - TTS:
tts_generate/tts_read_audio_base64
Backend entrypoint: src-tauri/src/main.rs.
- Frontend:
src/(Vite) - Rust backend:
src-tauri/ - Python bridge (TTS):
src-tauri/python/ - Tauri config:
src-tauri/tauri.conf.json(plus platform-specific configs) - Docs:
docs/(detailed user guides in Chinese and English)
docs/index.html: documentation entry and core capabilitiesdocs/stt.html: realtime STT and model configurationdocs/translation.html: realtime translation flow and settingsdocs/proofreading-summary.html: AI proofreading and smart summariesdocs/tts-voice-cloning.html: speech synthesis and voice cloningdocs/appendix.html: model downloads and API configuration guidedocs/context-landing.html: design philosophy, quick start, and best practices landing page
English docs are located alongside as *-en.html.
Local checks before submitting:
cd /path/to/boundless-flow
pnpm run type-check
pnpm run build
pnpm run tauri:build
pnpm run tauri:bundleRecommended submission: a single goal, reproducible steps, screenshots/logs (especially for UI/audio issues).
Copyright 2026 ZimaBlueAI & WaytoAGI-dev. All rights reserved.