本地优先 AI 语音输入工具 | Local-first AI voice typing tool. 支持 Whisper/SenseVoice/Parakeet/Qwen asr，离线可用，LLM 智能润色. Open-source speech-to-text with LLM post-processing, works offline.
See the codeREADME
Voconly — Voice in. Text out. Done.
Speak naturally. Voconly transcribes locally, refines with AI, and puts finished text right at your cursor — in any app.
No cloud. No quota. No copy-paste.
Your Thoughts Move Faster Than Your Fingers
Speaking is natural. Typing is slow. Most ideas die in the gap between them.
Voconly closes that gap — speak freely, and let AI handle the cleanup.
Most Voice Tools Stop at Transcription. Voconly Doesn't.
Traditional voice tools turn speech into a transcript — then you copy, paste, and edit.
Voconly turns speech into finished writing.
Speak naturally ↓ Local transcription ↓ AI understands and refines ↓ Finished text appears at your cursor
No switching windows. No copy-paste. It lands exactly where you're typing.
Two ways to finish, one key:
- Press to start → press again to stop: full pipeline, AI-polished output
- Press to start → double-press to stop: raw transcript, straight out — skip AI entirely
One Voice. Different Intentions.
The same voice can mean different things — a quick thought, a professional email, a translation, a meeting summary.
Assign each workflow its own hotkey and switch modes with a single keystroke.
| Mode | You say... | You get... |
|---|---|---|
| 📝 Light Polish | "I mean, I think this plan is actually pretty good, you know, it's just that the timeline is kind of a little too long, and maybe we could, like, make some adjustments and try to shorten it a bit." | "I think this plan is actually pretty good, it's just that the timeline is a little too long, and maybe we could make some adjustments and try to shorten it a bit." |
| 💼 Professional Polish | "I mean, I think this plan is actually pretty good, you know, it's just that the timeline is kind of a little too long, and maybe we could, like, make some adjustments and try to shorten it a bit." | "I think this plan is solid overall, but the current timeline is a bit too long. We could make some adjustments to shorten it. |
| " | ||
| 🌐 Translate | "I think the plan is pretty good overall, but the timeline is a bit too long. Maybe we could optimize it a little further." | "嗯，我觉得这个方案整体其实还不错，就是这个时间周期吧，感觉还是有点长，我们是不是可以再优化一下？" |
| 🗂️ Meeting Secretary | Meeting discussion | Structured summary with key points |
| 🛠️ Custom | Anything | Your own prompt, your own logic |
Each mode has its own hotkey and processing prompt, while your ASR model and LLM provider work across all modes.
Press the assigned key to start speaking, then press it again to finish.
Built for Everyday Use
⚡ See your words as you speak No matter which ASR model you use, Voconly gives you a live transcription experience. Text appears while you speak — and it's ready the moment you stop. No spinner. No waiting.
♾️ Dictate for hours, stay in the flow Speak for minutes or hours with a transcription experience built for long sessions.
📍 Works everywhere you can type Browsers, Word, Notion, VS Code, WeChat, your email client — if you can type there, Voconly works there.
Why Voconly?
| Voconly | Traditional dictation workflow | |
|---|---|---|
| Privacy | Audio never leaves your device | Often uploaded to the cloud |
| Limits | Unlimited. No quotas. | Free quotas or subscriptions |
| Output | Finished, polished text | Raw transcript |
| Workflow | Appears at your cursor | Copy → paste → edit |
| Long sessions | Stable for hours | Accuracy may degrade over time |
| Models | Choose the ones you prefer | Locked into one provider |
Use the Models You Prefer
You're not locked into a single AI stack. Choose between fully local workflows or connect your preferred cloud models.
- ASR: Whisper · SenseVoice · Parakeet · Qwen-ASR
- LLM: Ollama (local) + major cloud API providers
Get Started in 60 Seconds
- Download Voconly → Releases
- Launch it — pick a recommended model
- Press hotkey and start talking — press again for polished text, double-press for raw transcript
That's it.
Build from source (Windows):
.\setup.ps1 # check & install dependencies
.\start-dev.ps1 # start dev server
No GPU / no Vulkan SDK? Use .\setup.ps1 -SkipVulkan (CPU mode). GPU recommended for best speed.
Platforms
| Platform | Status |
|---|---|
| Windows 10 / 11 | ✅ Available |
| macOS | 🚧 Coming soon |
| Linux | — Not planned |
Roadmap
Done
-
Local ASR: Whisper / SenseVoice / Parakeet / Qwen-ASR
-
Real-time transcription for every model
-
Long-session transcription without drift
-
LLM post-processing: quick note / professional / translation / meeting / custom
-
Single-key hotkey modes with optional AI pass
-
Windows app
Coming soon
- macOS
Why Open Source?
I built Voconly because I wanted a voice tool that felt truly mine:
-
No subscription quotas
-
No audio uploaded by default
-
No dependency on a single company
-
No black box I couldn't control
So I open-sourced it.
Voconly is also my experiment in what one person + AI can build today. If it helps you, a ⭐ means a lot.
Contributing
Issues, ideas, and PRs welcome — let's explore productivity in the AI era, together.
License
MIT License — Copyright (c) 2026 Xing Yong
About
Built by Xing Yong (老幸.AI) — exploring a simple question:
What can one person build with AI today?
Voconly is one of those experiments.
📧 laoxingai@139.com
Tech Stack
Tauri 2.0 · React + TypeScript · Rust · Whisper.cpp
Contributors
326 commits
Similar repos
Languages
C++
56.9%
C
19.6%
Cuda
7.5%
Rust
5.1%
TypeScript
2.9%
Metal
2.1%
GLSL
1.5%
WGSL
1.0%
CMake
1.0%
本地优先 AI 语音输入工具 | Local-first AI voice typing tool. 支持 Whisper/SenseVoice/Parakeet/Qwen asr，离线可用，LLM 智能润色. Open-source speech-to-text with LLM post-processing, works offline.
See the codeREADME
Voconly — Voice in. Text out. Done.
Speak naturally. Voconly transcribes locally, refines with AI, and puts finished text right at your cursor — in any app.
No cloud. No quota. No copy-paste.
Your Thoughts Move Faster Than Your Fingers
Speaking is natural. Typing is slow. Most ideas die in the gap between them.
Voconly closes that gap — speak freely, and let AI handle the cleanup.
Most Voice Tools Stop at Transcription. Voconly Doesn't.
Traditional voice tools turn speech into a transcript — then you copy, paste, and edit.
Voconly turns speech into finished writing.
Speak naturally ↓ Local transcription ↓ AI understands and refines ↓ Finished text appears at your cursor
No switching windows. No copy-paste. It lands exactly where you're typing.
Two ways to finish, one key:
- Press to start → press again to stop: full pipeline, AI-polished output
- Press to start → double-press to stop: raw transcript, straight out — skip AI entirely
One Voice. Different Intentions.
The same voice can mean different things — a quick thought, a professional email, a translation, a meeting summary.
Assign each workflow its own hotkey and switch modes with a single keystroke.
| Mode | You say... | You get... |
|---|---|---|
| 📝 Light Polish | "I mean, I think this plan is actually pretty good, you know, it's just that the timeline is kind of a little too long, and maybe we could, like, make some adjustments and try to shorten it a bit." | "I think this plan is actually pretty good, it's just that the timeline is a little too long, and maybe we could make some adjustments and try to shorten it a bit." |
| 💼 Professional Polish | "I mean, I think this plan is actually pretty good, you know, it's just that the timeline is kind of a little too long, and maybe we could, like, make some adjustments and try to shorten it a bit." | "I think this plan is solid overall, but the current timeline is a bit too long. We could make some adjustments to shorten it. |
| " | ||
| 🌐 Translate | "I think the plan is pretty good overall, but the timeline is a bit too long. Maybe we could optimize it a little further." | "嗯，我觉得这个方案整体其实还不错，就是这个时间周期吧，感觉还是有点长，我们是不是可以再优化一下？" |
| 🗂️ Meeting Secretary | Meeting discussion | Structured summary with key points |
| 🛠️ Custom | Anything | Your own prompt, your own logic |
Each mode has its own hotkey and processing prompt, while your ASR model and LLM provider work across all modes.
Press the assigned key to start speaking, then press it again to finish.
Built for Everyday Use
⚡ See your words as you speak No matter which ASR model you use, Voconly gives you a live transcription experience. Text appears while you speak — and it's ready the moment you stop. No spinner. No waiting.
♾️ Dictate for hours, stay in the flow Speak for minutes or hours with a transcription experience built for long sessions.
📍 Works everywhere you can type Browsers, Word, Notion, VS Code, WeChat, your email client — if you can type there, Voconly works there.
Why Voconly?
| Voconly | Traditional dictation workflow | |
|---|---|---|
| Privacy | Audio never leaves your device | Often uploaded to the cloud |
| Limits | Unlimited. No quotas. | Free quotas or subscriptions |
| Output | Finished, polished text | Raw transcript |
| Workflow | Appears at your cursor | Copy → paste → edit |
| Long sessions | Stable for hours | Accuracy may degrade over time |
| Models | Choose the ones you prefer | Locked into one provider |
Use the Models You Prefer
You're not locked into a single AI stack. Choose between fully local workflows or connect your preferred cloud models.
- ASR: Whisper · SenseVoice · Parakeet · Qwen-ASR
- LLM: Ollama (local) + major cloud API providers
Get Started in 60 Seconds
- Download Voconly → Releases
- Launch it — pick a recommended model
- Press hotkey and start talking — press again for polished text, double-press for raw transcript
That's it.
Build from source (Windows):
.\setup.ps1 # check & install dependencies
.\start-dev.ps1 # start dev server
No GPU / no Vulkan SDK? Use .\setup.ps1 -SkipVulkan (CPU mode). GPU recommended for best speed.
Platforms
| Platform | Status |
|---|---|
| Windows 10 / 11 | ✅ Available |
| macOS | 🚧 Coming soon |
| Linux | — Not planned |
Roadmap
Done
-
Local ASR: Whisper / SenseVoice / Parakeet / Qwen-ASR
-
Real-time transcription for every model
-
Long-session transcription without drift
-
LLM post-processing: quick note / professional / translation / meeting / custom
-
Single-key hotkey modes with optional AI pass
-
Windows app
Coming soon
- macOS
Why Open Source?
I built Voconly because I wanted a voice tool that felt truly mine:
-
No subscription quotas
-
No audio uploaded by default
-
No dependency on a single company
-
No black box I couldn't control
So I open-sourced it.
Voconly is also my experiment in what one person + AI can build today. If it helps you, a ⭐ means a lot.
Contributing
Issues, ideas, and PRs welcome — let's explore productivity in the AI era, together.
License
MIT License — Copyright (c) 2026 Xing Yong
About
Built by Xing Yong (老幸.AI) — exploring a simple question:
What can one person build with AI today?
Voconly is one of those experiments.
📧 laoxingai@139.com
Tech Stack
Tauri 2.0 · React + TypeScript · Rust · Whisper.cpp
Similar repos
Contributors
326 commits
Languages
C++
56.9%
C
19.6%
Cuda
7.5%
Rust
5.1%
TypeScript
2.9%
Metal
2.1%
GLSL
1.5%
WGSL
1.0%
CMake
1.0%