openwhispr

Files (1)

• SKILL.md

SKILL.md (first lines)

# OpenWhispr

Real-time voice dictation using the [OpenWhispr](https://openwhispr.com) STT
engine, piped directly into the ox agent for hands-free coding input. Designed
for low-latency, privacy-first voice-to-command workflows.

## Installation

### Prerequisites

- ox platform installed (see `PLATFORM.md`)
- Microphone with audio input
- Python 3.10+ (for the OpenWhispr daemon)

### Step 1 — Install OpenWhispr daemon

```bash
# Via pip
pip install openwhispr-stt

# Or build from source
git clone https://github.com/openwhispr/openwhispr.git
cd openwhispr && python -m pip install -e .
```

### Step 2 — Configure ox voice backend

Set environment variables:

```bash
export VOICE_STT_BACKEND=whisper
export VOICE_MODEL_PATH=/path/to/whisper/model
export VOICE_LANGUAGE=en
export VOICE_LOCAL_ONLY=1
```

For OpenWhispr:

```bash
export OPENWHISPR_HOST=localhost
export OPENWHISPR_PORT=8799
```

### Step 3 — Register with ox

Add to `~/.config/ox/skills.json`:

```json
{
  "openwhispr": {
    "command": "npx @openwhispr/ox-dictation",
    "env": {
      "MODEL": "large-v3-turbo",
      "LANGUAGE": "en",
      "OUTPUT_MODE": "ox-stream"
    }
  }
}
```

### Step 4 — Start the daemon

```bash
openwhispr --start --port 8799 --model large-v3-turbo
```

Verify:

```bash
curl http://localhost:8799/health
# {"status": "ok"}
```

## Available Tools

| Tool | Description |
|---|---|
| `dictation_start()` | Begin listening for voice input |
| `dictation_stop()` | Stop recording and get transcript |
| `dictation_stream()` | Open a streaming transcription session |
| `dictation_status()` | Check current dictation state |
| `dictation_set_language(lang)` | Change STT language |
| `dictation_calibrate()` | Run microphone calibration |

## Usage

### Basic Dictation

```js
// Start listening
await tools.dictation_start();

// ... speak your command ...
// "create a React component named Header with a blue background"

// Stop and get the transcript
const { text } = await tools.dictation_stop();
// text = "create a React component named Header..."
```

### Streaming Mode

```js
// Open streaming session
const stream = await tools.dictation_stream();

// Stream partial results as they come in
for await (const chunk of stream) {
  if (chunk.transcript) {
    console.log("Partial:", chunk.transcript);
  }
  if (chunk.final) {
    console.log("Final:", chunk.final);
  }
}
```

### Hands-Free Coding

From SAI141003/ox, MIT.

More from SAI141003/ox