Source: https://harnsy.dev/features/dictation

> Speak into any text field of the harnsy dashboard: OpenAI or your own Whisper server recognizes it, the text lands at the cursor and is not sent on its own.

# Speak to the dashboard instead of typing.

Every text field has a microphone. harnsy sends your speech to the recognizer you choose and puts the text at your cursor. You edit it, then send it.

Dictation in the comment box of a work item: recording, with the level and the time. Sample data.

### A microphone in the field you are in.

A microphone appears in the corner of the text field where your cursor is. Click it, or press Ctrl+Shift+Space (on a Mac, ⌘+Shift+Space works too).

### Record, then recognize.

While you record, a level meter and a timer show. Press again to recognize; Esc cancels.

### The text lands at the cursor.

The recognized text is put where your cursor is. It is not sent on its own: you edit it and send it yourself.

### A message that says what to do.

No microphone, no access to it, no provider yet, a key the provider rejected: each has its own message and says where to fix it.

### Recordings become text.

An audio or video file attached to a task becomes text with timestamps, through the same recognizer you chose for dictation; long files are done in the background. If your recognizer is an online service, the file goes to it. Plugins can ask for it too, with the right to transcribe. harnsy charges nothing for it, in any edition.

### You add the provider.

harnsy brings no recognizer of its own. Add one under System › Connections › Dictation: OpenAI or your own Whisper server. Until you do, no microphone shows in the fields.

### OpenAI, with your key.

Speech goes to OpenAI with your API key; OpenAI bills it per minute of audio.

### Your own Whisper.

A Whisper server with an OpenAI-compatible API (for example Speaches), on this machine or in your network. The speech stays there.

### A language for each provider.

Choose the recognition language: Auto, where the provider detects it, or a fixed one. A fixed language is more reliable on short phrases.

### Several providers, one default.

Add more than one and make one the default; the default is the one the microphone uses.

### The key stays on this machine.

The API key stays on this machine, and the audio is not kept: it is held in memory only while it is recognized.

### The browser’s rule for the microphone.

A browser gives a page the microphone only over https or on this machine (127.0.0.1, localhost). Open the dashboard that way to dictate.

The recognized sentence lands at the cursor, after what was already typed; nothing is sent until you send it. Sample data.

System › Connections › Dictation: two providers, one is the default. Sample data.

## Limits, honestly

- A recording stops by itself at 5 minutes and is then recognized. Your provider may set limits of its own.
- You need a provider: OpenAI with your API key, or a Whisper server of your own.
- With OpenAI your speech leaves this machine for OpenAI; with your own Whisper it stays where that server runs.
- The browser must be able to record audio and be allowed to use the microphone.

## Your first team is one prompt away.

Installing takes one sentence to your agent. To run your agents, harnsy drives a terminal: tmux if it is installed, otherwise its own terminal host (on Windows, always its own).

`Install harnsy following https://harnsy.dev/llms.txt`
