Voice Input

Peon AI supports speech-to-text via any Whisper-compatible transcription endpoint. Recording is started and stopped with the microphone button in the chat toolbar.

Setup

Open Window → Preferences → Peon AI → Voice Input.

Enable Enable Voice Input — the microphone button will appear in the chat toolbar once enabled.

Fields

Field	Description	Default
STT Model	Model name sent to the endpoint	voxtral-mini-latest
Transcription Endpoint	Path appended to the base URL	/v1/audio/transcriptions
Base URL	Override the transcription host. Leave empty to reuse the main provider URL.	(empty)
API Key	Override the API key for transcription. Leave empty to reuse the main provider API key.	(empty)
Language	BCP-47 code for better accuracy (e.g. en, de). Leave empty for auto-detect.	(empty)

Base URL and API Key are only needed when your voice provider differs from your chat provider

If you use OpenAI for chat and Mistral for transcription, set Base URL to https://api.mistral.ai and API Key to your Mistral key. If both are the same provider, leave both empty.

Provider Examples

OpenAI

Field	Value
STT Model	whisper-1
Base URL	(leave empty — reuses model url)

Mistral AI

Mistral exposes a Whisper-compatible transcription endpoint.

Field	Value
STT Model	whisper-large-v3 or voxtral-mini-latest
Base URL	https://api.mistral.ai

LM Studio (local)

Load a Whisper model in LM Studio and point voice input at your local server.

Field	Value
STT Model	whisper
Base URL	http://localhost:1234

Ollama (local)

Ollama doesn't support the whisper models - gemma4 works.

Field	Value
STT Model	gemma4:e2b
Base URL	http://localhost:11434

If you use gemma4:26b or higher for coding - reuse the model.

Usage

Click the microphone button to start recording — the button turns red while active. Click again to stop; the audio is sent to the transcription endpoint and the result is inserted into the chat input.

macOS Microphone Permissions

macOS does not automatically grant microphone access to Eclipse. When permission is denied, the system returns silent (all-zero) audio without any error, so recordings appear to succeed but produce no transcription.

If the microphone button records without producing any text, run the following three commands in a terminal, then restart Eclipse:

/usr/libexec/PlistBuddy -c \
  "Add :NSMicrophoneUsageDescription string 'Microphone access for audio recording'" \
  /Applications/Eclipse.app/Contents/Info.plist

codesign --force --deep --sign - /Applications/Eclipse.app

tccutil reset Microphone epp.package.committers

After restarting Eclipse, macOS will show the standard microphone permission dialog the next time you use voice input. Grant access and recording will work normally. (For STS - replace Eclipse.app with SpringToolsForEclipse.app)

What each command does

• PlistBuddy — adds the NSMicrophoneUsageDescription key to Eclipse's Info.plist, which macOS requires before it will show the permission dialog.
• codesign — re-signs the app bundle so macOS accepts the updated Info.plist.
• tccutil reset — clears any previously denied microphone entry for the Eclipse bundle so macOS prompts again.