@arvoretech/pi-elevenlabs-stt

PI extension for push-to-talk speech-to-text using the ElevenLabs Scribe API

Packages

Package details

extension

Install @arvoretech/pi-elevenlabs-stt from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:@arvoretech/pi-elevenlabs-stt
Package
@arvoretech/pi-elevenlabs-stt
Version
1.1.0
Published
Jun 23, 2026
Downloads
399/mo · 57/wk
Author
joao.barros.arvore
License
MIT
Types
extension
Size
30.3 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./dist/index.js"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

@arvoretech/pi-elevenlabs-stt

PI extension for push-to-talk speech-to-text using the ElevenLabs Scribe API.

What it does

Registers a keyboard shortcut that toggles microphone recording. Press once to start recording from your default mic, press again to stop and transcribe. The resulting text is appended to the editor input.

  • First press: starts recording from the selected microphone via ffmpeg (shows 🎙 recording in the footer).
  • Second press: stops recording, sends the audio to ElevenLabs Scribe, and inserts the transcript at the cursor (⏳ transcribing…).

The transcript is inserted via the editor's paste pipeline (pasteToEditor), so it composes with custom editors such as pi-vim instead of overwriting the input buffer. With a modal editor, dictate while in insert mode — text is inserted at the cursor and the editor's mode/cursor state is preserved.

Commands

Command Description
/stt-devices List available microphone inputs and select one. The active device is marked with ✓; the system default is marked (default).
/stt-device-clear Reset back to the system default microphone.

The selected device is persisted in the session and restored on --resume.

Requirements

  • ffmpeg on PATH (used to capture microphone audio).
  • An ElevenLabs API key exported as ELEVENLABS_API_KEY.

The extension auto-detects the audio input backend:

Platform Backend
macOS avfoundation
Linux (PulseAudio / PipeWire) pulse
Linux (ALSA only) alsa

Configuration

Env var Default Description
ELEVENLABS_API_KEY — Required. ElevenLabs API key used for transcription.
ELEVENLABS_STT_SHORTCUT ctrl+alt+t (Linux/Windows), ctrl+alt+r (macOS) Keyboard shortcut that toggles recording. On macOS, alt is the Option key. Avoid super (Cmd) — terminals usually don't forward it to pi.

Some terminals and window managers reserve ctrl+alt+t / ctrl+cmd+t. If the shortcut doesn't reach pi, set ELEVENLABS_STT_SHORTCUT to another combination (e.g. alt+t, ctrl+alt+r).

Privacy

Recorded microphone audio is sent to the ElevenLabs API for transcription. Recordings are written to a temporary file and deleted immediately after transcription (and on session shutdown). Nothing else is stored.

Notes

  • Recordings shorter than 400ms are ignored.
  • Uses the scribe_v2 model with automatic language detection.
  • Requires interactive (TUI) mode for the shortcut and editor insertion.