SymvibeDocs
  1. Docs
  2. Orchestration
  3. Voice orders

Talk to the board

Hold Space, say what you want, let go. Your words are transcribed on your Mac and handled like a typed order, and the orchestrator can answer out loud.

Updated October 22, 20262 min read

On this page

Push to talk#

  1. Hold Space

    With the board in front and no console or text field holding the keyboard. You can also hold the mic button on the command bar, or tap it once to start and again to stop.

  2. Speak

    The bar shows that it is listening and writes your words as they come.

  3. Let go

    The order is sent. Esc cancels instead.

  • Space always talks to the command bar, never to an orchestrator card.
  • A recording stops at 60 seconds, and one cut off by that limit is not sent.
  • Turn Space off with SettingsVoiceHold Space to talk.
A spoken order, the planner's steps and a spoken reply.

Speech recognition on your Mac#

  • Speech is transcribed by Parakeet v3, a model that runs on your Mac. Symvibe downloads it once from Hugging Face; until it is ready, the bar says it is preparing.
  • Microphone audio never leaves your Mac.
  • Phrases speech models invent on silence are filtered out, and words that sound like a console's name are matched to that name.
  • SettingsVoiceSpeech language is Automatic by default, or one of 25 European languages. Pinning a language helps when results come back in the wrong alphabet.
  • Pick the microphone in SettingsVoiceInput device. If macOS blocks it, allow Symvibe in System SettingsPrivacy & SecurityMicrophone.

Spoken replies#

  • With SettingsVoiceConfirm commands out loud on, the default, the orchestrator answers with short lines: confirmations, questions, and who finished. It never reads an agent's reply aloud, only the agent's name.
  • Choose the voice in the same pane and press Test to hear it.
  • Replies and spoken lines come in English, Spanish, Portuguese, French, German or Italian. The planner answers other languages in English; the app's own short lines use whichever of the six it judges closest.
  • Each line is generated by Inworld through Symvibe's servers: the text goes out, the audio comes back. Lines are cached on your Mac, up to 100 MB per account, and a line unused for 30 days is removed; a cached line plays with no network and no credits. Cached audio has a Clear button.
  • Spoken lines use credits. When the allowance is used up, cached lines still play and the text still appears in the thread.

What leaves your Mac#

WhatWhere it goes
Microphone audioNowhere. It is transcribed on the Mac.
The transcribed order, recent thread lines and a snapshot of the boardSymvibe's planner service, which forwards it to Groq to run the model, when the order needs the planner
The text of a spoken lineSymvibe's servers, then Inworld, which returns the audio
The speech modelDownloaded once from Hugging Face

Data and privacy lists exactly what the board snapshot contains.

Keep going#

DictationType by voice without sending anything. Voice and typed commandsWhat to say, with examples. CreditsWhat spoken replies and the planner cost.

Something on this page does not match what you see in the app? Write to hello@symvibe.com or tell us in the Discord, and we will fix it.