Omo Research
Research
Experiment 03VoiceOpen source

Voice control from the phone

Run a team of agents hands-free: catch up, post, and undo by voice.

When

Oct 2026

Theme

Voice

We

Built it

Status

Open source

Illustration of voice control from the phone
IllustrationGenerated with ChatGPT image generation

01 / What it was

A phone app for running an agent team by voice while you're away from the desk. Ask “what did I miss?” and hear a short spoken catch-up with the real messages on screen, or say “tell the team we ship Friday” and it posts straight away.

We built three versions: a voice-only one, a visual one that shows the messages, and a redesigned commander that also tracks everything you asked for.

The open-source voice app's landing page, with an example conversation
ScreenshotThe open-source voice app's landing page, with an example conversation

02 / How it worked

  • The realtime voice model is only the ears and mouth. It hands each request to an agent with a small set of narrow tools.
  • Every answer is paired with the question it belongs to, so a slow answer is never spoken in reply to a newer question.
  • Posts go out at once, and saying “undo” reverses the last action. Background talk and filler words are filtered out.
  • About 70 words are spoken, then “want more?”. The full answer stays on screen.

03 / What we saw

Most spoken requests were catch-ups (“any updates?”, “what's new?”), not instructions. Reading every post back for confirmation felt slow, so posts now go straight out with an undo. And a voice app and a text chat on top of the same agent drifted into giving different answers.

04 / Where it led

Voice is now how we check in, not how we work. The detailed work moved into a space you can walk around.