In the room.
In context.
It hears the team’s discussion.
Address sparkie when you need it.
A voice
in the room.
Hands on
the work.
Meet sparkie
Explore the screen ↗
Your meeting’s fourth teammate.
Good conversations
deserve follow-through.
sparkie joins your Zoom call, follows the discussion, and takes on work in your project.
In the room.
Say the word.
Start the work.
“sparkie, turn that into an outline and save it in this project.”
Keep talking.
Keep making.
The agent works while the meeting continues. Ask for another edit as the plan evolves.
View transcript ↗
View completed tasks ↗
Read the report ↗From a conversation
to an actual file.
The recording is unavailable.
Open full-size report ↗
From the productA request.
A request.
A report.
Ready to share.
“Could you please make this a report? So that I can pass it to others?”
The saved report, opened in the workspace.
Screenshot from a local voice session.Imagine
your next meeting.
“sparkie, turn that decision into a product brief.”
Imagine: a brief to refine.
Example report ↗A voice up front.
An agent at work.
The conversation.
Zoom · Deepgram
Realtime · Gemini
The workbench.
Devin / Codex
Tools · Files · Commands
How it connects +
Separate Zoom participant tracks feed Deepgram transcription and Realtime semantic turn detection. Gemini checks whether the completed request addresses sparkie. GPT Realtime handles the voice; a separate Devin or Codex worker carries out delegated project tasks.
Each task starts with the transcript context available at delegation. Later decisions need an explicit update. Results return to the voice conversation; people review the actual output.
Leave with
work in hand.
Build with sparkieA working prototype.
macOS Zoom. English speech.