Spitball
Rapidly co-create lofi prototypes in ephemeral rooms with realtime shared canvases, built-in voice chat, and local voice and transcription AI.
YouTube Video
Project Description
Jon Berryman
Spitball: A Multiplayer lofi-prototyping and idea app with built in voice chat and a voice ai agent. Models for speech and transcription are local. AI can create and help you organize your thoughts. Users have realtime canvas access, both personal and shared spaces. Rooms are made to be easy to get in and do work. Everything is ephemeral. We use Kokoro to generate our agent’s voice so that he (Sal) can communicate easily with participants. Sal is an active listener, able to create an modify elements on a shared realtime canvas, and can respond to any questions.
The application is written in pythong and typescript with react for the frontend. we used pydantic ai agents and gpt-5-mini for our agent model with openrouter in case we change our mind.
We considered ourselves the users of the application, so not a ton of outside research was done, although we did consult with product people.
We have most basic functionality working, users can quickly ideate, organize thoughts, and build prototype screens, and so can sal. We need more sophisticated controls, and the ability to connect screens together.
Our loom video didn’t record the desktop output so you can’t hear sal in the demo video, but feel free to stop by to hear it all actually work, or just pick us to demo. :). https://www.loom.com/share/0d3ca05fb57948eab2aad13e6ecb7bbf
Prior Work
nothing existed before this weekend, we started completely from scratch