Moss Live Lab: Ship Voice Agents in Under an Hour
This is the second event in the Moss Live Labs series. Last time we built semantic retrieval from scratch. This time, we're going after voice.
Voice agents live or die on latency. Every 300ms of lag between "user finishes speaking" and "agent starts responding" is the difference between a conversation that feels human and one that feels broken.
Moss runs retrieval in-process. Sub-10ms. That's the difference between a voice agent that feels human and one that doesn't.
Saturday, May 2nd. One hour, live.
Here's the plan:
Build a basic voice agent from scratch. STT, LLM, TTS wired together, end to end.
Rebuild it with Moss and watch the latency drop.
Turn it into an Inpersona: a voice agent loaded with your context that can talk to people on your behalf.
Community demos from builders who brought their own.
We'll build voice agents with Moss, starting simple and layering on complexity as we go. Sit back and watch us build, or follow along in your own editor. Up to you.
Bonus: bring your own voice agent. Open a PR on usemoss/moss with your voice agent code, and you can demo it live to the Moss core team. You'll get 10 minutes to walk through your build, show it off, and get feedback or answers to any questions. Max 2 demo slots, first come first served via PR. Everyone who demos gets 6 months of Moss paid tier, free.
PR instructions: https://github.com/usemoss/moss/issues/184
Open to all levels. If you're building voice agents, or want to, this is the hour to spend.