Why it matters
A voice agent that runs in-browser via Wasm instead of round-tripping audio to a cloud STT/LLM/TTS pipeline is a direct latency and per-session cost story, and the guardrails talk speaks to the operational risk of agents that act without a checkpoint.
The tokenmaxxing angle
Compiling to Wasm to get cross-platform reach without committing to a hosted framework like Pipecat or LiveKit is worth tracking as a pattern for shifting inference-adjacent work off metered cloud infrastructure entirely.
From the organizers
The agenda names Shea Hawkins on ‘Building a Local-First, Open Source Voice Agent’ — Rust compiled to WebAssembly, contrasted against Pipecat and LiveKit — followed by Charles Faris on ‘Guardrails & Gates’, at Code Talent in Denver.