Underdog shows Woof, a 32MB on-device speech model, in a new demo

Why it matters

Underdog demoed Woof, a roughly 32MB speech model meant to run on-device for always-available voice without shipping audio to the cloud. TechCrunch’s Nov 6 write-up highlights the size claim and product framing more than published benchmarks. Edge speech that small matters for latency, privacy and offline use, but intelligibility and language coverage still need third-party tests. Pair with other Friday tooling drops (Vercel DeepSeek tools, Scope agent-database bench) as the practitioner layer of the hour.