Underdog released Woof as a text-to-speech model and described it as 31.7 MB—small enough to run in a browser. The model has 31 million parameters; Underdog says it can run on most devices without an internet connection and is available to download. A browser game demonstrates Woof by speaking text users type on a typewriter and generating different kinds of music when they rotate a tuner.
The demo’s room has a retro-tech theme, with a typewriter, CRT monitor and a desk lamp that makes a sound when clicked; its introduction warns about cloud providers. TechCrunch noted that a model this small is unlikely to replace larger models, but might be sufficient for local assistants or personalized podcasts. Weeks before the Woof demo, Underdog launched Bark, a separate 167 MB speech-to-text model designed for edge-device transcription and dictation.
Underdog founder Sigil Wen said he wanted to build an AI notetaking system that runs on users’ devices and lets them own their data, rather than relying on services such as Granola. Wen said he believed most capable models available at the time could run on existing devices within months, and described Underdog’s aim as training open, capable, American-made models for private AI. Underdog has also released an unnamed 27-billion-parameter model that runs on laptops and has a waitlisted app it positions as a locally hosted alternative to Muse and Instinct.