Underdog has released Woof, a speech generation model that weighs just 32MB, and it’s showing it off with a browser game. According to TechCrunch AI, the model has 31 million parameters and is open for anyone to download. The company says it can run on most devices without an internet connection.
That size is the story here. While most AI companies chase bigger models that need massive processing power, Underdog is building for the hardware you already have.
5 things to know about Woof
1. The demo is a tiny retro room
The game puts you in a room with a typewriter, a CRT monitor, a music tuner, and a desk lamp that makes a sound when you click its switch. It’s a playful way to show what the model does, and all of it runs in the browser through Underdog’s model.
- Typewriter: type anything on your keyboard, and the game speaks it back.
- Music tuner: rotate it to generate different kinds of music.
2. The intro takes a shot at the cloud
The game opens by warning users about “intrusive” cloud providers in the town. It’s a jab at cloud-based models, and it tells you exactly how Underdog wants to be seen. This is a company selling independence from big AI servers as much as it’s selling a model.
3. Woof has a sibling that listens
The demo comes weeks after Underdog launched Bark, a 167MB speech-to-text model built for transcription and dictation on edge devices (hardware like your phone or laptop, not a remote data center). Woof talks and Bark listens. Put them together, and you’ve got the two voice pieces a local assistant needs, in about 200MB combined.
4. It started with a notetaker
Founder Sigil Wen told TechCrunch that he started building the app because he didn’t want to use notetakers like Granola. He wanted his own system, built on models that run on users’ devices and let them own their data.
His ambitions go well past meeting notes. “I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI,” he said.
“Within months” is a big promise, and it comes from a founder with every reason to believe it. But it’s a claim worth tracking, because it’s easy to test.
5. Underdog’s lineup goes beyond voice
The company has also released a 27-billion-parameter model that can run on laptops. That’s roughly 870 times Woof’s parameter count, so Underdog isn’t only playing at the tiny end. There’s also a waitlisted app, which the company is positioning as a locally hosted alternative to Muse and Instinct.
Availability and caveats
- Access: Woof is open to download, and the game runs in a browser.
- Pricing: TechCrunch’s report doesn’t mention any pricing for the model.
- Offline use: the claim that it works without internet on most devices comes from Underdog.
- Limits: TechCrunch notes that a model this small likely won’t replace bigger ones.
That last point matters. Nobody should expect 32MB to match a large cloud model. The more useful question is whether it’s good enough for specific jobs, and TechCrunch points to two: driving local assistant experiences and creating personalized podcasts.
Why it matters
Small models change the math for developers. If speech runs on the user’s device, your voice feature can keep working offline, and the user’s data can stay on their hardware. That’s a real selling point for anything touching private conversations, which is exactly where Wen started.
A charming browser game doesn’t prove any of that at scale, though. The real test comes when developers download Woof and try it in their own products. If it holds up, Underdog’s bet on private, on-device AI starts to look a lot less like an underdog story. You can find more details in TechCrunch AI’s original report.