What if your AI assistant never sent your words to anyone’s cloud? That’s the bet behind Underdog, a personal AI assistant from Sigil Wen’s Conway Research that runs entirely on your own computer. The invite-only beta opened this week for Mac and Windows PCs, and it’s the most privacy-forward idea to hit the personal assistant race yet.
The pitch is disarmingly simple. Today’s assistants work by sending every prompt to a data center, where a giant model does the thinking and sends back an answer. Underdog flips that. The model lives on your machine. Your messages, documents, and emails stay on your hardware. Nothing has to leave your control for the assistant to do its job.
The privacy architecture
Here’s what actually makes this different. Underdog is built on Husky, an inference engine Wen developed himself for running AI models on consumer hardware. The claim: it moves less data between a computer’s processor and graphics chip than other local engines. That matters, because that data transfer is the bottleneck for running models on ordinary laptops.
On top of Husky sits a 27-billion-parameter reasoning model, fine-tuned from Qwen3.8-27B. That’s much smaller than the frontier models in data centers, but Wen argues it’s capable enough for everyday assistant work: comparing products, helping with homework, researching a topic. He points to benchmark comparisons with Claude Opus 4.6 to show how far compact models have come — though benchmarks vary enough that any single comparison is a data point, not a verdict.
The security design goes further than most. Underdog encrypts the access keys for services like email that you authorize it to use. The logic is sound: a personal assistant becomes genuinely useful exactly when it can see your sensitive information, and that’s precisely when you least want it phoning home to a server farm. Keeping the processing local dissolves that tension.
The business model experiment
Now the genuinely novel part. Underdog will be free and ad-free at launch. No subscription, no advertising, no data to sell. So how does it make money? A small cut of purchases made through the assistant, processed via Stripe.
Think about what that implies. Cloud AI companies pay for every request their users make — which is why they charge subscriptions. Underdog doesn’t pay per request because your computer does the computing. That collapses the cost structure and makes a transaction-fee model possible: the assistant earns when it helps you buy something, and only then.
It’s clever, and it’s a little dangerous. A free assistant funded by purchase commissions has an incentive to steer you toward buying things. Wen’s bet is that transparency and the user’s ultimate control over spending will keep that incentive honest. The beta will be the proving ground — recommendation quality, payment permissions, and fee transparency are the three things to watch. If Underdog recommends the best product rather than the most profitable one, the model works. If not, the privacy story won’t save it.
Who’s behind it
Sigil Wen is 21 and already one of the more interesting figures in AI infrastructure. A Thiel Fellow, he spent his late teens in Silicon Valley’s AI scene — living in a hacker house with researcher Andrej Karpathy, crossing paths with the founders of Perplexity and Midjourney. He once got GPT-2 running on an Apple Watch, which is the kind of thing you do either for fun or because you’re constitutionally incapable of leaving hardware alone.
Andreessen Horowitz led Conway Research’s first funding round, announced October 2, though the terms weren’t disclosed. That’s a meaningful signal: a16z doesn’t back on-device inference as a novelty, it backs it as a thesis about where computing goes next — that enough of the AI stack can collapse onto the device to make the cloud round-trip optional. Underdog is the consumer test of that thesis.
Why small models are having a moment
Step back and Underdog is part of a pattern. The industry is simultaneously making its biggest models cheaper and its smallest models smarter. A 27-billion-parameter model running on a laptop would have been a research curiosity two years ago. Now it’s the engine of a shipping consumer product with a16z backing.
That’s the quiet revolution. The frontier isn’t only about bigger models in bigger data centers. It’s also about smaller models getting good enough to live on your laptop, where the privacy problem and the cost problem both shrink at once. Underdog is the first serious consumer attempt to build a whole company on that insight.
What to watch next
Invite-only betas are easy to make look good. The test is whether Husky’s performance holds on ordinary consumer laptops rather than developer machines, and whether the 27B model handles the messy, ambiguous requests real users make. Linux, iPhone, and Android versions are promised but not here yet — and mobile is where an on-device assistant would matter most, and where it’s hardest to run.
The most encouraging thing about Underdog isn’t the technology, though the technology is genuinely impressive. It’s the business model. A free, ad-free AI assistant that doesn’t sell your data and doesn’t charge a subscription is a version of the future worth rooting for — if only because nobody else is building it.
Be First to Comment