Sigil Wen launches invite-only AI assistant that keeps data on-device
Sigil Wen, a Thiel Fellow and Silicon Valley AI figure, has launched Underdog, an invite-only AI assistant that operates entirely on users' devices rather than relying on cloud servers. This approach prioritises privacy by keeping user data local to their computers (initially Mac and Windows, with other platforms coming soon), addressing growing concerns about data collection and surveillance in AI services.
Underdog uses a 27-billion parameter model based on Qwen3.8 27B, which Wen claims performs comparably to Claude Opus 4.6 on certain benchmarks for everyday tasks like shopping research and homework help. Rather than charging subscriptions or selling advertising, Underdog's business model involves taking a small percentage of payment transactions made through the AI using Stripe's payment infrastructure, aligning the service's interests with users rather than advertisers or data brokers.
- AI assistant Underdog runs entirely on users' devices, not cloud servers
- Uses smaller but capable models; free with no advertising
- Revenue from tiny transaction fees, not data sales
New here? Start with this
Sigil Wen, a technology entrepreneur and former Thiel Fellow, has launched Underdog, a new artificial intelligence assistant designed to work differently from most AI services available today. Unlike popular AI tools that send your information to cloud servers operated by large companies, Underdog runs entirely on your own computer, meaning your conversations and data stay on your device rather than being uploaded elsewhere.
The shift to on-device processing reflects growing public concern about how AI companies collect and use personal data. Many people worry that established AI services monitor their activity and keep records of their conversations, which could potentially be used for advertising or sold to third parties. Underdog's design aims to address these privacy concerns by keeping users' information completely local.
Instead of charging monthly subscriptions or making money from advertising, Underdog generates revenue by taking a small commission from purchases users make through the service using Stripe's payment system. This business model means the company's interests align with users rather than advertisers or data collectors, since it only profits when users actively conduct transactions.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Advocates argue that on-device AI corrects surveillance capitalism by keeping sensitive data private and inaccessible to breaches or government demands. The transaction-based business model proves that privacy-respecting AI need not rely on advertising or data monetisation, potentially reshaping industry incentives away from user exploitation and towards genuine user alignment and autonomy.
The case against
Sceptics contend that on-device AI forces unacceptable compromises—the performance claims appear optimistic given inherent model limitations, and most users' privacy concerns can be addressed through cloud services with encryption and strong privacy commitments. The transaction model creates different perverse incentives, local devices lack computational efficiency for complex tasks, and without cloud updates, the AI quickly becomes outdated, making cloud-based solutions with proper safeguards the more practical choice.
Read the full article at the source →
Originally published by TechCrunch as “Silicon Valley’s AI wunderkind launches Underdog, the most private Instinct/Muse competitor yet”.