Silicon Valley’s New AI Star Unveils Underdog, a Highly Private On‑Device Assistant

Privacy‑First Design: How Underdog Keeps Your Thoughts Private

A 24‑year‑old coder from the Bay Area has announced Underdog, an on‑device AI assistant that rivals popular cloud‑based tools while keeping user data offline. The launch comes after Sigil Wen, who once shared a hacker house with Andrej Karpathy, built the software with backing from a roster of Silicon Valley heavyweights.

Wen began coding at 17 and moved to Silicon Valley to work in a shared AI lab. There he collaborated with future leaders of the industry, including Perplexity founder Aravind Srinivas and OpenAI researcher Noam Brown. While experimenting with early AI models, Wen discovered the privacy limitations of existing assistants. He spent months refining a lightweight, on‑device engine that can run on a standard laptop or smartphone without cloud access. Underdog promises fast responses, zero data upload, and the same conversational quality as mainstream competitors.

Is the Market Ready for a Fully Offline AI Assistant?

Underdog’s architecture is built around a local neural network that processes all inputs on the user’s device. The model is compressed to fit within 2 GB of storage, enabling it to run on mid‑range hardware. Wen explained that the system uses differential privacy techniques to avoid leaking sensitive information, even if the device is compromised. „We wanted an assistant that feels personal without the trade‑off of sending every query to a server,” he said. Early beta testers reported that the assistant could handle complex tasks—such as drafting emails, summarizing documents, and generating code snippets—without any noticeable lag.

The company’s funding round, led by a consortium of venture capitalists and AI pioneers, provided the capital to optimize the model for performance and security. Investors highlighted the growing demand for privacy‑oriented AI, especially as regulations tighten in Europe and the United States. Wen’s background in open‑source communities helped him attract talent skilled in efficient machine learning, ensuring that Underdog remains lightweight yet powerful.

The AI assistant market is dominated by cloud‑based services that rely on massive data centers. Critics argue that such models can never match the privacy guarantees of a local solution. Underdog’s entry may shift the conversation toward decentralised AI. Analysts note that the product could appeal to professionals who handle confidential data, such as lawyers, doctors, and government workers. However, the lack of continuous model updates could limit its ability to stay current with new knowledge, a challenge for any offline system.

What Happens Next for Underdog and the AI Landscape?

Wen acknowledges this trade‑off. „We plan to release periodic updates that users can download, but we’ll never store their data on our servers,” he said. The company’s roadmap includes support for multiple languages and the ability to integrate with existing productivity tools. If Underdog gains traction, it could force larger players to rethink their privacy policies.

Underdog is currently available for a limited beta on Windows, macOS, and Android. The team is working on an iOS version and a lightweight browser extension. If the product meets user expectations, it could spark a new wave of privacy‑focused AI assistants. The broader industry may respond by offering hybrid models that keep sensitive data local while leveraging cloud resources for heavy computation.

For Wen, the launch marks the culmination of years of experimentation and collaboration. „I’m excited to see how people use this tool in their daily lives,” he said. Whether Underdog becomes a mainstream alternative remains to be seen, but its emphasis on privacy and on‑device processing could redefine the way consumers interact with artificial intelligence.

Frequently Asked Questions

Can Underdog run on older smartphones? Yes. The model is designed to fit within 2 GB of storage and can operate on mid‑range devices, though performance may vary with hardware specifications.

What happens to my data if I uninstall the app? All local data, including conversation history, is deleted from the device upon uninstallation, ensuring no residual information remains.