More people than ever run AI on their own hardware, for privacy, for control, for cost. The apps for actually using those models have been an afterthought. Whelk is the company I am building to fix that. Whelk is a native client for iPhone, iPad, and Mac that talks to your own models, on-device with Apple Silicon or on your own servers over a VPN (like Tailscale). Any chat can be armed into a full agent work session in place, with live tools and approvals. Your models. Your agents. Your network. The trust model is the architecture, not a promise: no servers of mine, no accounts, no telemetry. Nothing leaves hardware you own. Background: I was CTO at Weedmaps through its IPO and ShopKeep through its $550M acquisition, and most recently at BuildOps. Private inference is the next platform. Whelk is the app I am building to win it. Free for individuals, with a Teams tier for organizations. I am here all day, including for the hard questions. Yes, it is closed source, and I am happy to explain why the trust model still holds without it.

Whelk: Private AI Platform

Your hardware. Your models.

Your hardware. Your models.