Khoury College of Computer Sciences
CS 4550 / CS 5610 · Web Development · Fall 2026
Tailscale Aperture: model routing on Vercel AI Gateway and Sandbox
- AI
- Course
Instructor-curated digest, not original reporting. Each post paraphrases a public article and links to the original source.
On September 11, 2026, Eric Dodds and Susan Aziz described how Tailscale built Aperture, a customer-facing model router that applies tailnet identity to AI model access. Add someone to the private network and they can use approved models; remove them and access disappears. Co-founder David Carney framed Tailscale as a networking company that chose not to rebuild provider routing; Remy Guercio leads the Aperture product.
Aperture sits on Vercel AI Gateway — one API for hundreds of models, with cost and usage in responses and zero data retention options including per-request zeroDataRetention — plus Vercel Sandbox for agent execution. The post uses the lethal trifecta framing for agents that can read private data, act on it, and reach the public internet, and mitigates that with an isolated sandbox plus identity controls. There is no cost markup on AI Gateway: customers pay the same rates as going direct. Tailscale migrated its own internal AI traffic to Gateway with a switch and reported that nobody noticed, going from prototype to paying customers in months.
The model router runs on the same Vercel deploy path you already use in Chapter 1 (and the Node server work in Chapter 5).
Read original: How Tailscale built a customer-facing model router on AI Gateway — Vercel