How Tailscale built a customer-facing model router on AI Gateway
Tailscale's Aperture ties AI model access to tailnet network identity, built on Vercel AI Gateway and Vercel Sandbox.
Tailscale built Aperture, a customer-facing model router, on top of Vercel AI Gateway and Vercel Sandbox. Aperture ties AI model access to a customer's tailnet identity: add someone to the network and they can use approved models immediately, remove them and access disappears.
Remy Guercio, who leads product for Aperture, compares a tailnet to "a VPC that can span any cloud, on-prem, your house, and your phone." Tailscale applied that idea to AI rather than building its own model routing and agent execution layers in-house. AI Gateway gives it one API for hundreds of models, and Sandbox gives agents an isolated place to run, which matters for agents that can read private data and reach the public internet, a risk Tailscale calls the "lethal trifecta."
Tailscale still maintains its own provider plumbing for a few customers who haven't migrated to Aperture yet, and says the gateway handles things providers themselves don't, like including the cost of a request in the response. Tailscale went from a model routing prototype to paying customers in months.
