Hey Product Hunt — I’m Pavel, founder of AnchorShell. I built Relay because AI infrastructure is getting more complicated than “send a prompt to one model.” Agents need to choose between models, stay inside provider limits, control cost, handle failures, and still make those decisions fast enough that the routing layer doesn’t become the bottleneck. Relay sits between your agents/apps and your models. It characterizes each request, makes a smart routing decision, then handles queue-first scheduling, RPM-aware request pacing, fallbacks, limits, guardrails, tokens, cost, and realtime tracing across hundreds of compatible model paths. One area we’re especially excited about is fast routing classification. Relay has an ultra-fast built-in classifier, and we’re also experimenting with pluggable System One models such as Laya, with Jev support planned. You can use the hosted version, or run the Community edition locally: npx @anchorshell/relay The Community version is self-hosted and source-available, and the GitHub repo is public. We’ve spent a lot of time trying to make Relay feel less like another proxy and more like an actual control layer for agents and teams, where you can make sure neither your agents or a team member accidentally breaks the bank. I’d especially love feedback on the routing model, queue-first approach, and anything you’d want to see before putting Relay in front of production agent traffic. Thanks for checking it out — and I’ll be here all day answering questions.

AnchorShell Relay

Fast, flexible smart AI routing for agents and teams

Fast, flexible smart AI routing for agents and teams