The idea
Make choosing an AI model part of the interface rather than hiding the choice behind a single assistant identity.
The connection
The product decision is transparency: model selection, response attribution and operating cost belong in the user experience. AI supplies the engines; the authored system decides how people encounter and compare them.
System design
The following structure makes the project concept inspectable and defines the implementation to demonstrate.
What would prove it
A useful demonstration runs the same clearly specified task through several available models, identifies each one correctly and shows how the interface handles a provider timeout or unavailable model.
The next complete milestone
A release-quality experience needs a verified provider matrix, predictable streaming cancellation, per-session usage limits and clear treatment of unavailable features. Comparative output should be shown as examples, not a universal model ranking.
Project record
A Cloudflare Worker chat platform that puts a curated dropdown of AI models in front of the user instead of locking them to one vendor's flagship. Built to be fast, cheap to run at the edge, and honest about which model you're actually talking to.
Source: the published APX project register, recovered 28 September 2026. This study explains the design direction and evaluation path; it is not a fresh audit of the product implementation.