Provider-Agnostic AI: Gemini by Default, Switchable in One Line

Different tasks want different models. Triage can run on a cheap, fast model; a deep security review might warrant a stronger one.

Our AI layer exposes a small model registry: a logical task name resolves to a concrete model, and provider is chosen by model-id prefix. Adding Claude or OpenAI is a one-line change plus an API key.

Gemini is the default everywhere for cost, routed through the Cloudflare AI Gateway for logging and rate-limiting. Nothing else in the product needs to know which model ran.