CASE WORKSPACE

Metered AI services

A single authenticated gateway: tier gate → rate limit → budget reservation → hardened prompt → charge → log.

Controlled demonstration

Run an assessment

ModelCase material or question (treated as untrusted input)

Plan: Free

Bearer-authenticated server function; no provider key reaches the browser.
Tier gate rejects models outside your plan before any upstream call.
Rolling-window limit: 8 requests / 60s per service.
Untrusted text is fenced, control characters stripped, injection markers logged.
Charges are idempotent per request id and written to an append-only ledger.
Logs are redacted: no tokens, emails, IPs or raw prompts.