You're looking at the kit's plumbing: streaming SSE, rate limiting, and per-request cost metering. See everything it ships with.
This public demo runs without an account. Sign in to attribute usage to your user and see it on the dashboard.
Ask anything to see token-by-token streaming.
This chat is the real Shipwright kit code in production, not a mock. Every message streams from Claude over SSE, and when a response finishes, the under-the-hood panel prices the request: input tokens, output tokens, cache reads and writes, stored as integer micro-USD so the money math is exact.
Signed-out visitors share a hard-capped free tier on the fastest model — the same quota and model gating shown on the pricing page. Hit the cap and the API answers with a 402 instead of a surprise bill.
The patterns behind it — disconnect-safe streaming, per-request cost metering, prompt evals — are written up on the blog, and the whole stack ships as a one-time purchase; see the landing page for what is in the kit.