Open-source LLM price catalog whose differentiator is provenance, not breadth: every price carries its source, the vendor page that was read, and the date it was read, and prices two sources disagree about are marked rather than quietly resolved. The pipeline is the product; a scheduled GitHub Action re-checks roughly 80 models across a dozen providers every morning and merges sources under an explicit trust order, and a run that loses a source or would shrink the catalog publishes nothing and fails loudly. Most are read against a vendor's own page; the rest stay labeled feed-sourced rather than laundering an aggregator's number into a first-party claim. It answers in six places: the website, a keyless public API with an OpenAPI 3.1 description, an MCP server published to npm and verified in the official MCP registry, a VS Code extension on the Visual Studio Marketplace and Open VSX, and native iPhone and Android apps. Since September 2026 it also ships the Receipt (promptspend.com/receipt): a fully visible instruction you paste into any AI conversation, which estimates what the visible exchange cost at live rates, names the two biggest cost drivers, and says "pricing unavailable" rather than guess. Sep 2026: native iOS and Android apps, free on the App Store and Google Play. Paste a sample AI conversation to forecast its cost across up to four models; pasted text stays on the device. 1,200+ automated tests, 150+ of them in a real browser across four viewports, axe accessibility auditing at WCAG 2.1 A/AA. No accounts, no analytics, no cookies. Live at https://promptspend.com — repository: https://github.com/AndrewAvery7/promptspend