SpendLint puts AI model costs in the pull request

Field notes

The site models costs from declared token counts, request volume, model selection, and contract rules rather than inspecting provider accounts. Its worked example uses 1,000 requests per day and a 30-day month: replacing Claude Sonnet 5 with Claude Opus 5 raises the projection from $360 to $900, causing a 150% regression and three rule violations. Price rows include provider labels and verification dates, while the project explicitly distinguishes projections from invoices.