Hi Product Hunt 👋 I built TokenFin because [our team's AI bill kept growing and nobody could say which project, person or prompt was behind it]. Provider dashboards show one total. Claude Code on a subscription shows almost nothing. How it works: • Run `npx tokenfin setup` once. It uses the OpenTelemetry export your coding agent already has, so nothing sits in your request path and we never hold your API keys. • Every prompt is stored in full, with its cost, model, tokens and tool calls, grouped into sessions with directory, branch and start/end times. • Set per-model limits. You get an in-app alert and an email at 90%, with an optional automatic model switch. • Subscription usage (Claude Code, Cowork) is priced at API rates but kept apart from your real bill, so you see value without double-counting. • Waste finder, what-if simulator, chargeback and FOCUS exports, plus a read-only MCP server so you can ask about your spend from chat. It's free, unlimited and MIT-licensed. Codex and Gemini support is in beta, so feedback from people using them is especially welcome. What would you want to track next?

TokenFin

See what your AI coding agents really cost, per prompt

See what your AI coding agents really cost, per prompt