Tokenwise

Tokenwise is a one-line LLM proxy (OpenAI-compatible baseURL) for makers and small teams who can never tell where their API bills come from. Add one line of code, or point your coding agents (Claude Code, Cursor, Codex) at it with no production changes, and you see every request: exact re-tokenized cost, input/output tokens, latency, status, model, provider, project, and custom tags. It clusters requests by prompt template so each call type is its own line. Then it acts. It recommends the cut — a cheaper model, a cache, a trimmed prompt — and proves it on your own recent traffic with an LLM judge scored against your rubric, not a public benchmark. Apply in one click as an A/B split, watch quality for 24h, and it auto-rolls-back if quality slips. A live counter tracks the dollars saved. The closed loop — see, then act, then prove quality held — is what makes it more than another dashboard. Free to start, no card required.