The expense report your AI proxy skipped

What it does CPA Usage Keeper is a bookkeeping layer for CLIProxyAPI. It consumes usage events from CPA’s Redis queue, writes them to SQLite, and serves a built-in web dashboard that breaks down request volume, token spend, cache hits, success rates, and per-model costs. It also syncs metadata—API keys, auth files, and provider configs—so the numbers stay aligned with whatever CPA is currently routing.

The interesting bit The project treats SQLite as a production analytics store rather than a toy, which keeps deployment weight minimal while still supporting filtering, pagination, export, hourly heatmaps, and quota refresh across multiple AI providers. It is essentially purpose-built glue code, but it is thorough glue: it even embeds into CPAMC as a plugin for unified management.

Key highlights

• Consumes CPA Redis usage events into SQLite with automatic metadata sync for keys, auth files, and providers.
• Built-in dashboard covers usage trends, cost estimation, model composition, API key breakdowns, and request-level event inspection.
• Supports time-range filtering, customizable columns, CSV export, and hourly heatmaps for forensic budget analysis.
• Optional password protection, SQLite backups, and subpath deployment for reverse-proxy setups.
• Ships as a single Go binary with Docker, Docker Compose, Homebrew, and systemd packaging.

Caveats

• Requires an existing CLIProxyAPI backend; it is not a standalone proxy and cannot generate usage data on its own.
• The README is mostly a deployment manual, so deeper architectural or behavioral details are sparse.

Verdict Worth a look if you already run CLIProxyAPI and need to answer “which key burned through the quota last Tuesday.” Skip it if you are not running CPA or need a general-purpose API gateway analytics tool.