Most teams routing traffic through CPA / CLIProxyAPI quickly outgrow ephemeral dashboards: transient queues drop diagnostic context, provider failures scatter across accounts, and per-model billing semantics make cost attribution hard. This project treats CPA as the control plane—persisting queue events, exposing account health, and providing local analytics so operators can inspect failures, attribute cost, and automate credential recovery without sending data to third parties.
What Sets It Apart
- Persistent request and failure history in local SQLite with recoverable migrations — so you can diagnose post-mortem after CPA queue expiry rather than losing traces.
- Cost and token analytics broken down by model, provider, account, API key, project and time range — so teams can attribute spending and spot inefficient models or leaking credentials.
- Account inspection, quota automation and credential queues for Codex/Claude/xAI and other providers — so expired/blocked credentials can be detected, queued for action, and selectively restored by automation owners.
- Two deployment modes (lightweight panel replacing the upstream UI, or Full Mode Manager Server) and single-container Docker/native builds — so you can run a minimal UI or a persistent observability stack depending on needs.
- No telemetry, local files + encrypted management keys for backups — so sensitive account metadata and request summaries remain on-premise.
Who It's For and Trade-offs
Great fit if you run CPA/CLIProxyAPI in production and need durable observability, cost tracking, and automation around multiple LLM providers (OpenAI-compatible, Gemini, Codex, Claude, xAI, Vertex, etc.). It helps SREs, platform teams, and small ops teams that must keep credentials and billing signals local and auditable.
Look elsewhere if you only need a simple UI replacement (use the lightweight panel mode instead) or if you expect the project to proxy/forward model traffic itself — CPAMP does not act as a runtime proxy. It also assumes you operate authorized accounts and are comfortable running Docker/native services and managing local SQLite backups and encryption keys.
Where It Fits
Positioned between the raw CPA runtime and organizational monitoring: not a model-serving layer, but the management/observability layer for multi-provider LLM operations and billing governance.
