Analysis updated 2026-07-03 · repo last pushed 2026-07-03
Monitor real-time request counts, success rates, latency, and cost across your AI gateway.
Diagnose failing Codex or Claude Code requests by viewing failure rates and redacted error summaries on the monitoring page.
Generate weekly AI spending reports with per-model cost rankings and token breakdowns.
Track multiple Codex accounts' quota limits, plan tiers, and credential status from one inspection page.
| seakee/cpa-manager-plus | cccyd2003-qwq/pinkbin | extend-hq/ui | |
|---|---|---|---|
| Stars | 1,199 | 1,181 | 1,255 |
| Language | TypeScript | TypeScript | TypeScript |
| Last pushed | 2026-07-03 | 2026-06-21 | 2026-07-03 |
| Maintenance | Active | Active | Active |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 3/5 | 2/5 |
| Audience | ops devops | developer | pm founder |
Figures from each repo's GitHub metadata at analysis time.
Requires an existing OpenAI-compatible AI gateway running alongside it, plus Docker or a native binary runtime.
CPA Manager Plus is a self-hosted dashboard that helps you monitor and manage your AI gateway traffic. If you route AI requests through a gateway like CLIProxyAPI or any OpenAI-compatible setup serving tools like Codex or Claude Code, this tool gives you a clear window into what is happening: how many requests are flowing through, what they cost, where failures occur, and whether your accounts are still healthy and within quota. The tool works by sitting alongside your existing gateway and recording every request that passes through. It shows a real-time dashboard with today's request counts, success rates, average latency, and estimated costs. You can filter by model, account, project, or time range. On the analytics side, it breaks down spend per model and per account, with token-level detail showing how much went to input, output, reasoning, and cached content. It also runs scheduled checks on Codex accounts to verify quota remaining, credential validity, and workspace status, and if an account hits its usage limit, it automatically pauses that account and re-enables it when the quota resets. This is built for teams or individuals who run an AI gateway and need visibility into cost and reliability. For example, if your Codex requests start failing, you can open the monitoring page to see the failure rate, which accounts or models are affected, and redacted summaries of what went wrong. If you need to report weekly AI spending, the analytics page ranks models by cost and breaks down tokens across categories. And if you manage multiple Codex accounts, the inspection page tells you each account's remaining quota, plan tier, reset schedule, and credential status so you know which accounts are usable. Notably, the project is designed with privacy in mind. It has no analytics SDKs, no cloud account requirement, and no registration flow. All data stays in local files on your host machine, gateway keys are encrypted before storage, and exported data never includes raw error bodies. It runs as a single Docker container or a native binary on Linux, macOS, and Windows, and only talks externally to the gateway you configure plus any optional integrations you explicitly enable, like model price syncing from LiteLLM or OpenRouter.
A self-hosted dashboard for monitoring AI gateway traffic, track request counts, costs, failures, and account quotas in real time, all stored locally with no cloud dependency.
Mainly TypeScript. The stack also includes TypeScript, Docker.
Active — commit in last 30 days (last push 2026-07-03).
No license information is provided in the repository explanation, so usage rights are unclear.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly ops devops.
This repo across BitVibe Labs
Verify against the repo before relying on details.