BingZi-233/check-cx

What it solves

Check CX addresses the lack of transparency and reliability when using AI model APIs, especially when routing through third-party relay providers. It prevents "fake" health checks where proxies return canned responses to appear online, and provides quantifiable evidence of latency and uptime to distinguish between provider outages and relay failures.

How it works

The system continuously probes AI APIs (such as OpenAI, Gemini, and Anthropic) using real streaming model calls. To ensure the API is actually functioning, it generates random language challenges (ranging across five difficulty tiers) that the model must solve correctly to be marked as "up." It measures the true time-to-first-token and ping latency, tracks uptime stats over 7, 15, and 30 days, and can sync with official provider status pages to correlate internal probe results with global outages.

Who it’s for

Developers and organizations that rely on multiple AI model APIs or third-party relays and need a public-facing or internal status board to monitor availability, performance, and model capability.

Highlights

• Random Challenge Validation: Uses dynamic language tasks to expose fake proxies that return static text.
• Real-Request Probing: Measures actual streaming performance and time-to-first-token rather than simple port connectivity.
• Capability Assessment: Samples high-difficulty challenges to assign an intelligence score to models without affecting their health status.
• Production-Ready Infrastructure: Features multi-node leader election to prevent duplicate polling, automatic retries, and a read-only status API for external integrations.
• Official Status Sync: Automatically polls OpenAI and Anthropic status pages for external verification.