OpenAI Monitoring

Community MKP for monitoring OpenAI service health, API usage, costs and performance with Checkmk 2.5.

This package provides two independently configurable special agents for monitoring OpenAI's public status infrastructure and authenticated OpenAI Platform APIs.

Go beyond service availability with visibility into token consumption, spending, prompt caching, API-key attribution and application performance.

What you get

• Service health: OpenAI availability, component status, incidents, maintenance and endpoint health
• API monitoring: Platform API connectivity and collection health
• Usage analytics: Requests, input/output tokens and model-level consumption
• Cost monitoring: Daily and month-to-date spending, including project-level allocation
• Prompt caching: Cached/uncached tokens and cache-read ratios
• Caller attribution: API-key usage and ownership tracking across users and service accounts
• Rate limits: Project/model rate-limit configuration
• Organization inventory: Projects, users, service accounts and API keys
• OpenTelemetry: Optional application-level latency, streaming and token metrics
• Dashboards: Preconfigured Checkmk dashboards for operational visibility
• Alerting: Configurable WARN/CRIT thresholds through Checkmk rules

Requirements

• Checkmk 2.5
• Outbound HTTPS access to OpenAI APIs
• OpenAI organization Admin API key for authenticated Platform monitoring
• Optional OpenTelemetry instrumentation and collection for application metrics

Public status monitoring requires no authentication or OpenAI account.

Installation

Install using mkp add / mkp enable, or via Setup > Extension packages.

Configure the appropriate special-agent rule:

• OpenAI Status Page — public availability and incident monitoring
• OpenAI Platform — authenticated usage, costs and organization monitoring

API credentials can be stored in the Checkmk password store.

Notes

Public service-status monitoring and authenticated Platform monitoring operate independently.

Rate-limit monitoring reports configured limits, not real-time remaining quota. OpenTelemetry monitoring requires separate application instrumentation.