Skip to main content
MetrixLLM logs every request that passes through the gateway in real time. This gives you complete visibility into exactly what your application is sending to the providers, and exactly what is coming back.

​
What is logged

For every request, the gateway captures:
  • Identifiers: request_id, timestamp
  • Routing: The upstream provider, exact model used, and any fallback logic triggered
  • Cost & Tokens: Exact prompt and completion token counts, and the precise cost in USD
  • Performance: End-to-end latency in milliseconds, and cache HIT/MISS status
  • Payloads: The full prompt text and response text (up to 10,000 characters)

​
Filtering and searching

The dashboard allows you to filter your logs across multiple dimensions:
  • Time range: View requests from the last hour, day, or a custom window
  • Provider & Model: Filter down to specific endpoints
  • Status code: Quickly find requests that returned 5xx errors or 429 rate limits
  • API Key: Identify which application or microservice generated the traffic
  • Custom properties: Filter by specific metadata tags you attached to the request

​
Data retention & privacy

By default, MetrixLLM logs the full text content of prompts and responses. This is incredibly useful for debugging. However, if you are handling sensitive data (PII, PHI) and cannot store payloads in a third-party system, you can disable Data Retention in your workspace settings. When Data Retention is disabled:
  • The gateway still logs all metadata (tokens, cost, latency, provider).
  • The prompt_text and response_text fields are immediately discarded and never written to the database.