1. We do not store your prompts or model responses
The prompt content you send through the API, and the response the model returns, are never written to our database. We do not retain them, analyse them, or use them for training. Once a request completes, that content no longer exists in our systems.
What we record for each call is metadata, used for billing and troubleshooting:
- Which model you called and which upstream channel served it
- Token counts (input, cached input, output, reasoning)
- Cost, duration, time to first token, HTTP status code, retry count
- The IP address the call came from, and the timestamp
None of these fields contains your prompt or the model’s output.