What is Tracked
Every log entry captures comprehensive metadata:| Field | Description |
|---|---|
| Timestamp | The exact time the completion originated |
| Provider | The AI provider used (e.g., OpenAI, Anthropic, Google) |
| Model | The specific model (e.g., gpt-4o, claude-3-5-sonnet) |
| Tokens (In/Out) | Prompt (input) and completion (output) tokens consumed |
| Duration | How long the request took to complete (ms) |
| Cost | Estimated cost based on the provider’s token pricing |
| Endpoint | The API endpoint called (e.g., chatComplete) |
Log Detail View
Clicking any log entry opens a detail panel showing the full request and response data: The detail panel includes:Overview
Overview
Status (Success/Failed), Trace ID, Span ID, provider name, model used, and the API endpoint.
Metrics & Cost
Metrics & Cost
Execution duration (ms), total tokens, total cost, and a breakdown of input vs output tokens.
Request Body
Request Body
The exact JSON payload sent to the model — including system prompts, user messages, and model parameters like temperature and max tokens.
Response Body
Response Body
The full model response content, including finish reason, token usage details, and any tool calls.
Filtering and Search
As your log volume grows, isolate specific issues or usage patterns by applying filters:- Status — Filter
FailedvsSuccessfulrequests to separate errors from valid traffic. - API Key — Filter by individual API keys to identify which application is driving specific traffic patterns.
- Provider & Model — View traffic for a specific provider (e.g., only Anthropic or only OpenAI requests).