Reserved Session Fields
| Field | Description |
|---|---|
metadata.num_events | Total events in the session |
metadata.num_model_events | Number of model (LLM) events |
metadata.has_feedback | Whether any event has feedback |
metadata.cost | Total LLM cost (from provider pricing) |
metadata.total_tokens | Total tokens (prompt + completion) |
metadata.prompt_tokens | Input tokens |
metadata.completion_tokens | Output tokens |
start_time | Earliest start_time of any event |
end_time | Latest end_time of any event |
duration | Calculated from end_time - start_time (unless manually set) |
Token and cost tracking
HoneyHive automatically rolls up token counts and cost from everymodel event in a session.
| Session Field | Aggregation |
|---|---|
metadata.total_tokens | Sum of prompt_tokens + completion_tokens across all model events |
metadata.prompt_tokens | Sum of input tokens across all model events |
metadata.completion_tokens | Sum of output tokens across all model events |
metadata.cost | Sum of per-call cost, calculated from provider pricing for the model used |
usage field). Cost is calculated server-side using up-to-date pricing tables for each model. If a session contains multiple LLM calls (e.g. a chain or agent loop), all calls are summed at the session level.
You can use these fields to filter, alert, and chart in the HoneyHive dashboard - for example, flagging sessions where cost > 0.10 or total_tokens > 10000.
Duration Calculation
Duration is calculated asmax(end_time) - min(start_time) across all events in the session. This works for both synchronous and asynchronous execution:
- Synchronous: Events chain sequentially (Session Start → Event 1 → Event 2)
- Asynchronous: Events branch in parallel from Session Start
duration on the session, that value is preserved.