What you see
- Outcomes: success, failure and retry counts, per turn and rolled up.
- Usage: tokens and cost for each turn.
- Speed: latency per turn.
- Artificial Experience activity: the kinds of memory events that happened, so it is not a black box.
- Tools and services: activity from tools and services where it applies to your workspace.
What you do not see
- Conversation content. None is recorded in Observability, by design.
- Which models engines use. Beunec does not disclose the models behind its engines.
Why it matters
Governed AI work needs an honest record. When a turn succeeds only on a retry, spends credits, or uses Artificial Experience, Observability makes it visible. That lets a team check cost and reliability without reading anyone's prompts.

