Are your logs used to train AI models?
No. Never. API inputs are processed and discarded under our provider agreement.
Does OperatorMesh store raw logs?
No. Webhook-triggered analyses run in memory and are discarded when the function terminates.
What is stored in the dashboard?
The analysis result only — root cause, confidence, actions. Never the raw log payload.
Can I delete my analysis history?
Yes. Any time. From the dashboard. Deletion is permanent and immediate.
Which AI model processes my data?
Anthropic Claude (claude-haiku-4-5 for speed, claude-sonnet-4-6 for complex analyses). No other models used.
How are confidence scores calculated?
Confidence is the model's own self-assessment, generated as part of its structured response — not a separately computed statistical score. Diagnosis confidence and remediation confidence are scored independently, because being certain about a root cause doesn't mean a fix is certain to work. We've adversarially tested this self-assessment against cases with strong evidence, contradictory evidence, and deliberately insufficient evidence — in testing, confidence dropped appropriately when evidence was weak or ambiguous, rather than defaulting to a high number regardless of input quality. That testing methodology and its results are not yet independently audited — it's our own internal evaluation, disclosed here rather than left unstated.
Is OperatorMesh output advisory or autonomous?
Advisory only. OperatorMesh never executes fixes, restarts services, or modifies infrastructure.
What happens to screenshot images?
Compressed in browser → sent to vision API → text extracted → image discarded. Never stored.