Runtime health
Agents, channels, adapters, heartbeats, model routes, and last-known runtime status are tracked as operated surfaces instead of disconnected implementation details.
// trace.observed_state
AI observability for operated runtimes
OpenOrange turns AI runtime telemetry into operated proof: observed-state snapshots, health checks, drift signals, request traces, model usage, billing attribution, and audit events.
AI observability in OpenOrange is not just another log stream. It is an operated view of runtime health, drift, requests, models, costs, redaction, and operator events tied to a private instance.
01
Observed-state snapshots show what the operator layer believes is true.
02
Health and drift signals show what changed and when.
03
Request traces preserve attribution and cost without raw payload exposure.
04
Operator events connect observations to plans, approvals, and applies.
Observed state
Snapshots give admins a stable read on the instance: what adapters are attached, what model access exists, what changed, what drifted, and which signals need review.
Trace without leakage
OpenOrange can keep actor, route, model, token, cache, cost, redaction state, and audit context while leaving sensitive message bodies local or redacted.
// questions
Is this just log aggregation?
No. The dashboard is built around operated state: health, drift, request attribution, model usage, billing proof, plans, approvals, applies, and rollback context.
Can observability work with multiple runtimes?
Yes. Runtime adapters normalize what the operator layer needs to observe while letting each runtime keep its execution model.