AI Objectives

Declared "keep this true" goals for your account. You state an objective — for example, keep p95 latency under 300 ms on an endpoint, or keep the error rate under 1% — and a reconciler evaluates it on a rolling window. When an objective is breached it either records the breach (observe), raises an Iris nudge that pre-seeds the assistant to diagnose and fix it (nudge), or — once autonomy is enabled for an actionable type — applies a dry-run-clean fix automatically (auto). Recovered objectives clear themselves.

Objectives are distinct from SLO targets (which drive the SLI display): objectives are multi-type, can be scoped to a single endpoint, and are actionable.

Account scope + permissions: objectives belong to your active account and require the traffic-inspector permission (read to list, write to change). Endpoint-scoped latency/error objectives need traffic capture enabled on that endpoint, otherwise they read back as insufficient data and never false-breach.

Objective types

List Objectives

Returns the account's declared objectives under an objectives array.

Create an Objective

Declares a new objective. The tuple (type, scope, endpoint) is unique per account — a duplicate returns 409.

Request body

Update / Delete an Objective

PUT tunes an objective's parameters (its type, scope and endpoint are immutable — they form the unique target). DELETE removes it. Both return 404 for an unknown UUID.

Errors