Dashboard: Observability

Two of the Observability screens are covered here; Audit Trail and Memory — both Admin/Auditor-only, audit-oriented feeds — have their own pages.

Metrics & Budget

Route: /observability/metrics (Metrics.tsx)

Sections: Platform Activity (agent/workflow rollup), Budget Utilization (licence transaction/spend budgets — see reference/api/usage-budget.md), Usage Summary, Agent Invocations, and Agent Latency charts. Backed by the same /usage, /usage/summary, and /licence/status endpoints the CLI/API surface exposes — this page doesn't compute anything the API doesn't already report.

Gateway & Providers

Route: /observability/providers (ProviderHealth.tsx)

LLM provider status, availability, latency — the same pool composition GET /gateway/pools reports (Local/Cloud split, driver-declared sovereignty disposition, live reachability per pool entry).

Also the read-only "effective configuration" panel (ADR-062 Decision §3): Platform Config never got its own standalone screen, because configuration is never hot-editable under any circumstances — changes go through config.toml plus a restart, full stop, with no console write path at all. Rather than a page that could only ever display and never edit, that content is folded in here, next to the live state it explains:

{
  "auth_mode": "local",
  "providers": 3,
  "registries": 1,
  "allow_local_service_keys": true
}

providers/registries are counts, not full listings — see reference/cli/providers.md and reference/cli/catalog.md for the actual inventories.

Related: