Pillar — Observatory

Two surfaces, one telemetry stream. The CFO dashboard and the SRE dashboard never disagree.

The Observatory pillar is Sagewai's show-your-CFO-where-the-money-goes surface. Two dashboards (Iron Man HUD + Grafana board) read the same OpenTelemetry stream so neither lies. Per-project tagging gives finance the per-tenant breakdown without manual reconciliation.

What the pillar does

• OpenTelemetry pipeline — admin backend emits OTLP; OTel collector receives + scrapes; VictoriaMetrics stores metrics; VictoriaLogs stores logs; Grafana visualises both.
• prometheus exporter + VM scraping — not prometheusremotewrite (which silently drops histograms and counters; see Issue #66).
• Per-project tagging — every span is tagged with sagewai.project_id; per-tenant rollups drop out of the dashboard.
• Iron Man HUD — agents on a graph, missions in flight, fleet posture across the top. The mission-control surface.
• Grafana board — Sagewai Admin board with 5 rows and 14 panels covering health, HTTP, status codes, OTel pipeline, logs.
• Audit trail — structured business events (setup.completed, auth.login.*, agent.created, agent.run.*, provider.test.*) land in VictoriaLogs.

What proves it works

Primary lighthouse

Observability and cost — Example 43 fires a mixed-tenant workload at the admin backend (real HTTP traffic, real OTel spans), and within three minutes both dashboards show real numbers. The page renders the dashboards from a real run, not pre-canned.

Sibling lighthouses

• Production multitenancy — the per-tenant tagging that makes per-project cost rollups possible.
• Train your own model — once the cost line is on the dashboard, the cost-down target is concrete.

Pattern examples

• Example 34 — observatory_cost_tracking — per-tenant cost tracking, the CFO line.
• Example 40 — fleet_under_load — 20+ workers, Iron Man HUD live state.
• Example 12 — budget_enforcement — per-user/team/project budget caps. The foundation companion.

Where to go to ship it

• Observatory overview — Iron Man HUD + Grafana, in detail.
• Admin Panel — the surface where Audit, Runs, Profiles, and dashboards live.

Related

• Fleet pillar — the per-worker telemetry that the Observatory aggregates.
• Sealed spine — the redaction layer that makes telemetry safe to export.
• Pillars overview — the other four pillars.