Observe Platform — Unified Observability Workspace
What Is the Observe Platform?
Section titled “What Is the Observe Platform?”The Observe Platform is a unified observability workspace embedded natively inside the AgentWatch admin console — not a separate product, portal, or licensed observability stack. Because it consumes the same usage, DLP, gateway, and fleet telemetry AgentWatch already records, an operator gets a single, org-scoped operational picture without exporting data anywhere or standing up an external time-series database. It is themeable (dark / light) and presents as an expandable Observe Platform menu in the left sidebar, with each surface a sub-item.
Reach it at:
https://<host>/{org-slug}/admin/dashboard → Observe Platform (sidebar)Access follows the same admin session and org scoping as the rest of the console — every panel is filtered to the caller’s organization, and no cross-org data is ever rendered.
The Nine Surfaces
Section titled “The Nine Surfaces”| Sidebar sub-item | Surface | What it answers |
|---|---|---|
| AI | AI Observability | Which models are in use, at what token volume, cost, and latency? What did agents actually prompt (lineage), how do they call each other (topology), and how good are the answers (quality evals)? |
| Services | Application Observability | Where is time spent across a multi-service request? A distributed-trace waterfall and a service-flow map. |
| Fleet | Infrastructure & Fleet | Are my deployed agents healthy? A honeycomb grid, one hex per agent, colored by health. |
| Optimize | AgentKeel Optimization | How much am I saving with embedded caching/compression, direct vs. optimized, per agent? |
| Business | Business Observability | Where is spend going — funnels by provider, team, and model? |
| Security | Security & Threat | What are my highest-risk DLP / WAF events, risk-ranked? |
| Insider Risk | Insider Risk (AI Agent) | Which AI agents are behaving riskily, and what enforcement is in effect? (§7.22) |
| Delivery | Software Delivery | Did the last release pass its guardrails — pass / warn / fail? |
| Logs | Log Analytics | Free-text query bar over recent activity plus a volume histogram. |
AI Observability in Detail
Section titled “AI Observability in Detail”The AI surface is the workspace’s anchor. It renders:
- KPI tiles — total requests, tokens (in/out), estimated cost, and p50/p95 latency for the selected window.
- Time series — request, token, cost, and latency trends over time.
- Model usage — a per-model breakdown of volume, tokens, and cost.
- Prompt lineage — a walkable view of what agents prompted and received, with a strict no-raw-PII guarantee: lineage previews are drawn from the same redacted/archived material the gateway already sanitizes (§7.1, §7.16), so sensitive values never surface in the observability layer.
- Agent topology — a graph of which agents invoke which models and tools.
- Quality evals — scored evaluations of response quality for the observed traffic.
Backend Endpoints
Section titled “Backend Endpoints”Every surface is served by org-scoped, session-authenticated endpoints under /admin/api/observability:
| Endpoint | Serves |
|---|---|
GET /admin/api/observability/kpis |
KPI tiles (requests, tokens, cost, latency) |
GET /admin/api/observability/timeseries |
Trend charts |
GET /admin/api/observability/model-usage |
Per-model breakdown |
GET /admin/api/observability/prompts |
Prompt lineage (PII-safe previews) |
GET /admin/api/observability/topology |
Agent / model / tool topology graph |
GET /admin/api/observability/threats |
Risk-ranked security events |
GET /admin/api/observability/logs |
Log analytics query + histogram |
GET /admin/api/observability/funnel |
Business spend funnels |
GET /admin/api/observability/provider/summary |
Provider-level rollups |
GET /admin/api/observability/verify/last |
Last verification / release-guardian result |
All responses are scoped to the requesting org; there is no tenant API key path into these endpoints — they require a valid admin session.
Step-by-Step: Open the Observe Platform
Section titled “Step-by-Step: Open the Observe Platform”- Sign in to the admin console for your org at
/{org-slug}/admin/dashboard. - In the left sidebar, click Observe Platform to expand it.
- Choose a surface (AI, Services, Fleet, Optimize, Business, Security, Insider Risk, Delivery, Logs).
- Use the time-range control to set the window; panels refresh in place.
- Toggle the theme (dark / light) from the console header — the workspace honors it.
Empty console? If you have not yet driven live traffic, enable Demo Mode (§7.23) to populate every surface with realistic sample data.