Monitoring & Cost Dashboards

Document ID: DASH-006
File Path: docs/10-dashboard/monitoring.md
Version: 1.0.0
Status: Draft
Owner: AI Platform Team
Last Updated: 2026-06-27

1. Purpose

This document specifies the monitoring and cost surfaces of the dashboard — the operational views that show platform health, live activity, and spend across agents, workflows, tools, memory, and model usage.

It is the human-facing layer over the platform’s telemetry and the planned Observability section.

2. Surfaces

View	Shows
Home / Health	Service health, error budgets, throughput
Activity	Live agent runs and workflow executions
Cost Explorer	Spend by tenant/project/agent/model over time
Usage	Tokens, tool executions, memory operations
Alerts	Active alerts and recent incidents

3. Health Overview

Aggregates /healthz//readyz//metrics and SLO status across services (API Gateway, Agent Runtime, Workflow Engine, LLM Gateway, Memory Engine, Tool Runtime, Plugin Engine):

Up/degraded/down per service
SLO attainment and error-budget burn (e.g. Tool Runtime SLOs)
Request rate, error rate, latency (RED metrics)

4. Live Activity

Real-time feed of in-flight work, streamed via the BFF websocket bridge:

Agent runs (status)
Workflow executions (status)
Tool executions (status)

Drill into any item to its trace/step inspector.

5. Cost Explorer

A first-class spend view fed by platform cost events:

Cost events (LLM Gateway · Tool Runtime · Memory)
        │  Event Bus → analytics
        ▼
   Cost Explorer (by tenant / project / agent / model / time)

Model spend from LLM Gateway token management
Tool execution cost from Tool Runtime
Cache savings from LLM Gateway and Memory compression
Budget/quota utilization vs. project quotas

Supports grouping, filtering, and CSV export.

6. Usage Analytics

Token usage by type (prompt/completion/cached)
Tool execution counts, durations, success/error rates
Memory operations, recall proxies, index growth
Trends and anomalies

7. Alerts

Surfaces active alerts (e.g. queue backlog, provider failover spikes, egress anomalies, budget breaches) with links to the relevant runbooks. Alert routing and rules belong to the platform Observability section; this view consumes them.

8. Drill-Down & Correlation

Every metric/alert links to underlying traces using the shared request_id/trace context (Overview §14), so an operator can go from a cost spike or error rate to the exact runs that caused it.

9. Scoping & Access

Views are tenant/project scoped; users see only their authorized scope.
Cross-project/organization roll-ups require org.admin/platform.admin.
All access is audited.

10. Embedding External Dashboards

Where deployments run Grafana/Prometheus (deployment architecture), the dashboard can embed those panels alongside native views for infrastructure-level metrics.

11. Dependencies

13. Revision History

Version	Date	Description
1.0.0	2026-06-27	Initial Monitoring & Cost Dashboards specification