18. Analytics Dashboard

The analytics dashboard provides a comprehensive view of tenant-wide LLM usage, workflow performance, knowledge base growth, and project health. It is composed of 10 panels fed by a single tRPC query (dashboard.getAnalytics) that aggregates data from workflows, discussions, insights, projects, auto-fix runs, refactor runs, and personas. The query auto-refreshes every 60 seconds.

18.1 Architecture

The AnalyticsDashboard component (src/components/dashboard/analytics/analytics-dashboard.tsx) renders a skeleton loader during fetch, then arranges panels in a responsive grid layout.

18.2 Data Type Definition

The full shape of the analytics response is defined in src/types/analytics.ts as the AnalyticsDashboardData interface. Key sections:

18.3 Hero Metrics

The hero section aggregates lifetime totals across all workflows and discussions for the tenant:

• Total Spend -- sum of costEstimate from all workflow steps plus all discussion messages
• Total Tokens -- accumulated from tokenUsage.total on each completed step
• Total Energy (Wh) -- computed via computeEnergy() from src/lib/workflow-metrics
• Time Saved (minutes) -- estimated via computeWorkflowAggregates(), which models manual effort displacement
• Workflows Run -- count of all workflows (any status)
• Success Rate -- completedWorkflows / allWorkflows.length * 100

18.4 Activity Timeline

A 30-day bar/area chart showing daily activity across three dimensions:

activityTimeline: Array<{
  date: string;       // "YYYY-MM-DD"
  workflows: number;  // Workflows created on this day
  discussions: number; // Discussions created
  insights: number;   // Workflow insights created
}>

The backend pre-fills all 30 days (even days with zero activity) to ensure the chart renders a continuous timeline with no gaps. Dates are bucketed using toISOString().slice(0, 10).

18.5 Provider Usage

Aggregates token consumption and cost across all completed workflow steps and assistant discussion messages, grouped by provider name. Both sources contribute to the same provider bucket:

providerUsage: Array<{
  provider: string;    // "anthropic" | "openai" | "google" | "ollama" | "unknown"
  totalTokens: number;
  totalCost: number;
  callCount: number;
}>

Results are sorted by totalTokens descending. Steps without a provider are grouped under "unknown".

18.6 Model Usage Table

The ModelUsageTable component (src/components/dashboard/analytics/model-usage-table.tsx) displays a sortable table with per-model breakdowns. It tracks seven columns:

Sorting is client-side. Clicking a column header toggles between ascending and descending order. The active sort column's icon is highlighted with text-nyx-accent.

18.7 Insight Pulse

Breaks down workflow insights along multiple dimensions:

• By Type -- counts per insightType (pain_point, solution, strength, pattern, decision)
• By Severity -- counts per severity level (blocker, high, medium, low)
• Paired Count -- number of insights that have a non-null pairedInsightId (pain points matched with solutions)
• Top Categories -- top 5 categories (e.g. "Security", "Architecture", "Performance") by insight count

18.8 Workflow Performance

Operational health metrics for the workflow engine:

• By Status -- distribution across pending, running, paused, completed, failed
• Avg Duration (ms) -- mean step execution time across all completed steps with non-null durationMs
• Total Retries -- sum of retryCount across all steps
• Steps With Retries -- count of steps where retryCount > 0
• Total Steps -- overall step count
• Top Errors -- top 5 error messages from failed steps (truncated to 100 chars), with occurrence counts. Sourced from the most recent 200 failed steps.

18.9 Knowledge Base Stats

Counts across the tenant's knowledge stores, plus 7-day growth indicators:

18.10 Project Health (Project Portfolio)

Per-project summary rendered by the ProjectPortfolio component:

projectHealth: Array<{
  id: string;
  name: string;
  status: string;        // "active" | "archived"
  workflowCount: number; // _count.workflows
  insightCount: number;  // _count.workflowInsights
  totalCost: number;     // Sum of workflow step costs for this project
  lastActivity: Date;    // project.updatedAt
}>

Project costs are computed by iterating all workflows, filtering by projectId, and summing computeWorkflowAggregates().totalCost. Projects are sorted by updatedAt descending (most recently active first).

18.11 Success Rates Panel

The SuccessRatesPanel component (src/components/dashboard/analytics/success-rates-panel.tsx) shows three progress-bar groups with color-coded thresholds:

Workflow: Single bar -- completed / total completion rate.

Auto-Fix: Two bars:

• Generate Rate: (fix_generated + pr_created + resolved) / total_issues
• Apply Rate: (pr_created + resolved) / total_issues

Status grouping is done via autoFixIssue.groupBy({ by: ["fixStatus"] }).

Refactor: Two bars:

• Generate Rate: (improved + applied) / total_opportunities
• Apply Rate: applied / total_opportunities

Status grouping is done via refactorItem.groupBy({ by: ["status"] }).

18.12 Persona Overview Panel

The PersonaOverviewPanel component (src/components/dashboard/analytics/persona-overview-panel.tsx) displays mini-cards for up to 20 personas (ordered by usageCount descending), including both tenant-specific and built-in personas.

Each PersonaMiniCard shows:

• Name and Level badge (Lv.{n})
• XP progress bar -- fractional part of the level rendered as a percentage fill
• Traits -- displayed as default-variant badges
• Specializations -- displayed as accent-variant badges
• Three stats: Uses (total count), Success (percentage or --), Last Used (relative time via formatRelativeTime or "Never")

Cards are arranged in a responsive 1-column (mobile) / 2-column (sm+) grid.