pi-insights: usage analytics and statistics page

[setkyar]

Problem / motivation

Users have no visibility into their pi usage patterns — how many sessions, token consumption, model preferences, time spent, peak usage hours, etc. Power users and teams want to understand and optimize their AI usage. The pi-insights project provides a reference implementation with token tracking, session analytics, and cost estimation.

Proposed solution

Add a pi-insights section to the Settings page (or as a standalone /insights page) that shows:

Summary stats:

• Total sessions (all time, this week, today)
• Total tokens consumed (input, output, cache read/write)
• Estimated cost (configurable per-model pricing)
• Total conversation time / active hours

Charts & visualizations:

• Token usage over time (daily/weekly/monthly bar/line chart)
• Model usage breakdown (pie/donut chart per provider/model)
• Session length distribution
• Peak activity hours heatmap

Session-level details:

• Sortable table of sessions with token counts, model, duration
• Click-through to the session detail page
• Filter by date range, model, project

Data source:

• Parse existing JSONL session files to extract token usage from message.usage entries.
• Aggregate and cache results in SQLite for fast queries.
• New GET /api/insights/summary, GET /api/insights/usage-timeline, GET /api/insights/sessions-table endpoints.

Inspiration: https://github.com/ygncode/pi-insights

Scope:

• Frontend: New web/src/insights/ or web/src/settings/insights.js with charting (use lightweight chart library or Canvas/SVG). Integrate into settings page as a "Usage Insights" section or dedicated route.
• Backend: New internal/server/insights.go with aggregation logic, SQLite caching, and API endpoints.
• Reuse existing session-parsing infrastructure from internal/sessions/.

PR requirements:

• E2E tests and screenshots showing the insights page with real session data (or seeded test data).
• Charts must render correctly in both dark and light themes.

Alternatives considered

• Could be a separate service/project entirely (like pi-insights already is), but integrating into pi-web gives a single-pane-of-glass experience without running another server.

Acknowledgement

• I understand the maintainer triages requests and will invite a PR only if this moves forward.