Analytics & margins

Two pages: Dashboard (at-a-glance) and Analytics (deep dive).

Dashboard

• KPI cards — active users, active subscriptions, requests in 24 h, jobs in flight.
• Traffic chart — 7-day request volume.
• Cost vs revenue (30 d) — revenue, provider cost, margin in BDT + margin %. This is your business health number.
• Top providers / top models — where the traffic (and cost) concentrates.

Analytics page

Filter everything by provider and time window (24 h / 7 d / 30 d):

• Latency percentiles — p50 / p90 / p95 / p99. Watch p95: users feel it.
• Requests + latency over time — dual-axis chart; failures plotted in red.
• Recent errors — the last 50 failed requests with provider, model, and error metadata. First stop when a customer reports "generation failed".

Margin management playbook

1. Margin % dropping? → Dashboard cost-vs-revenue → find which provider's cost grew (top providers) → check that model's provider pricing vs actual spend, and the affected plans' margin panel.
2. One model suddenly expensive? → the provider may have repriced. Update the model's pricing (its provider cost) or DISABLE it, then re-check plan margins.
3. Provider account draining fast? → Providers → row → Budget bar. Tighten dailyBudgetCents or reduce its routing weight.
4. A single user consuming everything? → Analytics top-users → user detail → check plan limits; consider a custom plan or a chat about enterprise pricing.

Upstream balances

For providers with billing APIs (e.g. OpenAI), the account row menu → account detail exposes upstream usage — your remaining balance at the provider, pulled live. Unsupported providers show "manual lookup required": check their own dashboard.