Per-key budgets and cost monitoring

Production cost-control patterns: per-app keys with model_allowlist + budget_cap, plus Slack alerts when spend crosses 80%/100% of monthly budget.

Outline

• One key per service: why proliferating keys is cheaper than one shared key
• Setting budget_cap_usd + model_allowlist via the dashboard or nex keys create
• What happens when a key crosses its budget (HTTP 402 + auto-suspend)
• Reading /v1/usage for cost rollups; per-key, per-model, per-day
• Slack webhook recipe: cron a 5-min job that polls /v1/usage and posts to your #ai-spend channel
• Per-provider spend dashboard via /admin/provider/spend (admin-only — for ops)
• Forecasting: linear-fit the last 14 days of spend and surface "trending to overshoot budget on day N"

Ship-date guess

Sprint 8 (paired with the upcoming team-budgets feature).