Quota & Cost Monitoring

Quota monitor — multi-provider

The sidebar Quota drawer (Developer Tools), with a separate panel per provider. Live quota visibility for all three major provider types. Claude Max connects a captured claude.ai browser session to show real 5-hour and weekly usage, with one-click Reconnect when the session expires. Antigravity polls agy /usage for per-model remaining quota with refresh timers. Codex uses provider-specific quota detection. When any model hits a confirmed quota wall, Orbit auto-retries the prompt on the tier’s configured fallback model and cross-links the two pipeline cards with an explanation badge.

Quota auto-fallback

On the Pipeline queue cards (an orange ”⤵ fallback” badge on the retry card, ”↻ retried with ” on the original). When a queued prompt’s primary model hits a confirmed quota wall, Orbit clones it into a fresh auto-approved row targeting the tier’s fallback model, shows a toast, and cross-links both cards. A fallback that also hits quota goes straight to FAILED with no second retry.

Dev Tools API cost monitoring (admin)

The Dev Tools drawer → API Costs / API Usage / Sivants Models. Admin views of monthly AI spend from sivants-api (total, P90, per-user), local per-provider token usage from the api_usage table, and editable internal Sivants model selections per call type.

Sivants Credits

The Credits drawer in the sidebar. A prepaid balance for Sivants-proxied AI features (image, video, premium models), purchased in bundles and never expiring; BYOK keys bypass credits entirely. A warm sidebar with a balance hero, a runway estimate, and a transaction ledger.

Deep dive: context/credits-system.md, context/pricing-and-tiers.md.