QuotaSentinel — AI coding quota watchdog

Warn before coding-agent quota runs dry and suggest a fallback

Problem

Developers on V2EX complain that one unfinished task burns through a whole codex plus/pro quota, that Claude accounts get banned, and they constantly debate how to stack model subscriptions. Heavy users switch between platforms without any way to predict burn rate — clients only tell you after you're throttled.

Solution

A small local proxy that sits between your coding agent and the model APIs, tracks token burn against remaining quota per task, warns before the threshold, and suggests a cheaper model or execution strategy. Everything stays local, with no account brokering.

Users

Heavy developers with multiple AI coding subscriptions, indie hackers, and tech leads at small teams.

MVP

Local proxy forwarding API calls; per-session burn vs. remaining quota; desktop/CLI alerts at thresholds; per-task cost log; monthly 'subscription value' summary.

Revenue

Subscription; free tier for stats and alerts, paid tier for multi-account switching advice and history reports.

Why now

AI coding subscriptions are now fixed monthly spend for developers and quota anxiety is constant. A local proxy is easy for a solo dev to validate.