Local AI Quota Real-Usage Tracker

Help heavy Claude/Codex users see weekly token burn and the 'downgrade' boundary

Problem

A V2EX user reported huge workloads in the past two weeks, mostly running luna + max, burning roughly a billion tokens per week, doubting the codex quota statistics and feeling the model degrade sharply when quota runs low.

Solution

A local-first tracker that aggregates usage logs from multiple AI coding CLIs like Claude and Codex, showing weekly/daily token burn, remaining quota, and model response quality trends. Alerts before you hit the degradation threshold so you can allocate models wisely.

Users

Heavy developers, indie hackers, and small teams using multiple AI coding CLIs at once.

MVP

Read local CLI logs and session files; compute weekly/daily token burn; show remaining quota; detect response quality drops and alert; pure local storage with CSV export.

Revenue

Free basic tier plus subscription for multi-device sync, team dashboards, historical comparison, and alerts.

Why now

Multi-model CLI use is the norm, official quotas are opaque, and there is no tool to quantify degradation. An indie developer can start purely local with no LLM or cloud costs.