StatusDrift — observability for AI coding quality and quotas

Track rate limits, quality drops, and outages across AI coding services so you know what to use today

Problem

On V2EX a developer reports Codex Plus weekly quota shrinking hard: previously a full day using two 5-hour windows stayed under 20% of the weekly cap, now one morning window leaves only 83%. The same day others complain that GPT/Claude feel "dumbed down" and one person literally built a service to monitor it. Add account bans, reseller sites going 502, and model renames like Astra changing writing style, and users have zero signal on whether today is usable or worth renewing.

Solution

A crowdsourced plus active-probe status board. Scheduled standardized prompts measure response quality, latency, and rate-limit thresholds, while anonymous clients report weekly-quota remaining curves. It produces a quality score and availability forecast per service, account type, region, and hour, with alerts when quota is about to run out or quality drops sharply.

Users

Developers and small teams who live in Claude Code, Codex, GPT, and Cursor every day, plus users dependent on third-party relay services.

MVP

Automated standardized prompt runs with scoring; crowdsourced weekly-quota and threshold charts; one red/yellow/green status plus 30-day trend per service; email and webhook alerts; a small anonymous reporter client that submits current remaining quota in one click.

Revenue

Free dashboard for reach; $5–10/month subscription for history, multi-account views, and early alerts; affiliate deals with API relay providers.

Why now

Model providers change quotas and routing constantly and official status pages only report outages, never quality regressions, so a solo dev with probes plus crowdsourcing can sell signal the vendors won't publish.