Agent Model Drift Sentinel

Monitors AI coding assistants for model swaps and cache anomalies, alerts and auto-fixes

Problem

Developers notice VSCode Copilot quietly switching away from Claude Haiku 5.5 without warning, changing behavior mid-task. Those using third-party API proxies see cache hit rates suddenly drop from normal to ~10%, causing endless loops and lost productivity. These issues are hidden, slow to diagnose, and break flow.

Solution

A lightweight local monitor that reads configs and request logs from AI coding clients (Copilot, Cursor, Claude Code, API proxies) to track the actual model in use, cache hit rate, and latency. On anomalies, it sends desktop notifications or webhooks and can optionally auto-revert to a pinned model or proxy.

Users

Heavy users of AI coding assistants, solo developers, small engineering teams, and anyone relying on third-party API proxies

MVP

Detect model changes from Copilot/Cursor configs; parse proxy logs for cache hit rate and latency; trigger system notifications on thresholds; one-click revert to pinned model; daily anomaly summary

Revenue

Free tier with desktop notifications and manual checks; subscription adds auto-switching, multi-device sync, historical trends, and webhooks

Why now

AI coding models iterate fast with little transparency, and users are increasingly intolerant of model drift and API cache instability. Indie developers are heavy users themselves, making validation easy and team size minimal.