Qarinah

I spent two years building a free tool that ranked well and earned almost nothing. The traffic was real. The audience just had no reason to buy anything. That was the lesson, and YearJersey is the same skill pointed at people who are actually shopping.

This is a very valuable ecosystem.

The evidence-linked append-only ledger and cited read models address a real weakness in long-running agent work: context can survive without blindly replaying every conversation. The difficult part is stale truth when a decision is later reversed. How does Qarinah mark evidence as superseded or resolve conflicting decisions while preserving the audit trail?

I built Qarinah after watching coding agents repeatedly lose decisions, replay huge project histories, and hand work between tools without durable evidence. Qarinah keeps a local, verifiable project memory so a fresh agent can recover cited context instead of starting from zero. I would love feedback from maintainers using multiple coding agents.