Show HN: Keep your credentials out of the AI agents you use and build

Security should be free! I was getting mad that all the security tools that meant to guard your agents are not local and those who are, just cost a fortune.

So I’ve built Weir. You can watch you sensitive data flow, your traces and your egress tools calls.

It doesn’t stop your ADLC, it only watch it locally (no egress).

The agents you use - I’ve set a Claude code hook that stripe your known keys from tool output before the model sees them.

The agents you build - Weir plants fake keys, runs your agents flows and report every place they leaked (traces, logs, model, services). It helped me find CVE-2026-116691 in MLFlow.

Would be super happy to know what you guys think!