I built a runtime guardrail that stops AI agents from doing dumb things

9 points by thomaslwang 4mo ago ↗ HN

Most AI security tools detect problems AFTER they happen. I built MoltGuard - runtime guardrails that block dangerous tool calls BEFORE execution.

Install: curl -s https://wry-manatee-359.convex.site/api/v1/download?slug=moltguard -o moltguard.zip

Currently at 16K+ downloads protecting AI agents from doing dumb things like deleting production databases or leaking credentials.

4 comments

[ 3.6 ms ] story [ 12.1 ms ] thread

_wire_ 4mo ago ↗

Thanks to the unbridled power of artificial intelligence, I just have to specify all the edge cases and the world will be mine!

thomaslwang 3mo ago ↗

Exactly! The challenge is that agents optimize for goals in unexpected ways. MoltGuard adds runtime guardrails to block the unexpected tool calls before they execute. Thanks for the comment!

gus_massa 3mo ago ↗

Licence?

Are you planing to publish it in GitHub?

[dead] Mooshux 3mo ago ↗

[flagged]