Hey everyone. I've been exploring SuperAGI for a small personal data pipeline project (moving some API data → BigQuery). I saw the new "Guardrails" module in the release notes.
My immediate thought was: is this just a fancy prompt wrapper, or does it actually add real control for production? In a typical ETL, I'd handle validation and constraints in the code or with dbt tests.
For example, if an agent is generating SQL for me:
* Does Guardrails just prepend "Please be safe..." to the prompt?
* Or can it actively intercept/validate the agent's proposed action against a schema?
I'm nervous about letting an AI loose on my warehouse without concrete boundaries 😅. Has anyone tried implementing it? I'm especially curious about:
* Can you define "guardrails" as YAML/JSON configs?
* Does it integrate with the action/step execution flow?
* Any performance overhead?
A minimal code example of how you'd set one up would be super helpful!