P
Pramagent
The LLM is not the final authority.
Used for this request only. Never stored. Never logged.
Scenarios
Pick one, then run the trust stack.Policies
The active scenario sets these automatically.PII Scrubbing
Redacts common PII before model contact.
Injection Guard
Blocks common override and exfiltration attacks.
Safety Rules
Applies deterministic block/escalate/redact rules.
Output Judge
Uses a second model call to withhold unsafe generated output.
HITL
Requires approval for consequential actions.
Safe output
AllowedAudit trace
prev:
this:
pip install pramagent
github.com/sriram7737/pramagent
Implementation status
Not a luxury. Mandatory.