from plyra_guard import ActionGuard guard.protect("file.write") # every tool call, governed. risk_level=RiskLevel.HIGH memory.recall("previous context") await trace.record(span) @guard.protect("db.delete") action: ALLOW | DENY | ESCALATE memory.remember("agent learned") # <2ms overhead. measure it.

plyra

Something is watching your agents.

Action middleware for AI agents in production.
Intercept every tool call. Enforce policy. Remember everything.

Guard

Memory

Trace

Get notified before launch.

agent.py

1from plyra_guard import ActionGuard
2from plyra_memory import Memory
3
4@guard.protect("file.write")
5async def write_report(content, path):