Skip to main content
AI agents execute real actions — deleting records, calling APIs, sending emails. There is no standard safety layer between the LLM’s decision and execution. plyra-guard is that layer. It intercepts every tool call your agent makes, evaluates it against your policy, and blocks, logs, or escalates — before anything irreversible happens.
plyra-guard runs in-process — no sidecar, no network hop. Every evaluation completes in under 2ms.
Plyra Guard Architecture

​
How it works

Every tool call passes through plyra-guard before execution:
  1. Your agent decides to call a tool
  2. plyra-guard intercepts the call in-process
  3. The call is evaluated against your policy — sub-2ms, no network hop
  4. Verdict: ALLOW, BLOCK, ESCALATE, DEFER, or WARN
  5. The decision is written to the audit log

Zero latency overhead

Evaluation runs in-process. No network hop. Sub-2ms per call.

Framework agnostic

Works with LangGraph, AutoGen, CrewAI, LangChain, OpenAI, Anthropic, and plain Python.

Policy as code

Rules live in your repo, reviewed in PRs, tested in CI. YAML or Python — your choice.

Full audit log

Every decision logged — allowed and blocked. Ships to OTEL, Datadog, or your own sink.

Quickstart

Protect your first tool call in 60 seconds.

Policy reference

Write rules that match your threat model.