Skip to content

Releases: hidearmoon/agentguard

v1.0.0 — Initial Release

17 Apr 15:15

Choose a tag to compare

AgentGuard v1.0.0

Runtime security layer for AI agents — inspect, control, and audit every tool call.

Install

pip install agentguardx
from agentguard import LocalShield
shield = LocalShield()

@shield.guard
async def send_email(to: str, body: str) -> str: ...

Highlights

  • 5-tier trust model with server-side enforcement
  • 3-layer intent consistency cascade (rules → anomaly → semantic)
  • Two-phase call architecture (data extraction ↔ action execution)
  • 22 built-in security rules + YAML Policy DSL
  • Merkle tree audit trail
  • MCP tool supply chain verifier (Ed25519 signatures)
  • EU AI Act compliance export (Article 12/14/9/17)
  • LocalShield: zero-dependency local mode, no server needed
  • SDKs: Python, TypeScript, Go
  • Integrations: LangChain, CrewAI, AutoGen, OpenClaw, MCP, Dify, AutoGPT, n8n
  • 398 tests (339 core + 59 SDK), all passing