Open-Source AI Agent Security

AgentWatch

Secure your AI agents with policy-based access control, data loss prevention, and full audit logging — open source, Apache 2.0 licensed.

View on GitHub Book Demo

Core Capabilities

A security layer between your AI agents and the tools they access — enforce policy, prevent data leaks, audit everything.

Role-Based Access Control

Define roles with specific tool permissions. Each user sees only the tools their role allows.

Least-privilege enforcement

Policy-Based Tool Filtering

YAML-driven policies control which tools are exposed to which agents and users.

Declarative security rules

Data Loss Prevention

Automatically detect and block sensitive data patterns in tool arguments before they reach upstream services.

PII and secret protection

Structured Audit Logging

Every tool call, policy decision, and DLP event is logged with full context for compliance and forensics.

Complete audit trail

MCP Protocol Native

Built for the Model Context Protocol. Sits between AI clients and MCP servers as a transparent security proxy.

Drop-in deployment

Docker-Ready

Ship with Docker Compose, Makefile, and health checks. Production-ready in minutes.

Deploy in minutes