Introduction

The Policy Compiler is a distributed system for securing LLM-based agents through context-aware authorization.

The Problem

Traditional authorization is stateless — it only sees “User X wants Action Y”. It has no visibility into:

• What conversation led to this request
• What data the agent has already seen
• The causal chain of messages

This creates security gaps that are impossible to address with conventional auth.

The Solution

This platform tracks the full message dependency graph — every message, tool call, and their causal relationships — enabling policies like:

Prerequisite Actions

“Allow FDA API only after registration approval”

Information Flow Control

“Block email if agent saw sensitive data AND untrusted input”

Key Features

• Zero code changes — Instrumentation patches your HTTP libraries invisibly
• Graph-based policies — Write rules in Datalog that query conversation history
• Credential isolation — API keys stay server-side, never exposed to agents
• Real-time evaluation — Differential Datalog for incremental policy checks

Next Steps

• Installation — Set up the platform
• Quick Start — Run your first demo
• Full Tutorial — Deep dive into the architecture