Rune Documentation

Rune is a runtime security platform for AI agents. It scans LLM inputs and outputs for prompt injection, data exfiltration, and policy violations — with under 10ms overhead.

Quick Start

Install pip install runesec and protect your first agent in 5 minutes.

Getting Started

Install the SDK, create an API key, and scan your first agent input in under 5 minutes.

SDK Reference

Complete reference for the Shield class, scanning methods, policy evaluation, and configuration.

Integrations

Drop-in security for LangChain, OpenAI, Anthropic, CrewAI, and MCP servers.

Policies

Write declarative YAML policies for tool access, data protection, and rate limiting.

REST API

Manage agents, policies, and alerts programmatically with the v1 API.

MCP Server

Use Rune as MCP tools in Claude Code, Cursor, and Windsurf. Scan, redact, manage policies.

Environment Variables

Complete reference for all SDK and dashboard environment variables.

Troubleshooting

Common errors, debugging tips, and frequently asked questions.

Dashboard Guide

Navigate the real-time monitoring dashboard: overview, agents, events, alerts, and settings.

How It Works

Your Agent (Python)
    |
    v
Rune SDK (scan + policy check, <10ms)
    |
    v
Cloudflare Worker (ingestion)
    |
    +---> ClickHouse (analytics, event storage)
    +---> Convex (alerts, policies, agent state)
              |
              v
         Dashboard (real-time monitoring)