Rune plugs into the frameworks you already use. Add runtime security to your AI agents in minutes — scan every input, output, and tool call for prompt injection, data exfiltration, and policy violations.
Rune provides native integrations for the most popular AI agent frameworks. Each integration hooks into the framework's specific middleware, callback, or interceptor system — not a generic wrapper. This means Rune can scan tool call parameters, inter-agent messages, retrieved documents, and streaming responses at the framework level, catching threats that generic solutions miss.
If your framework isn't listed, Rune's generic Shield class works with any Python-based agent. Wrap your input/output processing and get the same multi-layer scanning. Native integrations just make the setup easier and the coverage deeper.
Middleware-native security for LangChain and LangGraph agents
pip install runesec[langchain]Transparent security wrapper for the OpenAI Python SDK
pip install runesec[openai]Drop-in security wrapper for the Anthropic Python SDK
pip install runesec[anthropic]Two-layer security for multi-agent CrewAI workflows
pip install runesec[crewai]Security proxy for Model Context Protocol servers
pip install runesec[mcp]Native interceptor-based security for OpenClaw agents
openclaw plugins install @runesec/openclawRoute agent security incidents to your on-call team
Configure in Rune Dashboard → Settings → NotificationsUnified observability for agent security and infrastructure
Configure in Rune Dashboard → Settings → NotificationsEnterprise SIEM integration for agent security events
Configure in Rune Dashboard → Settings → NotificationsTrack blocked agent actions as Sentry errors
npm install @sentry/node @runesec/sdkDon't see your framework? Rune's generic Shield class works with any Python agent. See the SDK docs
Most integrations take under 5 minutes. Install the SDK with pip install runesec, import the framework-specific integration (e.g., ShieldMiddleware for LangChain), and wrap your existing agent. No code refactoring required — Rune hooks into your framework's existing middleware or callback system.
Yes. Rune's generic Shield class works with any Python-based agent. Call shield.scan_input() before sending data to your LLM and shield.scan_output() before returning results to the user. The framework-specific integrations automate this wrapping, but the core scanning is framework-agnostic.
Rune's L1 pattern scanning adds under 5ms of latency. L2 semantic scanning adds 10-20ms. Both run locally with no external API calls. L3 LLM-based judgment is optional and can run asynchronously (non-blocking). Event shipping to the dashboard is always asynchronous and never blocks your agent's response.
Yes. Create one Shield instance and use it across multiple framework integrations. For example, if you have a LangChain RAG pipeline feeding into a CrewAI multi-agent system, you can wrap both with the same Shield instance and see unified monitoring in the dashboard.
Pick your framework, install the SDK, and wrap your client. Every input, output, and tool call is scanned automatically.
Start Free — 10K Events/Month