Kastra

What is Kastra?

Kastra is a runtime authorization layer for AI coding agents that enforces policies with sub-1ms latency before any tool or API call executes. It covers Claude Code, Cursor, Codex, OpenClaw, and major SDKs through a single control plane.

Key features

• Sub-1ms policy enforcement on every tool call
• Covers Claude Code, Cursor, Codex, OpenClaw, Anthropic SDK, OpenAI SDK
• Prevents unauthorized tool use and prompt injection in real time
• Open-source runtime and policy pack library
• Single control plane for multi-agent governance
• Fine-grained controls over prompts, inputs, and outputs

Who it's for

• Teams running Claude Code or Cursor on shared/sensitive repositories
• Agentic workflows that call external APIs or execute system commands
• CI/CD pipelines where agent actions must stay within policy boundaries
• Solo founders managing multiple coding agents across projects

When not to use it

Not needed if you only use a single coding agent on your own machine with no sensitive data. The control plane adds architecture complexity best suited for teams or shared environments.

FAQ