ShrikeShrike Documentation
Everything you need to govern what your AI agents do — every tool call, command, and query evaluated against your policy before it runs. It catches prompt injection, data leakage, and unsafe actions along the way.
Quick StartStart Here
Get scanning in 5 minutes. Sign up, grab an API key, and run your first security scan.
Read more →Agent FrameworksFor builders
A starter per framework: three lines govern an agent on the Claude Agent SDK, OpenAI Agents SDK, Google ADK, LangChain or LangGraph, CrewAI, or the Vercel AI SDK, and the four-layer stack beneath it.
Read more →SDK Reference
Python and TypeScript SDKs. Integrate AI security in 3 lines of code.
Read more →API Reference
REST endpoint documentation for the Agent and Dashboard services.
Read more →MCP Server
Protect Claude Desktop, Cursor, and Windsurf with 14 security tools.
Read more →Agent Scope DeclarationNew
Declare what an agent is for — purpose, allowed tools, expiry. Out-of-scope calls route to human approval before they execute. You declare; Shrike enforces.
Read more →Local ModelsNew
Govern a model served by Ollama, vLLM, or LM Studio with the same SDK, policy, scope, and evidence record as a hosted one. A local model ships with no provider guardrails; this is the control that survives the move.
Read more →Cookbook — Working with ShrikeNew
Drop-in system-prompt block that teaches your agent how to react to Shrike verdicts. ~180 words. Copy, paste, ship.
Read more →OWASP Agentic Top 10 → ShrikeNew
Every risk in the OWASP Top 10 for Agentic Applications (2026) — ASI01 Agent Goal Hijack through ASI10 Rogue Agents — mapped to where Shrike enforces at the action layer.
Read more →Compliance
NIST AI RMF, SOC 2, HIPAA, and ISO 27001 alignment documentation.
Read more →Need help? Contact support or visit our GitHub.