What is Kyra?

Kyra is a runtime governance platform for AI agents. It sits outside your agent and intercepts tool calls before they execute, enforcing your organization's policies in real time.

What it does

When your agent decides to call a tool, Kyra intercepts that call and checks it against your configured policies. If the call is within policy, it proceeds. If not, Kyra blocks it or routes it to a human approver — before any action is taken.

Agent → tool call → Kyra → [allow / block / escalate] → tool executes (or not)

This happens externally to the agent. The agent does not need to be modified to understand your authorization boundaries — Kyra enforces them from the outside.

Why this matters

An agent can reason correctly about a task and still perform actions that cross organizational boundaries it was never told about. Kyra closes that gap without requiring you to re-engineer your agent.

What you need

• A Kyra account (sign up at kyratech.io)
• An API key from the dashboard
• One of our SDKs: Python, TypeScript, or Go

Start with the Quickstart.

What Kyra is not

• Not a prompt injection detector — Kyra governs tool calls, not LLM inputs
• Not an agent framework — works with any framework you already use
• Not a correctness checker — Kyra checks authorization, not whether the agent's reasoning is right

Next: Architecture Overview →