Shield Agent AI

← Docs hub

Customer guide

KYA quickstart

Shield Agent AI is Know Your Agent (KYA): identity, policy, custom tools by toolId, human approval, redacted evidence, and a trail for agents that can change production systems, on both IDE and runtime hosts.

The package is a checkpoint those agents call. It does not scan a network. Register the agent, wrap the tool, get Allow / Hold / Deny. Hold waits for a person. Agents that never call evaluate stay invisible. Walkthrough: how you use it.

First hour

  1. Pick a path: /install (npx first), light install (15 min), full local console, or marketplaces.
  2. Register an agent and your tools (any stable toolId).
  3. Point agents at MCP tools or OpenAPI (connectors).
  4. When policy says REQUIRE_APPROVE, the action waits for a person.
  5. Optional: open the hosted console for a visual approvals queue (/login) and Welcome tour.

A side effect that needs approval runs only after APPROVED. Otherwise it fails closed. Creating an agent is a tool (mint and spawn). A live session can lose Deploy or Build (clearance). Passports and session claims are signed (claims).

See also: How KYA works · Source.

Docs hub · Install · How KYA works · Source