Start with an authorization check

Conceptual check

Subject, action, resource, context. The snippet shows shape. It is not a pinned SDK, not an endpoint URL, and not a contract.

authorization.check({
  subject:  "agent:invoice-agent",
  action:   "payment.execute",
  resource: "payment:9281",
  context:  { amount: 240, env: "prod" }
})

Intended first path

1. 1 / 6

   Sign up

   Create an organization. Email or OIDC when those paths are public.

2. 2 / 6

   Create a tenant

   Isolated agents, NHI, policies, and resources under the organization.

3. 3 / 6

   Register an agent

   Give the identity an owner. If it can act, it must be findable.

4. 4 / 6

   Attach a policy

   Start from least privilege. Shadow is the default before protect.

5. 5 / 6

   Call authorization.check

   Subject, action, resource, context. Conceptual shape until the API host is live.

6. 6 / 6

   See the decision

   ALLOW, DENY, or REQUIRE_APPROVAL — and the event on the trail.

Decision chain

The SDK or gateway asks. Policy and risk inform. The outcome is allow, deny, or require approval.

• ALLOW

  The subject may perform the action on the resource in this context.

• DENY

  The subject may not. In shadow, this is recorded as a would-be deny. In protect, a PEP can block.

• REQUIRE_APPROVAL

  The action waits on a human. That is a decision outcome, not a notification after the fact.

Docs, API, SDK, console

Those hosts are not this website. Until they are public, the form is the path — or the example above.

• Docs

  Product documentation on a dedicated host. Not a docs engine on this website.

  Not public yet

  Read the docs

• API

  Authorization and registry APIs. Reference is not published on this host.

  Not public yet

  API reference

• SDK

  Python and TypeScript are the intended first languages. Public packages are not published yet.

  Not public yet

  Ask about SDKs

• Console

  Register agents, attach policy, read decisions. The product console is not public yet.

  Not public yet

  Open console

MCP

Authorize MCP tool calls as actions on resources, typically at an MCP gateway PEP. Detail: MCP Security. Agents as a category: AI Agent Security.

Related pages

• Authorization

  The check you integrate: subject, action, resource, context.

  Open authorization

• MCP Security

  When the first path is a tool gateway, not only an SDK.

  Open MCP Security

• Security model

  A check without a PEP still does not block.

  See security

• QData Identity Security

  Tenancy, modules, and what the product is not.

  See the product

Questions