Registry ¶

What this is¶

Registry is the inventory of every AI component the scanner has found: agents, MCP servers,
LLM providers, APIs, data sources and tools. It also lists shadow AI — namespaces whose pods
call AI services but are not yet opted in to scanning.

For each item you decide its governance state: Sanctioned (approved), Tolerated (seen, not
yet reviewed) or Unsanctioned (seen only from outside signals). New items arrive as

Tolerated.

What is on the screen¶

• Application dropdown (next to the page title) — All applications, or one application.
  The page shows only what belongs to your choice.
• Tabs — All, Agents, MCP Servers, LLMs, APIs, Data, Tools,
  A2AS Certificates, Shadow. Each tab except Shadow shows a count.
• Governance filter — All,
  Sanctioned, Tolerated,
  Unsanctioned, each with its count. Hidden on Shadow.
• Table / Treemap — switch between the table and a treemap.
• Baseline inventory — approves every Tolerated agent and MCP server of
  the application in one click. Pick one application first; it is disabled under All
  applications.

Columns¶

Every tab has Name, Type, Status (active, detected or unknown), Risk (the
highest severity of the item's open findings) and RPM (recent requests per minute; on Tools
it is Invocations). Agents, MCP Servers and All add Cert (A2AS certification);

Tools adds Sanctioned. Some tabs add their own columns, for example Model on Agents,

Spec and Latency on APIs, Data Class on Data, Category on All.

When this page is empty¶

On a new workspace every tab shows zero until the scanner reports. Shadow usually fills first:
it lists namespaces whose pods call AI services, before you opt them in. See
Getting started.

Detail panel¶

Click a row to open everything known about the item. The sections depend on the kind: agents show

A2AS Components, configuration (model, tools, owner, data class), performance and known CVEs;
MCP servers show transport, tools and how they were detected; APIs show detection source, protocol
and traffic; LLM providers show models and traffic; shadow detections show how the use was found
and which workload made it. The panel is read-only — change governance with the row buttons.

Statistics panel¶

With no row selected, the side panel shows statistics for what the current tab and filter show:

• Certification Rate — the share of items that are A2AS certified.
• Total, Sanctioned, Tolerated, Unsanctioned — counts.
• Risk Distribution — items per severity.
• By Type — items per subtype.
• Detection Coverage (APIs tab only) — how APIs were detected: source code, HTTP probing,
  eBPF, inferred.
• The footer — N of M not A2AS certified · X% shadow.

How to use it¶

Bring shadow AI under coverage¶

1. Open Shadow, or find a row of type Shadow Detection in the table.

2. Click Monitor to record the namespace without blocking, or Enforce to record and block
   by your policies. It takes effect on the scanner's next cycle.

3. After Monitor or Enforce, a namespace whose pods were already running shows
   ⚠ Restart to complete. Until then those pods are observed and connection-level policy
   applies, but their HTTPS is not decrypted. Click Adopt: it restarts the namespace's
   workloads (rolling restart) so their pods get the scanner proxy's certificate and
   content-level policy applies.

4. Revert removes coverage from the namespace.

Enforce needs the scanner installed with its proxy — see
Getting started.

Approve or revoke items¶

• Promote — on Tolerated and Unsanctioned rows; approves the item
  (Sanctioned).
• Demote — on Sanctioned rows; revokes the approval (Tolerated).
• Baseline inventory — approves all Tolerated agents and MCP servers of
  the selected application at once. Useful right after onboarding, when what the scanner found is
  trusted.

To find what still needs review: click Tolerated and sort by

Risk.

Check A2AS certification¶

Open A2AS Certificates, or open an agent and read its A2AS
Components. Each of the five components says why it is on or off:

• Behavior Certificates — on when the agent has a signed behaviour certificate that this
  workspace can verify and enforce. Open certificate opens it.
• Authenticated Prompts, Security Boundaries, In-Context Defenses, Codified
  Policies — each on only when the agent's A2AS profile turns it on in enforce mode and a
  request from the agent was instrumented with it in the last 24 hours.

Configure A2AS, or Open A2AS next to
a component, opens the A2AS page with the agent selected.

Where to go next¶

How you get here¶

• Sidebar → Discover → Registry.
• Briefing → a metric's detail → Govern agents in the registry.

Was this page helpful?

How can we improve this article? (Optional)

Information is unclear or confusing

Insufficient information

Outdated or incorrect information

0/240

Send