Cloud/AI Governance/Connector Policy
Connector Policy
The control tower of your fleet: how connectors behave globally, from four eyes deployment approval to tool exposition modes, response presenters and the Global Emergency Halt that shuts everything down in one confirmed action.
The eight Reports screens show what happened; the Settings group decides what may happen, and Connector Policy is its first screen. It governs connector behavior for the whole account: who approves deployments, how tools are exposed to AI clients, how responses are shaped, and the one switch that turns everything off. The console describes the screen in one line: "Configure how your connectors behave globally."

Deployment approval
When enabled, new deployments remain in pending status until a different team member explicitly approves them.
Tool exposition
Controls how tools are exposed to AI clients. Grouping reduces context window usage for large APIs.
Connectors above this tool count are grouped automatically. Connectors at or below it stay flat.
toolsResponse presenters
Automatically generates View schemas from OpenAPI response definitions. Requires a Swagger spec.
Configure how your connectors behave globally.
The mockup is live: flip the switches, switch the exposition mode, open the halt confirmation and type HALT ALL to see the fleet shut down state, then restore it. Nothing leaves this page.
Deployment approval: four eyes on every deploy
One switch with a precise rule: "When enabled, new deployments remain in pending status until a different team member explicitly approves them." Not any team member: a different one. The person who ships is never the person who approves, and no connector reaches production on a single pair of eyes.
Tool exposition: Flat, Grouped or Auto
This section controls how a connector's endpoints become AI tools, and it is a token budget decision as much as a security one:
- Flat. "Each endpoint becomes an individual AI tool. Best for small APIs (< 20 tools)." Every endpoint visible, nothing hidden, context cost grows with the API.
- Grouped. "Related endpoints share one AI tool with an action discriminator. Reduces token usage." Fewer tools in the model's context, same capability behind a discriminator.
- Auto, marked recommended in the console: "Automatically groups when a server exceeds the threshold below." A threshold input (5 to 200, default 20) decides the line: "Connectors above this tool count are grouped automatically. Connectors at or below it stay flat."
The console frames the whole section in one sentence: "Controls how tools are exposed to AI clients. Grouping reduces context window usage for large APIs."
Response presenters
A switch for "Automatically generates View schemas from OpenAPI response definitions. Requires a Swagger spec." With it on, connectors that ship a Swagger spec get typed View schemas for their responses; the AI client sees structure instead of guessing.
The Global Emergency Halt
The red block at the bottom is the kill switch for the entire account, and the console describes exactly what it does: "All active Connectors will be deactivated, all tokens will be revoked, and all running sessions will be terminated across your account. AI agents will immediately lose access to all Connectors. Use only in an emergency."
It is built so that nobody hits it by accident: the button opens a dialog titled "Halt all active connectors?" marked "This action cannot be undone", and the destructive button stays disabled until you literally type HALT ALL.
After the halt, the block turns green and honest about the recovery: "Previously revoked tokens remain invalid. New tokens must be issued for each server." Restore all connectors brings the fleet back online, but the old keys stay dead by design; you reissue them on each connector's Connection Tokens tab. A halted fleet comes back with a fresh set of identities, which is exactly what you want after an incident.
Save lives in the sticky footer next to the line "Configure how your connectors behave globally.", with its Saving... and Saved states.