We use optional cookies to understand website visits and identify visiting businesses. You can change your choice anytime using Cookie preferences.
Apollo records page visits and sets a first-party cookie to identify visiting companies globally. If person-level tracking is enabled in Apollo, it may also identify visitors from the U.S. Read Apollo's privacy policy.
Your browser's Global Privacy Control signal is enabled, so Apollo tracking is off.
Use cases / Google Cloud
Keep Google’s controls. Put shared business limits, owner approvals and decision evidence around consequential actions—even when everything stays on Google Cloud.
One cloud or many. Your business stays in charge.
Build the agent on Google. Decide its authority with GovernorAI.
customer-data-agent v3 · BigQuery
Send a customer table to an unapproved Cloud Storage bucket.
analytics-prodExtract jobcrm.customersgs://unapproved-export-bucketThe export stops before dispatch. The agent can continue its approved analysis without moving customer data outside policy.
DATA-BOUNDARY v14 · Customer data may only be exported to approved destinations.
The action is held for its owner, who approves or rejects it.
Permission is ready. The job has not been submitted.
{
"sample": true,
"provenance": "local_interactive_marketing_example",
"schema": "governorai-example/v3",
"release_authority": {
"record_id": "example-release-03",
"agent_version": "customer-data-agent/v3",
"snapshot_id": "example-snapshot-03",
"policy_pack": "google-data-controls/r14",
"status": "approved",
"provenance": "synthetic_fixture",
"candidate_version": "customer-data-agent/v4",
"candidate_status": "blocked_required_assessment_missing"
},
"record_id": "example-export-initial",
"agent_id": "customer-data-agent",
"provider": "gcp",
"project_id": "analytics-prod",
"normalized_action": "bigquery.jobs.insert",
"action_kind": "extract",
"resource": "analytics-prod.crm.customers",
"arguments": {
"job_type": "extract",
"source_table": "analytics-prod.crm.customers",
"destination_uris": [
"gs://unapproved-export-bucket/customers.json"
]
},
"policy": {
"id": "data-boundary"
},
"decision": "DENY",
"reason": "Unapproved destination",
"approval": {
"status": "not_required",
"required": false,
"required_role": null,
"releases_call": false
},
"dispatch": {
"status": "not_dispatched",
"permitted": false,
"downstream_job_id": null
},
"execution_outcome": {
"status": "not_executed",
"observation": "no_execution_observation"
},
"evidence": {
"cryptographically_signed": false,
"chain_verification_performed": false,
"contains_real_cloud_events": false
}
}Why GovernorAI—even on one cloud
“Single cloud” describes where your agents run. Your business still decides what they may do, who answers for it, and what the record must prove. GovernorAI brings those decisions together.
Keep a business limit attached to the customer, case or process—not reset by a new agent, session or project. Count confirmed actions and outstanding commitments together.
Business exposure. Not just model spend.Let Finance or the data owner approve the request without becoming a cloud administrator. Separate policy ownership from agent development, and hold the action for its owner.
An approval is not a blank check.Connect the approved agent version, policy, human response and execution result. Keep an exportable record as runtimes, configurations and teams change.
The business rule outlives the deployment.Try one request
A Google Cloud agent issues service credits through a Cloud Run API. GovernorAI decides each request before the credit API is called, and records the decision. Choose a request.
Each request is evaluated on its own, before the credit API is called. An approval is not an execution: this example never calls the API.
$600 is above the $500 automatic threshold, so it is held for Finance before the credit API is called. A request nobody decides expires after 15 minutes.
{
"sample": true,
"provenance": "local_interactive_marketing_example",
"schema": "governorai-decision-example/v1",
"scope": {
"tenant": "example-tenant",
"provider": "gcp",
"project_id": "customer-ops-prod"
},
"tool": "credits.issue",
"args": {
"amount_minor": 60000,
"currency": "USD",
"account": "acct-C-1042"
},
"policy": {
"name": "finance.credits",
"rule": "approval-over-500"
},
"decision": "HOLD",
"approval": null,
"dispatch_status": "not_dispatched",
"evidence": {
"contains_real_cloud_events": false,
"cryptographically_signed": false
}
}In the product
Recorded on a demo tenant: one governance policy previewed for a provider target, the same policy for a second provider, every agent on one map, Google Cloud included, and decisions from every platform in one log.
Real product screens from a demo tenant with seeded activity. The provider outputs are previews of what the policy becomes for each target; nothing was applied.
Google provides identity, access, content and tool-use governance. GovernorAI connects those controls to the business case: shared exposure, an accountable owner and the record of the result.
Agent Identity, IAM, and Agent Gateway establish the native access boundary.
GovernorAI addsA trusted business scope and shared limits across agents and sessions, while Google’s native access decisions remain in force.
Model Armor inspects and sanitizes content. Semantic Governance evaluates proposed tool use against intent and business constraints.
GovernorAI addsBusiness-owned policy and approval holds. Native risk signals inform the decision; they cannot turn a hard policy denial into permission.
Cloud Logging and Trace help teams understand activity across their Google Cloud environment.
GovernorAI addsThe business case, reservation, approval, dispatch and observed result in one exportable decision trail.
Connect business authorization through Agent Gateway’s custom egress extension. Preserve Google’s access and content checks.
Google’s extension contract ↗Use an ADK tool integration, a customer-controlled Envoy path, Apigee or MCP to apply the same business decision model.
Explore integration paths ↗Native denials stay effective. Track the control that made each decision and the business authority behind an approval.
Explore the policy lifecycle ↗01 / Connect
Connect a project, folder or organization with federated identity and infrastructure your team can review.
Use Workload Identity Federation and short-lived credentials. Keep access scoped to the work it needs to do.
Use Terraform and Infrastructure Manager. Your team approves the resources, permissions and deployment plan.
Validate the deployed path, expose missing permissions and carry a clear connection state into ongoing operations.
Connect the inventory, the identity and the control path. Find workloads that need attention before they become production surprises.
Join each workload to its owner, identity and action path. Distinguish discovered, observed and enforced coverage. Name missing permissions, stale evidence and ungoverned routes.
Agent Registry and runtime APIs identify deployed resources. Gemini Enterprise adds registered employee-facing agents. Cloud Audit Logs and Workspace Admin Reports contribute activity and consent signals. Key metadata and IAM context connect access to ownership.
From managed agents to custom workloads, put control where an AI decision becomes a real-world action.
Govern model selection, usage and request constraints at the provider boundary.
provider_gcp · model-call policyAdd business authorization through the native egress extension, alongside Google’s controls.
custom egress · ext_procEvaluate tool calls and delegated work before they reach data or production systems.
tool boundary · A2A delegationKeep control with your custom workloads using runtime integration or a deployed Envoy path.
runtime · Envoy ext_procApply policy through the gateways your team already operates.
forward-auth · LiteLLM · PortkeyGovern Google-managed MCP, MCP Toolbox and your own tools, argument by argument.
mcp_invocation · action policyThe provider proxy constrains calls to Gemini and Vertex AI. Recheck the exact request after an applied constraint.
Use the tool, runtime or gateway path for BigQuery, Cloud Storage and enterprise actions. Bind identity, resource and arguments to the decision.
IAM, network boundaries and native denials remain in force. A GovernorAI allow is never an override of Google’s access decision.
Turn one business policy into native controls and runtime rules. Keep a written account of what translated, what was approximate and what still needs enforcement.
Map the supported rule subset to the target’s actual authorization context.
Validated subsetCarry supported tool allow/deny rules into the gateway configuration.
Validated subsetTranslate configured protections; retain the detector and assessment source.
Signal-basedCarry intent rules into semantic checks without treating a model judgment as certainty.
ProbabilisticKeep transaction rules at the action boundary and decide before dispatch.
Runtime ruleGive platform, security and risk teams one connected answer: what was requested, who authorized it, whether it was dispatched and what outcome was observed.
This trail follows the action selected in the interactive demo.
Authorization is not completion.
Send decision, approval, dispatch and outcome events to Google SecOps and Cloud Logging. Raise selected, actionable SCC findings.
Correlate SCC and Model Armor findings with the agent’s actions. Preserve the source and freshness of each risk input.
Keep the decision record beside a hash-chained control-plane audit log.
06 / Assure
Pin the agent version, policy and acceptance criteria. Run the required checks in Cloud Build or your existing pipeline. Missing evidence is not a pass.
Explore assurance ↗exit 1 · not ready to promoteA deploy step that consumes this verdict keeps v3 under its approved policy. Shown as a demo, not yet the automated stop in your pipeline.
07 / Developers
Work with native policy or OPA/Rego. Review changes in code, test the decision model and carry the same policy into your agents and evidence.
Value now. Portability next.
When the workflow reaches AWS, Azure, Databricks, private infrastructure or SaaS, carry the same business scope, policy ownership and evidence model with it. Standardize governance—not the runtime your teams use.
Start inside Google Cloud. Extend to hybrid infrastructure, MCP and enterprise applications.
Because one cloud does not mean one agent, one session or one business owner. GovernorAI connects the shared limit for a customer case, the person authorized to approve a consequential action, and the record of its outcome. The service-credit example above stays in one Google Cloud project. No second cloud is needed.
No. Keep Google’s identity, access, content and tool-use controls. GovernorAI provides the business-policy workflow around them: shared case state, scoped human authority, native-control mapping and connected evidence. A GovernorAI allow never overrides a native denial.
You could assemble parts of it, and the native controls stay essential: IAM, Model Armor, Agent Gateway and Semantic Governance keep doing their jobs. Then your team owns the shared ledger across sessions and agents, the approval lifecycle, policy mappings, coverage checks and evidence assembly, and keeps them working as those services change. GovernorAI brings them into one supported governance workflow. Start with one real workflow and compare the operating effort, coverage and traceability against your native implementation.
No. Keep your runtimes and tools. Begin with discovery and shadow-mode decisions, review the findings with the policy owner, and enable enforcement for the agreed workflow.
Move AI agents from pilot to governed production
Bring one Google Cloud workflow, its business limit and the person accountable for it. Observe, agree the control and prove the result in a scoped six-week design-partner engagement.