Integrations · Cloud and infrastructure
Govern Vertex AI and every AI call, inside your Google Cloud project.
Compliance first, ready on day one. For Google Cloud teams: one Control room with compliance built in, running inside your Google Cloud project and VPC. Your apps, agents and people keep the SDKs they use, and Vertex AI and Gemini models stay native, in your region.
Why Google Cloud teams choose it
One deployment, not a services project.
Compliance built in
Personal, financial and health data checked before it leaves. Every call recorded, ready for the auditor.
One view of every call
Every prompt, decision, cost and record in one place, across every vendor you use.
Guardrails that cost less
VidaiGuard's small models run on your own compute, with no metering per check.
Freedom to choose
Vertex AI and Gemini models, your own models or another cloud. Try a new model on live traffic without code changes.
Pilots in weeks
One deployment to stand up, on infrastructure you already run.
Inside your account
Containers and PostgreSQL in your Google Cloud project and VPC. Nothing phones home.
Where it sits
Inside your Google Cloud project and VPC.
Who calls it
- Apps on GKE or Cloud Run
- Vertex AI Agent Builder
- ADK agents
- People in VidaiChat
Where it runs
Containers on GKE or Cloud Run, state in Cloud SQL for PostgreSQL, secrets in Secret Manager. Private networking, set up your way.
What it calls
- Vertex AI and Gemini models, native and in your region
- Your own models, so sensitive work stays in-house
- Other providers, only if you allow them
Assemble or arrive
Several services to wire together, or one place, ready on day one.
Assembled from Google Cloud services
- Apigee: API front door and token quotas
- Model Armor: prompt attacks, harmful content, sensitive data
- Sensitive Data Protection: discovery and masking of personal data
- Vertex AI request logging: prompts and responses to BigQuery
- Cloud Monitoring: metrics and alerts
- Cloud Audit Logs: API activity for audit
- Cloud Billing reports: spend by project and label
- Your own glue: routing, chargeback, dashboards
Vidai, one deployment
- Routing, fallback and A/B experiments
- VidaiGuard guardrails, in the path of every call
- Policies per team, agent and application
- Cost ledger and budgets
- An audit record for every call
- One view of every call, across every vendor
- VidaiChat for your people
Agents
Your agents, governed with one setting.
Vertex AI Agent Builder
Tool and action webhooks call models through Vidai. Keep the key in Secret Manager. Each agent or project is registered in Vidai as an application, so its spend appears as its own line in chargeback.
client = OpenAI(
base_url=VIDAI_URL,
api_key=os.environ["VIDAI_API_KEY"],
) Google ADK agents
With Vidai's maintained ADK fork, one environment variable routes every model class.
# one setting for every model class
export ADK_LLM_BASE_URL=https://vidai.your-co.internal Into your tools
Your monitoring, security and BI keep working.
Metrics
Prometheus metrics into Google Cloud Managed Service for Prometheus.
Security events
Signed events to your SIEM, or any system that accepts a webhook, such as Google Chat or PagerDuty.
BI push
Calls, daily cost and the audit log to BigQuery and Looker.
Freedom to move
Start on Google Cloud. Keep the choice.
Two-way translation
Apps keep the OpenAI, Anthropic or Gemini SDK format they use. Vidai translates both ways, so the same code can be served by another cloud's models or your own.
Try a model on a slice of traffic
Send 10% of live calls to a candidate model, compare cost, speed and errors, then switch. No code change.
The Google Vertex AI provider is in alpha: tested end to end, with known gaps. Vidai ships as containers; a Helm chart and Terraform are available to Enterprise customers. Running on Google Cloud managed compute and databases is the intended pattern, with several servers behind a load balancer for high availability. Metrics, event webhooks and BI push come with Enterprise. Product names are trademarks of their owners.
Setup steps for every SDK, provider and integration are in the docs.
How we start together
Start with one team, on your own AI traffic.
- 01
Pick one app or team
Point it at the Control room. One configuration change; nothing else changes.
- 02
Watch for two weeks
Every call priced, checked and recorded. Rules start in observe mode.
- 03
See the evidence
What it cost, what it caught and what it would have stopped, from your own traffic.