Google Cloud · Cloud service
Vertex AI with OrchestrAI
Catalog exported 2026-09-02
Run Vertex AI from chat: datasets, training jobs, model registration, endpoint deployment, and predictions.
OrchestrAI exposes 12 Vertex AI operations: 7 are low-risk (read-only or low-impact), and 5 create or modify resources and run only after you confirm the plan. 3 of them also carry a step-level approval gate.
What teams use it for
ML engineers use OrchestrAI to kick off a Vertex AI training job, check on it later by listing jobs, and cancel it if the loss curve looks wrong. Once a model is trained, the same conversation can register it, deploy it to an endpoint, and send a test prediction to verify the response shape. There is no undeploy or delete-endpoint operation, so tearing down a serving endpoint after a test still requires the console.
Every Vertex AI operation, with its risk level
| Operation | What it does | Risk | Step-level approval |
|---|---|---|---|
Create Vertex AI Dataset |
Create a Vertex AI dataset for training | Low risk | No |
List Vertex AI Datasets |
List datasets | Low risk | No |
List Vertex AI Endpoints |
List endpoints | Low risk | No |
List Vertex AI Models |
List models | Low risk | No |
List Vertex AI Training Jobs |
List training jobs | Low risk | No |
Upload Vertex AI Model |
Upload/register a model | Low risk | No |
Vertex AI Predict |
Make prediction | Low risk | No |
Cancel Vertex AI Training Job |
Cancel a training job | Creates resources | No |
Create Vertex AI Training Job |
Create a training job | Creates resources | Yes |
Deploy Vertex AI Model |
Deploy a model to a Vertex AI endpoint | Creates resources | No |
Delete Vertex AI Dataset |
Delete a dataset | Modifies existing | Yes |
Delete Vertex AI Model |
Delete a model | Modifies existing | Yes |
Risk tiers come from the catalog: low is read-only or low-impact, medium creates resources and is reversible, high modifies existing resources, destructive may lose data. Every plan that creates or changes resources is shown with its cost estimate and waits for your confirmation. Operations marked with a step-level approval pause again on their own step. Destructive operations require a typed risk phrase.
Prompts that work
- Create a Vertex AI tabular dataset called churn-2026 from gs://ml-data/churn.csv in us-central1
- List my training jobs in us-central1 and cancel the one named churn-xgb-run-7
- Deploy the churn-model to the churn-endpoint with an n1-standard-4 machine and send a test prediction
Before anything runs
Every mutation shows its plan, cost estimate, and blast radius, then waits for your confirmation. Destructive operations require a typed risk phrase. Credentials are minted per run through OIDC federation and discarded afterward; nothing you create here is invisible later, because every resource lands in the desired-state ledger where drift is detected and can be converged. Details on the security page.
Frequently asked questions
- Which Vertex AI operations require confirmation in OrchestrAI?
- Creating a training job waits for confirmation because it starts billable compute. Deleting a dataset or a model is rated high risk and also waits for your approval.
- Can OrchestrAI call a deployed Vertex AI endpoint?
- Yes, the predict operation sends an instance to an endpoint and returns the response, and it is a read-only, low-risk action.
- Which Vertex AI operations need an extra approval step?
- 3 operations carry a step-level approval gate on top of plan confirmation: Create Vertex AI Training Job, Delete Vertex AI Dataset, Delete Vertex AI Model. None of them is classed destructive.
Other Google Cloud services
Related integrations
Try it on your own account
Connect your cloud read-only and see your resources, drift, and costs before anything runs. $5 minimum to start. Unused credits refunded in your first 14 days.
Unused credits refunded in your first 14 days.