Cloud generative AI platform
How to use
Vertex AI.
Building and operating production generative AI and ML systems with Gemini and Model Garden models, grounding, RAG, evaluation, tuning, agents, and Google Cloud controls.
Vertex AI is a Google Cloud platform for generative AI models, agents, retrieval, tuning, evaluation, and ML operations. Building and operating production generative AI and ML systems with Gemini and Model Garden models, grounding, RAG, evaluation, tuning, agents, and Google Cloud controls. This guide covers the whole path in one place: official access, a first session that produces something reviewable, the checks that make output trustworthy, and the permissions worth limiting before you connect real work.
AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.
Access & setup
Find, install, and sign in to Vertex AI
Get into the official Vertex AI experience with the right account and a setup you understand.
- 01
Start at https://cloud.google.com/vertex-ai and confirm the domain before entering account or payment information.
- 02
Availability: Vertex AI is accessed through the Google Cloud console, regional APIs, Google Gen AI and Vertex AI SDKs, CLI, notebooks, and managed services.
- 03
A Google Cloud project with billing, enabled Vertex AI API, approved region and models, least-privilege IAM identity, quotas, and secure application credentials.
- 04
Sign in with the account you intend to keep using, then review plan, data, notification, and permission settings.
- 05
Run one low-risk test task before connecting sensitive files, repositories, or workspace data.
- Use official download pages.
- Review permissions during setup.
- Keep installers and applications updated.
First session
Your first useful Vertex AI session
Learn the interaction loop using a small task with a clear outcome.
- 01
Create an isolated development project with billing alerts, quotas, an approved region, and least-privilege Vertex AI access.
- 02
Define a small benchmark and select a generally available model appropriate for the task and data classification.
- 03
Send a minimal request through the current Gen AI SDK with a pinned model ID, explicit settings, and non-sensitive fixtures.
- 04
Review output support, safety, latency, token use, errors, and retention requirements before widening access or traffic.
- State the outcome before the background.
- Provide the real source material.
- Review the result before expanding the task.
Quality control
Check the quality of Vertex AI output
Establish that an autonomous run did the right thing, not merely that it finished.
- 01
Define what the run should achieve and what it must never touch before granting it a single tool.
- 02
Read the full execution trace: which tools were called, with what arguments, and in what order.
- 03
Verify the side effects directly in the target system rather than trusting the agent's own report of success.
- 04
Confirm failures surfaced as failures — a silent retry loop or a swallowed error is more dangerous than a crash.
- 05
Re-run the same task and compare: an agent that behaves differently across identical runs is not yet production-ready.
- Verify side effects in the system of record, not in the agent's summary.
- Require human approval for any irreversible or outward-facing action.
- Log every tool call so a run can be reconstructed afterwards.
Privacy & permissions
Use Vertex AI safely
Bound what an autonomous system can reach before you let it run unattended.
- 01
Enumerate every tool, credential, and system the agent can reach, and remove the ones it does not need.
- 02
Require explicit human approval for irreversible actions: sending, publishing, paying, deleting, or deploying.
- 03
Run against non-production data until behaviour is predictable across repeated runs.
- 04
Set hard limits on spend, iterations, and runtime so a failure loop cannot run unbounded.
- 05
Treat anything the agent reads from the web or a document as data, never as instructions it may follow.
- Vertex AI capabilities, models, retention conditions, and data locations vary by feature and region. Use isolated projects and service accounts, protect credentials, confirm data governance and zero-retention conditions for the exact feature, enforce quotas and billing alerts, preserve source permissions in retrieval, evaluate safety and grounding, and regression-test every model, prompt, index, and SDK change.
- Follow your organisation's approved-use policy.
- Never treat fluent output as authorization to act.
Core workflows
Step-by-step ways to use Vertex AI for the work it does best.
Each workflow below has its own outcome, steps and review checkpoints.
Compare Gemini and approved Model Garden options under consistent data, safety, quality, latency, and cost conditions.
↓ 02 WorkflowBuild a governed RAG workflow on Vertex AIGround model responses in approved data while preserving access, source identity, retrieval evaluation, and deletion handling.
↓Workflow 01
Benchmark a generative model on Vertex AI
Compare Gemini and approved Model Garden options under consistent data, safety, quality, latency, and cost conditions.
Outcome: Select and integrate a pinned model using evidence from a reproducible evaluation rather than a playground impression.
- 01
Define the use case, data classification, region, model eligibility, output schema, evaluation set, quality thresholds, latency, and budget.
- 02
Create an isolated Google Cloud project and service account with least-privilege Vertex AI permissions, quotas, billing alerts, and approved networking.
- 03
Call candidate models through the current Gen AI SDK with pinned IDs, explicit system rules, bounded generation settings, and non-sensitive fixtures.
- 04
Evaluate factuality, grounding, safety, multilingual behavior, structured output, long context, streaming, errors, latency, and token use consistently.
- 05
Record model and configuration versions, deploy behind application validation and monitoring, and require regression review before any model change.
What good use looks like
- Use one benchmark across models.
- Pin model and generation settings.
- Set quotas and billing alerts early.
Workflow 02
Build a governed RAG workflow on Vertex AI
Ground model responses in approved data while preserving access, source identity, retrieval evaluation, and deletion handling.
Outcome: Return citation-ready answers from an authorized corpus with measurable retrieval and answer support.
- 01
Define source ownership, user permissions, location and retention requirements, expected questions, citations, and an unanswerable-aware benchmark.
- 02
Choose the RAG Engine or approved retrieval components, configure least-privilege identities and encryption, and ingest a small versioned corpus.
- 03
Preserve document IDs and permissions through parsing and chunking, then test retrieval recall and access filtering before generation.
- 04
Generate from only the selected passages with citations and abstention rules, and test prompt injection, contradictions, stale content, and missing evidence.
- 05
Monitor retrieval and answer metrics, version index settings, verify refresh and deletion propagation, and promote corpus changes through staging.
What good use looks like
- Preserve source and permission metadata.
- Test retrieval before generation.
- Verify deletions leave the index.
Official references
Check the current product documentation.
Features, plan limits, availability, and data controls change. These official pages are the starting points used for this guide.