Cloud platformsGoogle Cloud

Build on Google Cloud

Agents grounded in your data,
built on Google Cloud.

Gemini Enterprise Agent Platform, BigQuery and the Google Cloud foundation, engineered into agents that answer from your data and act through governed tools.

Gemini, Claude and Model Garden models feed the Gemini Enterprise Agent Platform, surrounded by Agent Runtime, ADK, Memory Bank, Agent Identity, Agent Gateway and Model Armor, above a knowledge and data layer, with a human approval stepASKED IN GEMINI ENTERPRISEWhy did the North region miss its target?BIGQUERY · VIA MCPSELECT region, week,SUM(revenue) AS revFROM sales.ordersGROUP BY region, weekNORTH · WEEKLYtargetGROUNDED ANSWERTwo large orders moved to thenext quarter, and volume dippedin week 5.[1] orders table[2] quarter reviewAnalyst reviews before it is sharedAGENT PLATFORMADK · Agent Runtime
Talk about your projectExplore the Google Cloud stack
Agent platform
Gemini Enterprise Agent Platform, with agents built in the Agent Development Kit.
Models
Gemini, Claude and other Model Garden models, compared on your tasks.
Knowledge
BigQuery, Agent Search and grounding in your own sources, with Google Search where it helps.
First step
One data-heavy workflow as a Challenge24 prototype, or a review of a Vertex AI project.

Vertex AI, renamed

Same platform, new names.

In 2025 and 2026 Google renamed most of its AI platform. Existing Vertex AI projects keep working. This is how the names map.

BeforeNowWhat it is
Vertex AIGemini Enterprise Agent PlatformThe platform itself. Existing projects and APIs keep working.
Vertex AI Agent EngineAgent RuntimeWhere deployed agents run, with Sessions and Memory Bank.
Vertex AI SearchAgent SearchManaged search over your documents and websites.
Vector Search 2.0Agent RetrievalVector retrieval for agents and RAG.
Vertex AI StudioAgent StudioWhere prompts, models and agents are tried and tuned.
Gen AI evaluation serviceAgent Platform EvalsEvaluation of models and whole agents.
AgentspaceGemini EnterpriseThe workplace app where people use agents.
BigLakeCross-Cloud LakehouseApache Iceberg tables across clouds.

Source: Google Cloud name changes (opens in a new tab)

How we see it

The best agent answers come from data you already trust.

Google Cloud has rebuilt its AI platform around agents. Vertex AI is now Gemini Enterprise Agent Platform, with a runtime, identities, a registry and a gateway for every agent, and BigQuery has its own data agents.

The value still comes from the data underneath. We connect agents to your warehouse, documents and operational systems with clear permissions, so answers are grounded and actions go through tools you control.

An illustrative design

An analyst agent over your warehouse.

A manager asks why a region missed its target. The agent queries BigQuery, reads the related reports and drafts an explanation with its sources. An analyst reviews it before it is shared.

  1. Manager asksGemini Enterprise
  2. Agent plans the stepsADK on Agent Runtime
  3. Queries the dataBigQuery via MCP
  4. Reads the reportsAgent Search · Cloud Storage
  5. Screened and loggedAgent Gateway · Model Armor
  6. Analyst reviewsReview before sharing
  7. Traced and scoredAgent Observability · Evals

What changed on Google Cloud

The release notes that matter.

The releases that change how we design and build on Google Cloud, with a link to each announcement.

  1. Agentspace became Gemini Enterprise

    Google’s workplace agent app was relaunched with a no-code agent designer and a marketplace for agents.

    Read the announcement (opens in a new tab)

  2. Gemini 3 for enterprise

    Gemini 3 arrived for business use, alongside Google Antigravity for agent-led software development.

    Read the announcement (opens in a new tab)

  3. Vertex AI became Agent Platform

    At Next ’26, Vertex AI was renamed Gemini Enterprise Agent Platform, with Agent Runtime, Agent Identity, Agent Registry and Agent Gateway.

    Read the announcement (opens in a new tab)

  4. FinOps agent in Gemini Cloud Assist

    Cost anomalies can now be detected and explained by an agent inside Google Cloud’s own console.

    Read the announcement (opens in a new tab)

  5. Agent Gateway and Observability GA

    Agent traffic can be routed and governed through one gateway, with OpenTelemetry tracing on by default.

    Read the announcement (opens in a new tab)

The Google Cloud AI stack

Six layers. We build across all of them.

Open a layer to see the Google Cloud services involved and the part we take on.

ExperienceWhere people ask, review and approve.
  • Gemini Enterprise
  • Google Workspace
  • Looker
  • Your web and mobile apps

What we build. We place the agent where people already work, in Gemini Enterprise, Workspace or your own application, and design how it shows evidence and asks for approval.

AgentsWhere the plan is made and the steps are run.
  • Agent Development Kit
  • Agent Runtime
  • Sessions and Memory Bank
  • A2A

What we build. We build agents in ADK, run them on Agent Runtime and decide what each one remembers, for how long and why.

Tools and actionsWhere the agent touches your systems.
  • MCP servers
  • Agent Gateway
  • Cloud Run
  • Apigee
  • Sandboxes

What we build. We expose your APIs as MCP tools, route agent traffic through Agent Gateway and run code and browser steps in sandboxes.

Knowledge and dataWhere answers come from.
  • BigQuery
  • Agent Search
  • RAG Engine
  • AlloyDB AI
  • Cloud Storage

What we build. We connect warehouse data, documents and operational records, design retrieval with permissions and freshness, and ground answers in sources people can check.

ModelsWhere reasoning happens.
  • Gemini
  • Claude
  • Model Garden
  • Grounding with Google Search

What we build. We evaluate Gemini, Claude and open models on your tasks, including multimodal inputs, and choose by quality, latency and cost.

Trust and operationsWhere you keep control.
  • Agent Identity
  • Agent Registry
  • Model Armor
  • Agent Platform Evals
  • Security Command Center

What we build. We give agents their own identities, register them, screen prompts and responses with Model Armor and run evaluations on every change.

Our services on this platform

From first idea to production on Google Cloud.

How each stage of our work maps to Google Cloud. Start at whichever stage fits where you are.

01Start

Decide what is worth doing.

  • Challenge24

    A Gemini or Claude prototype over your own data, in three business days.

    Agent Platform · BigQuery · ADK

  • UI/UX Design Studio

    The experience around the agent, designed and tested with your users.

    Gemini Enterprise · Looker · your apps

02Build

Build the system.

03Strengthen

Make it dependable.

  • Enterprise AI Hardening

    Agent identities, gateway policies, Model Armor and evaluations before real users.

    Agent Identity · Agent Gateway · Model Armor

  • Legacy Modernization

    Applications and databases moved with assessments you can defend.

    Migration Center · Database Migration Service

04Run

Keep it healthy.

  • SRE as a Service

    Agent traces and service signals watched by agents, decided by our engineers.

    Agent Observability · Cloud Monitoring

  • AI FinOps

    Model, BigQuery and compute spend attributed and kept within budget.

    FinOps hub · Gemini Cloud Assist

  • AI Governance and Evaluation

    A registry of agents with evaluation records and risk reviews.

    Agent Registry · Agent Platform Evals

05Extend

Add capacity.

  • Engineering teams

    Engineers who know BigQuery, GKE and Agent Platform, inside your team.

    Terraform · Cloud Build · Antigravity

Claude on Google Cloud

Claude on Gemini Enterprise Agent Platform

Claude models, including Claude Opus 5.5 and Claude Sonnet 5.5, are available through Model Garden on Gemini Enterprise Agent Platform. They run under the same IAM, VPC Service Controls and logging as Gemini, so we can test both on your task and use whichever performs better.

How we build with Claude

Under the agents

The cloud foundation we build on.

Compute

Cloud Run · GKE · Compute Engine · TPUs

Serverless, containers or accelerators, matched to the workload and the people running it.

Data and analytics

BigQuery · Cloud Storage · AlloyDB · Spanner · Looker

Analytical and operational data with deliberate access, retention and cost controls.

Events and streaming

Pub/Sub · Dataflow · Eventarc

Events and data streams with explicit delivery, recovery and monitoring behaviour.

Networking

VPC · Private Service Connect · VPC Service Controls

Private paths to models and data, with perimeters around sensitive projects.

Security and identity

Cloud IAM · Cloud KMS · Security Command Center

Access for people and agents, customer-managed keys and AI-aware security posture.

Delivery

Terraform · Cloud Build · Artifact Registry

Repeatable releases for code, containers, agents and infrastructure.

Questions

Common questions.

What happened to Vertex AI?
In April 2026 Google renamed Vertex AI to Gemini Enterprise Agent Platform. Existing Vertex AI projects keep working, and several components have new names, such as Agent Runtime for Agent Engine and Agent Search for Vertex AI Search. We can map your current setup to the new platform.
Gemini Enterprise or a custom agent?
Gemini Enterprise gives your people a ready workplace agent app with a no-code designer. Custom agents built with ADK make sense when the work needs deeper integration, strict evaluation or a specific user experience. Custom agents can also be published into Gemini Enterprise.
Can you work with our BigQuery data?
Yes. We connect agents to BigQuery with scoped permissions and cost controls, and use BigQuery’s own AI functions and data agents where they are the simpler answer.
What does grounding mean?
Grounding ties a response to source information. We ground answers in your approved documents and data, add Google Search grounding where public information helps, and test whether the cited sources support the answer.
How do you choose between Gemini and Claude?
Both are available on Agent Platform. We test them on your real inputs, including documents and images where relevant, and choose by quality, latency and cost. Some systems use both.
How do you secure agents on Google Cloud?
Agents get their own Agent Identity, their traffic goes through Agent Gateway, prompts and responses are screened by Model Armor, and sensitive projects sit inside VPC Service Controls perimeters.
Can you support sovereign or air-gapped requirements?
We assess your obligations against Google’s options, from data boundaries to Google Distributed Cloud air-gapped deployments, and design the architecture around the one that fits.
Can you migrate workloads to Google Cloud?
Yes. We use Migration Center assessments, plan the move in stages and define validation, cutover and rollback with your team.

Explore another platform

Amazon Web ServicesMicrosoft AzureIBM Cloud & watsonx

Let’s start with the work you want to improve.

Tell us about your application, your Google Cloud environment, or the idea you want to explore.

Talk about your project