Join our Newsletter — 33% off our NHI Course
Home› Guides› Gemini Enterprise Agent Platform (formerly Vertex AI Agent…
Guide AI Platforms

Gemini Enterprise Agent Platform (formerly Vertex AI Agent Builder) Explained: Components, Pricing and What Changed

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 15 November 2026 13 min read
On this page

Gemini Enterprise Agent Platform is Google Cloud's managed platform for building, deploying, governing and optimising AI agents, launched on 22 April 2026 at Google Cloud Next 2026 as what Google calls "the evolution of Vertex AI". The rename left a trail of names: Vertex AI, Vertex AI Agent Builder, Vertex AI Agent Engine, Agent Platform, and "Gemini platform", which is the heading on the current product page. The old cloud.google.com/products/agent-builder address now declares the gemini-enterprise-agent-platform page as its canonical URL. This page explains what the platform contains, how its components fit together, which features became generally available and when, how agent hosting is priced and how it relates to ADK and to the separate Gemini Enterprise product. It is written for architects, platform teams and the people who approve cloud AI services. Every fact comes from Google Cloud's product, pricing, blog and release notes pages. Last checked: 2026-11-15.

This is an independent guide from NHI Mgmt Group, based on Google Cloud's public documentation as of November 2026. It is not sponsored or reviewed by Google. Product features change, so confirm settings against Google Cloud's current documentation.

Key takeaways

  • Vertex AI now ships through Agent Platform. Google's launch post says "all Vertex AI services and roadmap evolutions will be delivered exclusively through the Agent Platform, rather than as a standalone service".
  • Four pillars. Google groups the components as Build (Agent Studio, ADK, Agent Garden), Scale (Agent Runtime, Sessions, Memory Bank, Agent Sandbox), Govern (Agent Identity, Agent Registry, Agent Gateway) and Optimize (Agent Simulation, Agent Evaluation, Agent Observability, Agent Optimizer).
  • Governance went GA in July 2026. A 29 July 2026 Google Cloud post describes Agent Identity, Agent Gateway and Agent Registry as "newly generally available", with Memory Bank and Agent Runtime "available for everyone".
  • Agent hosting is metered by resource. Agent Compute lists at $0.085 per vCPU-hour and Agent Memory at $0.009 per GiB-hour, with idle time between turns not billed.
  • It is not the same product as Gemini Enterprise. The product page lists Gemini Enterprise separately, as the place to register and govern agents for employees; Agent Platform is the developer platform underneath.

What is Gemini Enterprise Agent Platform?

Google's product page describes the platform as "our open and comprehensive platform that empowers businesses to rapidly build, scale, govern and optimize enterprise-grade agents grounded in your enterprise data". The 22 April 2026 launch post, by Google Cloud vice presidents Michael Gerstenhaber and Michael Bachman, says it brings Vertex AI's "model selection, model building, and agent building capabilities" together "with new features for agent integration, DevOps, orchestration, and security".

In practice the platform carries two families of capability. The first is the model and MLOps estate that Vertex AI customers already used: Model Garden, custom training, Model Registry, Pipelines, Feature Store, Vector Search, Colab Enterprise and Workbench notebooks, and model evaluation and monitoring. The second is the agent stack, which the launch post groups into four pillars:

PillarComponents named by GoogleWhat they do
BuildAgent Studio, Agent Development Kit (ADK), Agent Garden, Workspaces, native ecosystem integrations, batch and event-driven agents (BigQuery and Pub/Sub), programmatic interface for coding agentsDesign agents visually or in code, start from templates, connect data and tools
ScaleAgent Runtime, long-running multi-day agents, Agent Sandbox, agent-to-agent orchestration, Agent Memory Bank (with Memory Profiles), Agent Sessions (with custom session IDs), WebSocket bidirectional streamingHost agents, keep context and memory, run model-generated code in a hardened sandbox
GovernAgent Identity, Agent Registry, Agent Gateway (with Model Armor protections), Agent Anomaly Detection, Agent Threat Detection, Agent Security dashboard (Security Command Center)Give each agent an identity, index agents and tools, enforce policy on traffic
OptimizeAgent Simulation, Agent Evaluation, Agent Observability, Agent OptimizerTest with synthetic users, score live traffic, trace behaviour, refine system instructions

Key facts at a glance

As of 15 November 2026, from Google Cloud's product page, pricing page, blog and release notes.

FieldValue
OwnerGoogle Cloud
TypeManaged agent and AI development platform
Launched22 April 2026, Google Cloud Next 2026 (Google Cloud blog)
PredecessorVertex AI, including Vertex AI Agent Builder; Google describes the platform as "the evolution of Vertex AI"
Names in useGemini Enterprise Agent Platform (blog, docs, pricing), Agent Platform (short form), "Gemini platform" (product page heading)
Agent frameworkAgent Development Kit (ADK), open source
Models"200+ Google and third-party AI models and tools" in Model Garden (product page)
Agent hosting price$0.085 per vCPU-hour and $0.009 per GiB-hour (list, USD)
Free usage$300 trial credit for new customers (90 days); monthly agent runtime free tier per billing account
Governance GAAgent Identity, Agent Gateway, Agent Registry described as newly GA on 29 July 2026
Release notesdocs.cloud.google.com/gemini-enterprise-agent-platform/release-notes

Plans and pricing

There are no seats or editions: the platform is billed by usage on a Google Cloud billing account. Model calls are billed per model under their own SKUs, and the product page lists Standard PayGo, Flex PayGo (discounted, latency-tolerant), Priority PayGo (higher reliability) and Provisioned Throughput subscriptions. Agent infrastructure is metered on three resources. The table copies the pricing page as of 15 November 2026, in US dollars.

ItemList priceBilling notes from the pricing page
Agent Compute$0.085 per vCPU-hour1-year rate $0.0765; 3-year rate $0.068
Agent Memory$0.009 per GiB-hour1-year rate $0.0081; 3-year rate $0.0072
Agent Storage$0.30 per GiB-monthQuoted for Memory Bank, Sessions and Skill Registry
Agent RuntimeCompute plus memory allocated to the agentRounded to the nearest second; idle time between turns not billed
Code Execution and Computer Use sandboxesCompute plus memoryNo separate sandbox SKU listed
Agent Gateway1 vCPU-hour per 15,000 API calls or authorisation requestsBilling effective 13 July 2026; prorated per hour per project
Memory Bank and SessionsStorage plus 1 vCPU-hour per 3 million reads or 1 million writesBilling started 1 September 2026; memory generation and embedding tokens billed separately
Semantic governance policies1 vCPU-hour per 15,000 response evaluationsBilling "starts later in 2026"; evaluation-model tokens billed separately
Agent SearchAbout $4.00 per 1,000 queries plus storageFrom the product page
PipelinesFrom $0.03 per pipeline runFrom the product page

Worked example: an agent allocated 1 vCPU and 2 GiB of memory for one active hour costs $0.085 plus $0.018, or $0.103 at list price, before model tokens. Free usage: the product page lists a $300 trial credit for new customers valid for 90 days, and a monthly free tier per billing account covering the first 50 hours (180,000 vCPU-seconds) of compute, 100 hours (360,000 GiB-seconds) of RAM and 1 GiB of storage for agent runtimes. The pricing page did not show a separate Agent Registry price as of November 2026.

Models available

Model Garden is the model catalogue, which the launch post put at more than 200 models. Models are added through the release notes; recent dated entries show the range:

Date (release notes)ModelStatus stated
13 Oct 2026Gemini 4 ArgonGA
14 Oct 2026DeepSeek-V4.1-FlashPublic Preview
27 Oct 2026Grok 4.7 (global and US endpoints)GA
9 Nov 2026Claude Fable 5.2GA
9 Nov 2026Meta Muse models (Muse Wonder, Muse Ember, Muse Spark 1.3)Preview
10 Nov 2026Gemini 4 CarbonGA
12 Nov 2026Gemini 4 HydrogenGA

The product page also names Gemma (open model), Imagen (image generation) and Anthropic's Claude model family. The release notes record Claude Sonnet 5.5 (28 September 2026) and Claude Haiku 5.5 (7 October 2026) arriving in Model Garden. For model detail see the Gemini 4 Argon profile.

Features and modes

  • Agent Studio: a low-code visual interface for designing and testing agents and prompts. The launch post says Studio logic can be exported to ADK for code-level work.
  • Agent Development Kit: the open source framework, upgraded at launch with graph-based sub-agent orchestration. See the Google Agent Development Kit (ADK) page.
  • Agent Runtime: the managed host. The 29 July 2026 post says agents can run continuously for up to 7 days for multi-day, asynchronous tasks; the launch post cites sub-second cold starts.
  • Memory Bank and Sessions: long-term memory with structured schemas, and conversation sessions with custom IDs. A dedicated Memory Bank console page, with Agent Runtime memory controls, arrived on 6 November 2026.
  • Agent Sandbox and Workspaces: hardened environments for model-generated code, computer-use tasks, bash commands and file management. The computer use tool reached GA on 30 October 2026.
  • Agent Identity: the launch post describes a unique cryptographic ID per agent; the July post calls it a native IAM type for least-privilege permissions with non-repudiable auditing and automatic lifecycle management.
  • Agent Registry and Agent Gateway: a central library of agents, servers and connections, and a control point that applies IAM conditions, natural-language rules and Model Armor protection to agent traffic.
  • Semantic governance policies: release notes from September to November 2026 add custom denial messages, VPC Service Controls support (Preview, 25 September), user intent matching and cross-project governance.
  • Agent Evaluation and Observability: online evaluation monitors with pre-built, custom Python, LLM-as-a-judge and adaptive rubric metrics, plus end-to-end tracing and dashboards.

Supported IDEs and platforms

The platform is a Google Cloud service used through the Google Cloud console, APIs and SDKs (the product page's sample uses the google-genai Python client). Developer surfaces named by Google include:

SurfaceWhat it is
Agent Development KitOpen source framework in Go, Java, Kotlin, Python and TypeScript, deployable to Agent Runtime
Agent StudioBrowser-based low-code design and testing in the console
Colab Enterprise and WorkbenchManaged notebooks integrated with BigQuery
Google AntigravityA desktop application and CLI to steer and orchestrate agents, using standard Google Cloud credentials
Programmatic interface for coding agentsListed in the Build pillar of the launch post

Agent Gateway is extending beyond Agent Runtime: Agent Gateway for GKE entered Preview on 12 November 2026, and Agent Gateway for cross-cloud, which governs agents on non-Google CNCF Kubernetes clusters, entered Limited Preview on 11 November 2026.

Enterprise and admin controls

Controls follow the Google Cloud model: IAM roles on projects and resources, organisation policies, VPC Service Controls perimeters, Cloud Audit Logs and Security Command Center. The agent-specific additions are Agent Identity (agents as their own IAM principals), Agent Registry (an inventory of agents, tools, servers and skills), Agent Gateway (policy enforcement on calls between agents, tools and models) and the Agent Security dashboard in Security Command Center. A 28 October 2026 release note asks teams to use a single regional Agent Registry instance per Agent Gateway for Gemini Enterprise apps. For the identity model see the Google Cloud Agent Identity guide; for perimeters and logs see the VPC Service Controls Guide and the Google Cloud Audit Logs and Log Sinks Guide.

Data handling and privacy

DataWhere it livesCustomer control
SessionsAgent Sessions storage, billed per GiB-monthCustom session IDs (launch post)
Long-term memoryMemory Bank, including revisionsMemory Profiles and structured schemas; Agent Runtime memory controls in the console
Prompts and responsesSent to the chosen model endpointChoose regional or global endpoints; inference request metadata logging (GA 30 October 2026)
Abuse monitoring signalsEnd-user safety identifier labelsGA 11 November 2026 (release notes)
Agent trafficThrough Agent GatewayModel Armor protection, semantic governance policies, Cloud Trace integration (Preview)

Model-specific data terms, such as provider retention for third-party models, follow each model's terms in Model Garden. Confirm residency needs against the regions listed for each model and service.

Release history

Dated milestones from Google Cloud's blog and release notes, as of 15 November 2026.

DateEvent
22 Apr 2026Gemini Enterprise Agent Platform launched at Google Cloud Next 2026 as the evolution of Vertex AI
13 Jul 2026Agent Gateway billing effective
29 Jul 2026Agent Identity, Agent Gateway and Agent Registry newly GA; Memory Bank and 7-day Agent Runtime available for everyone
1 Sep 2026Billing starts for Memory Bank, Sessions and Skill Registry
25 Sep 2026Semantic governance policies with VPC Service Controls (Preview)
30 Sep 2026Cloud Trace integration with Agent Gateway (Preview); App Topology API GA
30 Oct 2026Computer use tool GA; inference request metadata logging GA
6 Nov 2026Dedicated Memory Bank console page with Agent Runtime memory controls
10 Nov 2026Cross-project governance for semantic governance policies
11 Nov 2026Agent Gateway for cross-cloud (Limited Preview); end-user safety identifiers GA
12 Nov 2026Agent Gateway for GKE (Preview)

Limits and known constraints

  • Names differ across Google's own pages: the product page heading reads "Gemini platform", the blog, docs and pricing say Gemini Enterprise Agent Platform, and older links still use vertex-ai paths. Search documentation under both names.
  • Preview features carry Pre-GA terms: Agent Gateway for GKE, Agent Gateway cross-cloud, Cloud Trace for Agent Gateway and semantic governance with VPC Service Controls were Preview or Limited Preview as of November 2026.
  • Long-running agents have a ceiling: the July 2026 post states continuous runs of up to 7 days.
  • Billing changes landed during 2026: Gateway, Memory Bank, Sessions and Skill Registry billing started between July and September 2026, and semantic governance billing is due later in 2026, so earlier cost estimates may be low.
  • Tokens are billed on top: memory generation, embeddings and governance evaluations consume model tokens billed under model SKUs, separate from the vCPU metering.

Security guidance

This page explains what the platform is and how it is priced. Hardening is covered in the Google Vertex AI Security Guide, which covers IAM roles, service agents and data controls for the platform, and the framework layer in the Google Agent Development Kit Security guide. In platform terms, give each production agent its own Agent Identity with least-privilege IAM, register agents and tools in Agent Registry so the inventory is complete, route agent traffic through Agent Gateway with Model Armor and semantic governance policies, and keep data services inside a VPC Service Controls perimeter. Apply Google Cloud security updates and bulletins promptly. For employee-facing agents in the separate Gemini Enterprise product, see the Gemini Enterprise Security Guide.

Practitioner checklist

  • Map every Vertex AI project, endpoint and Agent Builder or Agent Engine deployment to its Agent Platform name so runbooks and documentation search use current terms.
  • Re-estimate agent costs with Agent Compute, Agent Memory, Gateway, Sessions and Memory Bank charges, plus model tokens for memory and governance.
  • Check the monthly free tier and any 1-year or 3-year consumption rates before sizing budgets.
  • Assign an Agent Identity to each production agent and grant least-privilege IAM roles; avoid shared service accounts across agents.
  • Register agents, MCP servers and tools in Agent Registry, using one regional registry per Agent Gateway for Gemini Enterprise apps.
  • Put Agent Gateway with Model Armor in front of agent traffic, and test semantic governance policies before relying on them.
  • Track Preview features you depend on through the release notes, and confirm Pre-GA terms are acceptable.
  • Turn on Agent Observability and online Agent Evaluation for production agents, and export audit logs to your SIEM.

Standards and references

Related NHI Mgmt Group resources: Google Agent Development Kit (ADK), Google Vertex AI Security Guide, Google Cloud Agent Identity, Gemini Enterprise Security Guide.

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 15 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.