Google Gemini Enterprise Agent Platform

The Gemini Enterprise Agent Platform is Google Cloud's rebranded and expanded Vertex AI, announced in April 2026. It consolidates model selection, model building and agent building into one stack spanning build, scale, govern and optimise, with Agent Studio, the Agent Development Kit, Agent Runtime, Memory Bank, an Agent Gateway and 200-plus models for enterprises putting production agents in front of real users.

The Gemini Enterprise Agent Platform is Google Cloud's enterprise agent stack, announced on 22 April 2026 at Cloud Next as the rebranding and expansion of Vertex AI. It is organised into four layers. Build offers Agent Studio, a low-code visual builder, alongside the Agent Development Kit, an open-source code-first framework supporting graph-based sub-agent networks, multimodal streaming and sandboxed bash execution; Agent Garden supplies pre-built templates for code modernisation, financial analysis and invoice processing, with batch and event-driven agents running through BigQuery and Pub/Sub. Scale provides Agent Runtime with sub-second cold starts and support for long-running agents that hold state for days, Agent Sandbox for safely executing model-generated code, Memory Bank for persistent curated long-term context, Agent Sessions, and bidirectional WebSocket streaming. Govern is the layer that most distinguishes it: Agent Identity issues a unique cryptographic ID to every agent with auditable action trails, Agent Registry indexes approved agents, tools and skills, and Agent Gateway enforces consistent policy with Model Armor protections against prompt injection and data leakage, backed by agent anomaly and threat detection and a security dashboard powered by Security Command Center. Optimise closes the loop with agent simulation, continuous evaluation against live traffic, observability tracing and an optimiser that clusters real-world failures into instruction fixes. Model Garden exposes more than 200 models including Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3, Gemma 4 and Anthropic's Claude family. Google migrated all SDKs, billing and APIs without breaking changes, and names Comcast, PayPal, L'Oréal, Payhawk, Color Health, Geotab, Burns & McDonnell and Gurunavi as production customers.

Category

AI Agents & Orchestration

Pricing

Usage-based, Contact for pricing

Target Market

CTOs, CIOs, Platform Engineering Leaders, Enterprise Developers, Data Scientists, Security Architects

Deployment

Cloud-only, API-based

Founded

2008

Headquarters

Mountain View, United States

Team Size

500+

• ✓
  Agent Development Kit (ADK)

  Open-source code-first framework for graph-based sub-agent networks, multimodal streaming and sandboxed bash execution in agent workspaces

• ✓
  Agent Studio

  Low-code visual builder that lets non-specialists assemble agents and hand off cleanly to code-first development later

• ✓
  Agent Runtime

  Managed execution with sub-second cold starts, seconds-to-provision agents, and support for multi-day long-running workflows

• ✓
  Memory Bank and Agent Sessions

  Curates persistent long-term memory from conversations and maps session history to internal databases or CRM records

• ✓
  Agent Identity, Registry and Gateway

  Cryptographic IDs per agent, a central registry of approved assets, and policy enforcement with Model Armor injection defences

• ✓
  Agent threat and anomaly detection

  Statistical models plus LLM-as-judge flag unusual reasoning, with real-time malicious-activity detection via Security Command Center

• ✓
  Simulation, evaluation and optimiser

  Tests agents against synthetic interactions, scores live traffic continuously, and clusters real failures into instruction improvements

• ✓
  Model Garden

  Access to 200-plus models including Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3, Gemma 4 and Anthropic Claude

• •
  Customer service agent modernisation

  Comcast rebuilt Xfinity Assistant on ADK with a multi-agent troubleshooting architecture running on Agent Runtime

• •
  Regulated finance workflow automation

  Payhawk's Financial Controller Agent uses Memory Bank and cut expense submission time by more than half

• •
  Agent-based commerce

  PayPal builds multi-agent workflows on ADK and the Agent Payment Protocol for secure agent-initiated transactions

• •
  Enterprise knowledge agents

  Burns & McDonnell combines deterministic rules with probabilistic reasoning in an internal knowledge agent built on ADK

• •
  Clinical eligibility triage

  Color Health's Virtual Cancer Clinic assistant screens breast cancer eligibility using ADK and Agent Runtime

• •
  Governed agent rollout at scale

  Platform teams register every internal agent and enforce one security policy across environments through Agent Gateway

This platform is billed by consumption, not by seat, and that distinction is the single most common source of budgeting confusion because Google ships a separate seat-priced product with a near-identical name. Agent compute — the vCPU time behind Memory Bank, Sessions and the Skill Registry — runs about $0.085 per vCPU-hour, stored session events and memories cost $0.25 per 1,000, and model tokens are billed on top at wildly different rates depending on tier, from roughly $0.10-$0.50 per million input tokens on flash-class models to $1.25-$12.50 per million on Pro-class, with output tokens several times the input rate. Every consumption resource carries a monthly free allowance — the first 50 vCPU-hours of compute, 100 GiB-hours of memory and 1 GiB-month of storage are free — and new Google Cloud accounts generally get a $300 / 90-day trial credit that applies here. Gemini Enterprise, the seat-priced agent and enterprise-search product, is a different SKU starting around $21 per seat per month for Business and $30-plus for Standard and Plus; do not budget the two as one line item.

This page was written from 6 sources, 3 on domains other than cloud.google.com.

1. 1.cloud.google.com — introducing gemini enterprise agent platformvendor
2. 2.cloud.google.com — gemini enterprise agent platformvendor
3. 3.github.com — 1857
4. 4.medium.com — vertex ai is dead long live gemini enterprise agent platform
5. 5.hpcwire.com — google unveils gemini enterprise agent platform
6. 6.cloud.google.com — offeringsvendor