Rank 6 of 7
Google Gemini Enterprise
What is Google Gemini Enterprise?
Google offers two products that a platform leader has to keep apart. The employee product is the Gemini Enterprise app, and Google's FAQ frames it as an advanced agentic platform meant for the whole workforce. Builders get a separate product, and Google says the Agent Platform is the evolution of Vertex AI.
In this article
Google Gemini Enterprise: a suite to buy, and an agent platform to build on
Google offers two products that a platform leader has to keep apart. The employee product is the Gemini Enterprise app, and Google's FAQ frames it as an advanced agentic platform meant for the whole workforce1. Builders get a separate product, and Google says the Agent Platform is the evolution of Vertex AI2. The two must be purchased separately3. In build-versus-buy terms, the app is the "buy" and the Agent Platform is the "build", and a Google Cloud customer can do both. Gemini Enterprise ranks sixth on this site.
What the app does
Google's docs fold three jobs into the app, finding things on the intranet, answering as an assistant, and hosting agents4. Search gives permissions-aware access to enterprise information5, and the product ships prebuilt connectors to commonly used third-party applications6. A companion product, Gemini Notebook Enterprise, handles research and summarizing across a team's sources7.
Any employee can make a custom agent without code, using what Google calls a workbench8. Two agents arrive ready-made, Deep Research for digging into hard questions on a user's behalf9 and, still in preview on Standard and Plus, Data Insights10. Agents made by Google partners can be picked up from a marketplace inside the product11.
This is where the build side joins: agents coded with the Agent Platform or ADK can be published into the app by Standard and Plus customers12, and agents living elsewhere can plug in over A2A13. That is the "both" path: engineers build specialized agents in code, and employees reach them in the same app they use for search and chat.
Editions
Editions are where buying gets complicated, because Google offers several, each tuned to a different size and type of buyer14. Even the documentation is divided, since Business has its own support center while the main docs serve the other editions15. The Business edition caps out at 300 users, a size Google pitches at small companies and departmental teams16. Moving to Standard or Plus removes the seat ceiling and adds bigger quotas, outside agents and Gemini Code Assist17, with a Frontline add-on available for frontline workers18,19. Public bodies and schools are steered toward special editions such as Gemini for Government20.
Setup and licensing
Under the hood, the commercial plumbing is Google Cloud's own. Procurement happens in two moves: the subscription is purchased on a billing account21, and an admin then allocates its licenses by project and location22. Licenses stay bound to that project and location23, just as the subscription stays bound to its billing account24. Licenses can be assigned automatically or by hand25. Seats can be committed monthly or annually26, and a single subscription cannot exceed the seat cap that the Cloud console enforces27. Before licenses go out, IAM has to be in order, since each recipient needs at least the Gemini Enterprise User role28.
Business is simpler. Sign-up needs only a work email address29, and there are neither project numbers nor project IAM roles to administer30. Business, Standard and Plus each come with a 30-day free trial31,32, and Plus can be trialed with no sales contact at all33.
For a platform team already on Google Cloud, this setup will be familiar. For one that is not, the project, location and billing account model is a new layer to learn before the first agent runs.
Integrations
Out of the box, Business links to Google Workspace and to rivals' tools, Microsoft 365, HubSpot and Notion among them34. On the enterprise side, the docs give four examples of third-party connectors that come prebuilt: ServiceNow, SharePoint, Jira and Confluence35. Google's own suite is covered with live syncing of Gmail, Drive, Calendar and Groups36. Analytics data can be reached through a federated connector to BigQuery37.
Google does not publish a total connector count on the pages opened. The named list centers on documents, collaboration and IT service tools. Nexus publishes a figure above 4,000, spread over CRM, ERP, messaging and custom API connections38.
Security and certifications
Google's certification list is long. The compliance documentation, scoped to Standard, Plus and Notebook Enterprise39, is specific: its list runs from HIPAA and FedRAMP for regulated US work through the SOC family, ISO 27001 and payment-card standards40. A team in a regulated US industry will find this the most direct answer here on HIPAA and FedRAMP.
Encryption controls are deep. Key control covers CMEK41, delivered through Cloud KMS on Standard and Plus42, with EKM and HSM-backed keys as further options43. CMEK and Access Transparency are both unavailable when the global region is chosen44, which makes region choice a security decision.
Google commits to keeping prompts, outputs and connected company data out of foundation model training45, and to walling each customer's data inside its own Cloud project46. Access is controlled by Cloud IAM47, document-level access controls are enforced48, and Model Armor is included in all editions49. Prompt injection is handled by Model Armor, which screens both incoming prompts and tool output in real time50. Where connectors reach outside Google, the advice is to authenticate them through Workforce Identity Federation51. Sensitive Data Protection, applied to connectors, prevents flagged data from entering the index in the first place52.
As a hosted service, the app gives customers no way to insert network controls of their own53. Connectors to non-Google systems travel over the public internet to those systems' endpoints54. Inside a VPC Service Controls perimeter, an org policy has to whitelist each Data Cloud connector55; until it does, no data store can be created56. In federated query mode, data in Data Cloud connectors stays where it is57.
Governance
Governance is meant to be centralized, and the FAQ claims the app gives administrators one place to oversee agents across all teams58. On the enterprise editions, administrators work through Cloud IAM, use Workforce Identity Federation for SSO, and rely on each source system's existing ACLs to decide what a user may see59. Business-edition admins manage a shorter list: adding and removing members and deciding team access to connectors60. Admins can cap consumption per user61.
On the developer platform, agents are treated as principals, each issued its own SPIFFE ID62 and protected out of the box by Context-Aware Access63. Network control for agents is concentrated in Agent Gateway, the single entry and exit point the docs describe64, which suits a security team that wants one place to inspect. Tool-use policy for agents is expressed as ordinary sentences instead of code65. Sensitive Data Protection works across all editions66.
That identity and gateway model is a strong foundation for a platform team that wants to run many agents under one policy. It does assume the team builds on Google's stack.
Region, retention and training
In the editions table, residency is offered in three settings, us, eu and global67, and the FAQ confirms that Standard and Plus customers choose their region68. The editions table lists Cloud Audit Logs and Access Transparency among the enterprise features, both useful to an audit team69. The no-training rule holds across editions, for enterprise data70 and for what Business users type and receive71. Business itself lives on the web at business.gemini.google72.
According to the security overview, data a user asks to remove is gone within 60 days73. Capacity planning should treat storage and indexing as one pool per project and location, shared across users74. Developer-side data gets the same protection: Google's zero-retention page promises the same for Agent Platform models, with training or fine-tuning ruled out absent prior permission or instruction from the customer75. Some retention does apply, short-term storage for Models as a Service76 and up to three days of logs when answers are grounded in Google Search77, while logging of requests and responses stays off unless enabled78. Data protection scanning works in the global region and in the EU and US multi-regions79.
Pricing
Only Business has a list price, starting at 21 dollars per seat per month on its own page80. The app's invoice is computed per seat each month, which means every person using it needs an active license81. We could not find a Standard or Plus seat price, which is not publicly stated on the pages we opened; the pricing link itself was broken on our check date82.
Developer usage is metered instead, charged as it is consumed83. Release notes confirm that pay-as-you-go carries no license fee84, and the licensing docs add that the bill arrives monthly for actual consumption85. Calls to partner models are metered the same way86, and online billing accounts can now prepay with credits87. Notebook Enterprise has its own subscription route, with a ceiling of 5,000 licenses88.
A platform leader costing a Google build therefore has three lines: app seats, Agent Platform consumption and the engineering time to build agents on it. Nexus removes that third line as a separate cost, since its charge per role automated already covers the consultants who build89.
Stated limits
Daily allowances differ by edition, 160 assistant queries per license on Standard90 against 200 on Plus91. Apart from storage and indexing, those quotas are shared across the users of an edition92. Customers counting on Gemini Code Assist should note the date it stops being bundled into existing Standard and Plus plans, March 31, 202793. The scanning has blind spots: a data store created the older way escapes the content rules, whether its source is a bucket, a BigQuery table or a website94, and encrypted Office files pass through without a content scan95.
Who it suits
Google aims the app at knowledge workers, naming the commercial, finance, people and technical departments as its main audience96. Google's examples are everyday office work, such as a recap of last year's top campaigns97, inventory tracking with automatic purchase orders98, customer emails drafted with Search context99, and new-hire onboarding through Notebook100.
Build, buy or both on Google Cloud
The buy path is short. A team can open a Business edition account with a work email29, pay from 21 dollars per seat per month80 and try it free for 30 days31. Staff get search, assistants and a no-code agent workbench8 without a project to configure30. For a group of up to 300 people16, that may be all that is needed.
The build path runs through the Agent Platform. It is billed on consumption83, with partner model usage also pay-as-you-go86. Your engineers write agents with ADK or the Agent Platform and publish them to the app12, bring in agents from other stacks over A2A13, and govern them with an identity issued to each agent62, a single gateway64 and tool rules written in plain words65. That is a capable stack for a team prepared to build on it, and the work of designing, coding and maintaining each agent stays with that team.
Doing both is the pattern Google designs for: the app as the front door, the Agent Platform behind it, and partner agents from the marketplace in between11. The cost has three parts, seats, consumption and engineering time, and only the first two appear on Google's bills.
A platform team should also plan for the network edges. Custom network controls cannot be added to the SaaS app53. Any connector to a non-Google system will leave Google's network for a public endpoint54. Teams inside a VPC Service Controls perimeter must allow-list Data Cloud connectors by org policy before any data store can exist55,56. CMEK requires a specific region44. Each is manageable, and each is a task for the platform team.
Nexus approaches the same estate from the other side. Its layer runs over the systems a company already has101, its model provider can be changed102, and its consultants assemble the first agent together with your staff, sparing your engineers that work89. A company that wants Google models inside a cross-system workflow without staffing the build could run that workflow through Nexus and keep Gemini Enterprise for search and assistants.
Verdict
Gemini Enterprise is the right shortlist entry for an organization that already runs on Google Cloud and wants one place for search, assistants and custom agents. It has a long certification list, including HIPAA and FedRAMP40, regional residency67 and a well-defined agent identity model62. Sixth place reflects four things: two separate purchases3, no public price for Standard or Plus82, a named connector list weighted toward collaboration tools35, and custom agents that your own engineers must still write on Google's stack. Nexus ranks above it for teams that want agents across many systems built with them, on a committed timeline89.