Gemini for government services: Google's five-part agent stack for U.S. agencies launches in preview

• Who: Google Cloud and Google Public Sector, announced by Google Cloud CEO Thomas Kurian on the Google Cloud blog.
• What: Gemini for government services, a preview offering that bundles skills, secure connectors, and partner agents into one governed environment for U.S. federal, state, and local agencies.
• When: Announced October 20, 2026 at the Public Sector Summit '26 in Washington. No general-availability date yet.
• Scope: 5 components — skills for government work; connectors to agency systems; partner agents; a governed control plane (VPC Service Controls, CMEK, Assured Workloads); and an open partner ecosystem.
• Named integrations: 9 systems — the Federal Register, regulations.gov, eCFR, Oracle E-Business Suite, Oracle PeopleSoft, Salesforce, ServiceNow, Google Workspace, and Microsoft 365/SharePoint. Partner agents come from SZNS Solutions (FOIA discovery and redaction); 13 named systems integrators are on the partner list.
• Early users: The State of Missouri and the U.S. Department of Health and Human Services. Federal agencies can keep using Gemini at $0.47 per agency per year until November 15 under the extended OneGov deal.

Google Cloud took its agent push to the public sector on Tuesday. The company announced Gemini for government services, a preview offering aimed at federal, state, and local U.S. agencies, unveiled at its Public Sector Summit '26 in Washington.

The idea is blunt: government work runs on statute, regulation, and agency guidance, across siloed legacy systems, under strict data-sovereignty and compliance rules — conditions a general-purpose agent handles badly. Google's answer is not one model but a bundle: reusable skills written by domain experts, secure connectors into the systems agencies already run, pre-built partner agents, and a single governed control plane underneath it all. It is, in effect, the Gemini agent for work, rebuilt for the way bureaucracies actually operate.

Five pieces, one governed package

The launch has five components. Skills for government work are reusable instructions and context — written by domain experts — that teach Gemini to run a specific task, carrying the statute, regulation, and agency guidance that govern it. Connectors to agency systems link Gemini to case management, ERP, HR, and collaboration systems, plus authoritative federal data sources; Gemini queries each system in place, so no data is migrated, and access follows the identity and permission controls the agency already has. Partner agents combine skills and connectors into deployable agents — from public sector software providers — with a person approving each consequential step. A governed control plane gives security and compliance teams one place to enforce policy, including VPC Service Controls, CMEK, and Assured Workloads, with outputs grounded in source documents and citations. And an open partner ecosystem of integrators and ISVs helps agencies customize without vendor lock-in.

On the security side, the service can run inside Google's Assured Workloads FedRAMP High Data Boundary, enforcing U.S. data residency and personnel controls. Google says agency records, constituent data, and model outputs stay private to the organization and are never used to train or fine-tune its foundation models without permission.

Skills that read like a bureaucrat

The concrete workflow list is where the product gets interesting. Policy and regulatory research answers questions drawn from the CFR, the Federal Register, state administrative codes, and internal manuals — with page and paragraph citations, and an explicit call on whether federal or state rules govern when both apply. FOIA record discovery searches SharePoint, Drive, and email archives for records responsive to a request and assembles a review set for the records officer, work that often takes weeks. Redaction for public release finds sensitive constituent data across a batch of documents, groups findings for batch clearance, and proposes redactions labeled by statutory exemption; after a reviewer confirms, it produces the exemption log and audit-ready copies.

Then there is the money side: vendor bid evaluation checks multi-volume bids against RFP requirements, FAR guidelines, and evaluation rubrics, returning scoring justifications with citations; grant and program expense audits check invoices, receipts, and ledger entries against 2 CFR 200 Uniform Guidance, flagging unallowable costs and missing documentation with line-item citations; document validation at intake classifies pay stubs, W-2s, utility bills, and leases on arrival and flags inconsistent numbers or missing signatures; and plain-language notices turn policy memos into public notices aligned to the Plain Writing Act of 2010 and Title VI language access guidance.

It plugs in where the data already lives

The connector list reads like a tour of government IT. On the federal-data side: the Federal Register (legislative actions, regulations, executive orders as they post), regulations.gov (rulemaking dockets, public comments, agency response deadlines), and the eCFR (grounding answers in the governing regulation, with citations down to the paragraph). On agency systems of record: Oracle E-Business Suite for procurement, payables, and ledgers; Oracle PeopleSoft for HR; Salesforce for benefits, licensing, permitting, and contact centers; and ServiceNow for IT, security, and employee workflows. Productivity connectors cover Google Workspace and Microsoft 365/SharePoint. A BLEN connector searches the Federal Register, regulations.gov, and the eCFR, according to Google's announcement.

On the partner side, SZNS Solutions ships two pre-built agents — a FOIA and public records discovery agent, and a document redaction agent that works inside connected repositories and exports permanently redacted, audit-ready copies without the processing leaving the agency's cloud perimeter. The integrator bench is long: Accenture Federal Services, Clutch, Dark Wolf, Daston, Deloitte, Egen, Game Plan Tech, Knexus, Onix, Resultant, SAIC, Slalom, and SZNS Solutions. Knexus, a Google Cloud Premier Partner, announced it is a launch partner for the service on the same day.

Missouri and HHS are in the first cohort

Google named two agencies working closely with it to develop the offering: the State of Missouri and the U.S. Department of Health and Human Services, and says it is letting more agencies into the preview to gather feedback. Missouri's director of AI and innovation, Tim Marczewski, put the pitch in one sentence: "Our citizens expect the government to move at the speed of modern life, not the speed of paperwork."

Nextgov/FCW's reporting on the launch adds two useful details Google's own blog leaves out: general availability is expected later this year with no set date, and federal agencies can keep using Gemini at $0.47 per agency per year until November 15 under the recently extended OneGov deal.

The context: agents are in government, but not everywhere

The preview builds on Gemini for Government, a broader offering Google introduced in August 2025, as TokenPost notes — so this is the second act: same government motion, now with the agent layer made explicit. And the timing fits the data: Google Public Sector cites survey research with theCUBE and ECI showing 96.3% of surveyed agencies have deployed AI agents in some form, but only 9% have achieved agency-wide adoption (via TechUpScale). The gap between a pilot and an agency-wide deployment is exactly what a skills-and-governance package is designed to close.

Earlier this month Google introduced the universal Gemini agent at its Gemini at Work event; the government offering is its first industry-tailored packaging of that agent, alongside purpose-built solutions for other industries. For a company that operates the full stack — models through application layer — routing cheap document-parsing work to fast models and hard policy analysis to reasoning models is the lever for keeping public-sector spend under control, and Google's blog post says so explicitly.