Google links Looker to Gemini Enterprise for chat data

Google has integrated Looker with Gemini Enterprise, bringing Looker's semantic layer into its workplace AI platform.

The integration lets Looker analysts and administrators publish conversational agents into Gemini Enterprise using the Agent-to-Agent protocol. Employees can then ask questions in natural language across structured company data and unstructured documents from a single chat interface.

At the centre of the change is Looker's semantic layer, which defines business metrics and database logic before a query reaches the underlying data source. That matters because large language models often work well with emails, documents and PDFs, but can produce unreliable results when left to infer how corporate databases are organised.

In practice, when a Gemini Enterprise user asks for a key business measure such as revenue, the request is routed to a Looker agent. The system then generates SQL from version-controlled business logic instead of relying on a model to infer which tables, joins or filters should apply.

Reducing guesswork

Google is positioning the integration as a way to reduce inconsistent answers from AI systems used for business reporting. Standard natural-language-to-SQL models can return different results for the same question when they have to make assumptions about schema relationships or metric definitions.

By contrast, the Looker layer serves as a governed source for those definitions. Teams querying data through Gemini Enterprise should therefore receive answers tied to the same approved logic used elsewhere in the business.

The integration also supports structured databases and unstructured content in one environment. Users can pair hard metrics from databases with contextual information from documents without switching between separate tools.

Security model

Governance and access control are central to the design. Gemini Enterprise does not ingest, replicate or store the underlying database records when a user queries data through a Looker agent, according to Google.

Instead, the system uses a pass-through architecture over the Agent-to-Agent protocol. Users must provide one-time OAuth consent to bind their Gemini Enterprise session to their Looker credentials, allowing existing row-level and column-level permissions in Looker to remain in force.

That approach is meant to preserve the same restrictions users already face inside Looker. If someone lacks permission to see payroll data, regional financial rows or other sensitive records in Looker, the same restriction applies in Gemini Enterprise.

Even when an agent is published into an agent gallery to improve discoverability, established security controls remain in place, Google said. The company cast this as a way to broaden access to AI tools without loosening data governance rules.

Charts and workflows

The release also adds support for native charts inside Gemini Enterprise. Users can ask for visual trends such as monthly sales performance or regional distribution, and the Looker agent can return interactive charts within the chat interface.

That moves the product beyond text-based responses into basic presentation of results in the same workspace. It also reflects growing demand among enterprise users for AI systems that can present information in a form suitable for meetings and internal reporting, rather than only generating narrative answers.

Another part of the release is interoperability with other agents. Looker agents published into Gemini Enterprise can share governed insights with other Google Cloud agents and external third-party agents, allowing structured data to feed wider workflows such as operations, productivity or supply-chain tasks.

This points to a broader strategy around multi-agent orchestration in enterprise software. Rather than treating analytics as a standalone destination, vendors are increasingly trying to embed governed data access into wider AI-driven business processes.

Broader push

For Google, the integration connects two parts of its enterprise software portfolio: business intelligence and workplace AI assistants. It also addresses a persistent problem in corporate AI deployments, where enthusiasm for natural-language interfaces has often collided with the need for consistent metrics, auditability and strict permissioning.

Rivals across the software market are pursuing similar goals by linking AI assistants to semantic models, data catalogues and governed analytics systems. Businesses want conversational access to data, but they also want confidence that a request for a metric such as churn, margin or revenue will produce the same answer across teams.

The latest move suggests Google sees governed analytics as a necessary layer for enterprise AI adoption, especially in large organisations where reporting definitions are tightly controlled and errors can undermine trust in automated systems. It also gives Looker a more direct role in day-to-day employee workflows by bringing its output into a general-purpose chat environment rather than keeping it inside a dedicated analytics tool.

Google said users interacting with Looker agents in Gemini Enterprise can receive "native, interactive data charts" directly inside the chat interface.