Businesses can connect OpenAI tools to company data in three ways while retaining controls over access and spending, Databricks said in a blog post published on Nov. 13.
The arrangements let employees query business records, give engineers access to coding tools and help teams build agents that carry out tasks, the company said. Databricks, which provides a data and artificial intelligence platform, titled the post “OpenAI and Databricks: Enterprise AI on your governed data.”
The company described product integrations rather than findings from a study, offering no adoption figures, measured performance gains or research period. Its account tied the partnership’s usefulness to connecting OpenAI tools with company definitions, records and access permissions.
For employees seeking business figures, Databricks said its Genie plugin lets ChatGPT retrieve data and return charts without requiring users to write database queries. Genie uses approved business definitions, including how a company measures revenue or identifies an active customer, to interpret questions.
The company said users can see only information they already have permission to access, and answers link back to sources in Databricks. It described Genie Ontology, which stores business definitions and relationships between data, as the layer supplying that context. A Databricks administrator must enable Genie before users connect it to ChatGPT.
Databricks said ChatGPT Work can also recommend follow-up steps, identify people to involve and share findings through Slack or email. Actions through connected tools require users’ approval, putting a limit on what the system can carry out independently.
For engineering teams, the company said Unity Gateway lets developers use OpenAI’s Codex coding tool while managers track requests and spending. Developers sign in with Databricks credentials, while the gateway holds the OpenAI access key. Managers can set team budgets and select which GPT model handles a task.
Databricks said the gateway records every request, addressing its description of a problem with coding tools spreading across teams without central oversight. The arrangement is intended to give managers visibility into usage, costs and where code and data go.
For teams building their own agents, the company said it runs OpenAI models alongside business data under Unity Catalog, its system for managing data access. Teams can build and evaluate agents in Agent Bricks or write them with OpenAI’s software development kit and deploy them as Databricks Apps.
The company said customers do not need a separate external access key when Databricks serves the model. It also described Lakebase, its managed database, as storing agents’ memory and working state. Databricks said it would publish further posts over the following few weeks with architecture details, setup instructions and lessons from customers.
