OpenAI and Databricks: Enterprise AI on your governed data
OpenAI and Databricks integrate enterprise AI tools with governed business data, enabling employees to query data directly via ChatGPT and Codex while maintaining security and governance standards.
OpenAI and Databricks have partnered to allow enterprises to use frontier AI models like ChatGPT and Codex on their governed data within Databricks. Historically, employees relied on analysts or external AI tools lacking company-specific definitions, slowing decision-making and risking errors. The new integration uses Databricks’ Genie Ontology to provide business context, ensuring AI responses align with internal definitions such as revenue or active customer metrics. This enables non-technical users to ask direct questions in ChatGPT, receive accurate answers with visualizations, and act on insights without manual data translation or analyst dependency.
The Genie plugin connects ChatGPT to Databricks, allowing users to query data using company-specific definitions stored in Genie Ontology. Responses are generated with data the user is permitted to access, and each answer includes source links for verification. Beyond answering questions, ChatGPT can recommend next steps, identify stakeholders, and facilitate sharing findings via Slack or email. The plugin operates on the Genie MCP, ensuring consistent governance and security across all interactions.
For engineering teams, Unity Gateway manages the deployment of OpenAI’s Codex, enabling leaders to control access, set budgets, and monitor spending by team. Developers sign in with Databricks credentials while OpenAI keys remain secure within the Gateway, with every request logged for audit purposes. This setup allows enterprises to scale AI coding agents across teams without losing visibility into usage or costs, addressing common challenges in agent rollouts.
Enterprises can also build and deploy custom AI agents using Agent Bricks or the OpenAI Agents SDK within Databricks Apps. Databricks serves recent OpenAI GPT models, including the GPT-6 family and open-weight gpt-oss models, directly alongside governed data under Unity Catalog governance. No external API keys are required, and all interactions adhere to existing security and access controls, ensuring compliance and reducing operational overhead.