AUMOVIO centralized access to artificial intelligence models through a single gateway to simplify oversight and cost tracking, Databricks said in a blog published on Nov. 18.
Teams had previously used different provider connections and gateways for particular tasks, adding complexity as AI use grew, the blog said. Databricks, which sells the data and AI platform used by AUMOVIO, published the account under the title “How AUMOVIO centralized AI access with Databricks Unity Gateway.”
A task force representing security, data protection, business units and IT chose a central gateway as the common route to large language models, the blog said. Such models are trained on large amounts of text to generate responses.
The authors gave no timeframe for the change or figures for its cost or savings. AUMOVIO chose Unity Gateway partly because its existing company data platform already ran on Databricks.
The blog said the setup reused AUMOVIO’s existing Microsoft Entra ID system for corporate identities. Applications can gain gateway access through that system, and model access ends when the associated corporate identity ends, the authors said.
Model approvals, access policies, usage tracking and logs are managed centrally, the authors said. Logging of the data sent to and returned by models is switched off by default and requires explicit approval.
The authors said AUMOVIO also wanted to retain a choice of model providers. A catalog gives applications consistent names for approved models, allowing the company to redirect a name to another provider or newer model without changing the application’s code, they said.
The gateway can route requests to models hosted by Databricks and services from Google, Microsoft and Amazon, among others, the authors said. Teams still need to test their applications against the models they select.
The authors said AUMOVIO built three capabilities on top of the gateway: cost allocation, a model catalog and an employee AI app. The cost system assigns usage to cost centers or projects and displays model use and spending on dashboards, they said.
The employee app runs in production and gives staff access to AI without writing code, the authors said. Developers can use a compatible connection to access the same central models from their own tools.
The authors said the company’s next phase involves sharing AI agents within projects and connecting them to knowledge sources, web search and other tools. That expansion raises questions about what agents can retrieve, which tools they can use and whose permissions apply.
The authors said AUMOVIO remains responsible for its architecture, security decisions and application quality. Running a gateway as a company-wide service also requires continuing maintenance, security work and user support.


