AUMOVIO centralizes AI access across providers through one Databricks gateway
The deployment connects a production employee app and developer tools to approved models. Stable model names simplify routing, but applications still need testing when models change.
AUMOVIO’s shared AI gateway lets applications use approved model aliases across Databricks-hosted models and external providers, while administrators can change the destination without changing application code. The setup ties access to corporate identity and routes usage into cost-center and project billing, giving the company a common operating layer without forcing workloads onto one model provider. It does not capture request contents by default, and teams still need to test applications when models change.
01
AUMOVIO reported the deployment in a Databricks case study published November 18, 2026; its 1ADP AI App is already running in production.
02
Microsoft Entra ID authenticates applications, while approved permission grants control gateway access and access ends with corporate identity.
03
Payload logging through inference tables is off by default and requires an explicitly approved justification under data protection agreements.
AUMOVIO is standardizing how employees and applications reach AI models, rather than moving every workload to one provider. In a November 18, 2026 deployment case study published by Databricks, the company described a shared gateway for model access, spending records and governance. An employee AI app already runs in production on that infrastructure.
Separate provider integrations and use-specific gateways had complicated operations, cost attribution and governance, AUMOVIO says. A taskforce spanning security, data protection, business areas and IT chose Databricks Unity Gateway as the common route.
A stable name, not an interchangeable model
Applications use approved model aliases beginning with 1adp-*. These names route requests to Databricks-hosted models, Google Vertex AI, Azure OpenAI, AWS Bedrock or other providers. Administrators can redirect an alias to another provider or a newer model without changing the consuming application's code, AUMOVIO says.
That flexibility does not remove application testing. AUMOVIO's catalog lists approved models, what each alias maps to and its consumption tier. Teams no longer build infrastructure to reach a model, but still test how their applications behave with it.
Corporate identity connects access to spending
Applications authenticate through Microsoft Entra ID, AUMOVIO's corporate identity system. A token exchange grants gateway access without separate accounts or distributed API keys. Adding an application requires a permission grant; model access ends when the corporate identity does.
Internal billing: Users assign consumption to a cost center or project. Pipelines combine those assignments with token-usage records and Databricks system tables to charge the right budget. Dashboards show model usage and costs, with attribution in near real time.
Employee access: The production 1ADP AI App gives employees an interface for using AI without writing code, through the shared gateway.
Developer access: An OpenAI-compatible endpoint connects compatible AI coding tools to approved models, while leaving teams a choice of development interface.
Buying the gateway did not remove the operating work
AUMOVIO's build-versus-buy reasoning centered on maintaining provider integrations, security patches and user support. Databricks already ran its 1ADP data platform, so the choice avoided another vendor, network path to review or isolated technology stack.
Engineers still manage configuration as code using Terraform. Provider destinations, fallback order, rate limits and guardrails sit in version-controlled files. Changes undergo review, making configurations reproducible and traceable when production fails.
The next planned phase is sharing agents within 1ADP projects and connecting knowledge sources, web search and other tools through Model Context Protocol connections. AUMOVIO identifies the resulting decisions plainly: what an agent may retrieve, which tools it may call and whose permissions it uses. Those controls sit beyond the model-access consolidation already in place.
Sources
databricks.comHow AUMOVIO centralized AI access with Databricks Unity Gateway
Reader comments
Newest comments first. Replies stay oldest first.