GitLab has expanded Duo Self-Hosted so companies can run its AI development features against models deployed in their own Microsoft Azure environment. The integration uses Microsoft Foundry and supports model families from OpenAI, Anthropic, Meta and Mistral.
A self-managed GitLab instance sends requests through the organization’s own AI Gateway to its chosen model endpoints. That architecture is aimed at teams with data-residency, sovereignty, regulatory or network-isolation requirements, because source code, prompts and responses do not have to travel through GitLab-managed model infrastructure.
Administrators can assign different models to different Duo features, such as code completion, agent tasks or high-volume work. They can also replace a deployment without changing the basic developer workflow. The flexibility comes with more operational responsibility: customers must manage capacity, availability, networking, credentials and the model lifecycle themselves.
Access to a model in Microsoft Foundry does not automatically mean GitLab supports it. The two catalogs evolve separately, so teams must check GitLab’s compatibility matrix before deploying. The release gives regulated organizations more control, but it is not a turnkey substitute for operating the underlying AI stack.