GitLab has expanded its self-hosted AI development platform to work with models deployed through Microsoft Foundry, letting companies operate GitLab's AI coding features on infrastructure they control within their own Azure environments. The integration announced by GitLab supports model families including OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral, giving enterprises greater choice over which model provider they use, where it runs, and what path their data follows. The move matters most for organizations facing data residency, sovereignty, regulatory, or network isolation requirements.
Instead of routing AI requests to GitLab-managed model infrastructure, GitLab Duo Self-Hosted now connects to an organization's own AI Gateway and model deployments. The setup involves three core components: a self-managed GitLab instance, a self-hosted GitLab AI Gateway, and one or more model endpoints running through Microsoft Foundry. The gateway sits between GitLab Duo and the chosen models, decoupling individual Duo features from direct ties to any single model provider. Organizations can assign different models to different GitLab Duo capabilities—using a code-focused model for Code Suggestions, another for agentic workloads, and a lighter model for higher-volume tasks. Model deployments can be swapped out without fundamentally altering the GitLab development workflow.
The report notes an important trade-off with self-hosted AI: giving organizations control over models and infrastructure delivers greater flexibility, but it shifts more responsibility onto engineering and platform teams. Those teams must now manage model deployments, capacity, networking, credentials, availability, and model lifecycle on top of the GitLab environment itself. The report also points out that model availability doesn't automatically mean GitLab Duo compatibility—Microsoft Foundry's catalogue can change faster than GitLab's supported-model matrix, so organizations need to verify compatibility across both platforms before selecting a model.
GitLab's approach reflects a wider shift away from bundling AI development tools and foundation models into a single service. Microsoft Foundry offers access to models from multiple vendors, while GitLab supplies the development and DevSecOps layer around them. GitHub Copilot, by comparison, increasingly supports multiple underlying models but keeps its standard experience tightly integrated with GitHub's managed service; GitLab's self-hosted model approach instead emphasizes controlling the AI infrastructure and network path. Platforms like Amazon Bedrock and Microsoft Foundry provide multi-model infrastructure but aren't substitutes for an integrated DevSecOps platform such as GitLab. This turns the development environment into something resembling a model-agnostic control layer: GitLab manages the developer workflow and AI features, while the organization decides which models power them underneath.
The significance of the announcement extends beyond another model integration, according to the report. As AI embeds itself deeper into software engineering, enterprises increasingly must decide not just which AI capabilities developers use, but where the models run, where source code and prompts travel, who controls the credentials, and which jurisdictions process the data. GitLab's Microsoft Foundry integration addresses part of that problem by allowing the AI model layer to sit within an organization's chosen Azure environment. Enterprise AI tooling is moving toward model choice, deployment control, and data sovereignty, rather than assuming that the best development experience requires a single centrally managed AI provider. The question for enterprises is no longer simply whether to adopt AI-assisted coding, but how to architect the boundaries around it—and GitLab's bet is that those boundaries will increasingly run through infrastructure teams own and operate. For organizations where compliance and control outweigh convenience, that architectural choice may become the default rather than the exception.