GitLab Duo Self-Hosted now supports models deployed through Microsoft Foundry, letting organizations run coding assistants within their own Azure environments instead of routing requests to GitLab-managed infrastructure. The integration opens access to model families including OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral, with feature-level model selection—teams can assign different models to Code Suggestions, agentic workloads, or high-volume tasks independently.
The architecture sits three layers deep: a self-managed GitLab instance, a self-hosted GitLab AI Gateway, and model endpoints in Microsoft Foundry. The gateway acts as the intermediary between GitLab Duo and the selected models, decoupling individual features from any single provider. This matters for organizations with data residency, sovereignty, regulatory, or network isolation requirements—requests and responses stay within the chosen Azure environment, and administrators control where source code and prompts travel.
The trade-off is operational weight. Self-hosting shifts responsibility for model deployments, capacity, networking, credentials, availability, and model lifecycle from GitLab onto the engineering and platform teams running the stack. Model deployments can change without breaking the GitLab workflow, but teams must verify compatibility across both GitLab's supported-model matrix and Microsoft Foundry's catalogue before selecting a model. Foundry's offerings can shift faster than GitLab's compatibility list, creating a moving target for validation.
The move reflects a broader shift in enterprise AI tooling. Rather than bundling development environment and foundation models as a single service, platforms are splitting into layers: GitLab manages the developer workflow and AI features, while organizations choose which models sit underneath and where they run. GitHub Copilot remains tightly integrated with GitHub's managed service; Amazon Bedrock and Microsoft Foundry provide multi-model infrastructure but lack the integrated DevSecOps layer. GitLab's approach positions the development environment as a model-agnostic control layer, letting teams make independent decisions about model provider, deployment location, and data path.
The significance extends beyond another model integration. As AI embeds deeper into software engineering, enterprises need to answer not just which AI capabilities developers use, but where models run, who controls credentials, and which jurisdictions process data. GitLab's Foundry integration addresses the infrastructure and sovereignty piece by allowing the model layer to sit within an organization's chosen cloud boundary. For teams with strict data residency or network isolation requirements, this removes the friction of choosing between cloud-managed AI and no AI at all.
If your team needs coding assistants without cloud dependency, map the three-layer stack—GitLab instance, AI Gateway, Foundry models—and verify model compatibility before deployment.