GitLab Duo Self-Hosted Integrates Microsoft Foundry for AI Model Deployment

Expanded Self-Hosted AI Capabilities

GitLab has updated GitLab Duo Self-Hosted to include support for models deployed via Microsoft Foundry. This change enables organizations to utilize GitLab's AI development features with models hosted within their own Azure environments, rather than relying on GitLab-managed infrastructure.

The integration covers various model families, including OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral. This offers enterprises increased flexibility in choosing their preferred model provider, deployment location, and managing data flow.

Addressing Data and Regulatory Requirements

This update is particularly beneficial for organizations facing strict data residency, sovereignty, regulatory, or network isolation requirements. By using their own AI Gateway and model deployments, administrators can maintain greater control over the processing of AI requests and responses, as well as the deployment of underlying models.

The architecture involves a self-managed GitLab instance, a self-hosted GitLab AI Gateway, and model endpoints hosted through Microsoft Foundry. The gateway acts as an intermediary, decoupling individual Duo features from specific model providers.

Feature-Level Model Selection

A key aspect of this integration is the ability for organizations to select different models for various GitLab Duo capabilities. For instance, a code-focused model can be used for Code Suggestions, while a different model handles agentic workloads, and a smaller model can be assigned to high-volume tasks. Model deployments can also be modified without altering the GitLab development workflow.

This approach aligns with a broader industry trend of separating AI development tools from foundation models, as Microsoft Foundry itself offers access to models from multiple vendors.

Trade-offs and Responsibilities

While self-hosted AI provides greater flexibility and control, it also shifts more responsibility to engineering and platform teams. These teams become responsible for managing model deployments, capacity, networking, credentials, availability, and the model lifecycle, in addition to the GitLab environment.

Organizations must also verify model compatibility, as Microsoft Foundry's model catalog can change independently of GitLab's supported-model matrix. Model availability in Foundry does not automatically guarantee GitLab Duo compatibility.