AI
GitLab Duo Gains Self-Hosted AI Models via Microsoft Foundry Partnership
GitLab's AI coding assistant can now run against models deployed in customers' own Azure environments, addressing enterprise demands for data residency and infrastructure control.
InfoQ
GitLab has expanded its AI development features to integrate with Microsoft Foundry, enabling organizations to run GitLab Duo against models hosted within their own Azure infrastructure. This move opens access to multiple model families—OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral—while keeping code and requests within organizational boundaries.
Meeting Enterprise Data Requirements#
The integration directly addresses concerns around data sovereignty, regulatory compliance, and network isolation. Rather than routing AI requests through GitLab's managed infrastructure, organizations can now process them entirely within their own environment. This shift gives administrators direct oversight of where source code, prompts, and model responses are processed, and how underlying AI systems are deployed.
How the Architecture Works#
The setup involves three layers: a self-managed GitLab deployment, a self-hosted GitLab AI Gateway, and model endpoints on Microsoft Foundry. The gateway functions as a proxy between GitLab's AI features and the chosen models, decoupling individual capabilities from any single model provider.
A notable flexibility is feature-level model selection—organizations can assign different models to different tasks. Code Suggestions might use one model optimized for programming, while another handles agentic workloads, and a lighter-weight option handles high-volume queries. This granularity allows teams to balance performance, cost, and capability without disrupting workflows.
The Trade-off: Control vs. Responsibility#
Self-hosting brings autonomy but transfers operational burden. Platform teams now own deployment, capacity planning, networking, credential management, availability, and model lifecycle alongside their GitLab infrastructure. Additionally, compatibility cannot be assumed: Microsoft Foundry's catalog evolves faster than GitLab's supported-model matrix, so organizations must validate models across both systems before adoption.
