GitLab has expanded GitLab Duo Self-Hosted to support AI models deployed through Microsoft Foundry, letting organizations run GitLab's AI development features against models hosted in their own Azure environment. The integration supports OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral model families, giving enterprises control over model provider, deployment location, and data path. The architecture uses a self-managed GitLab instance, a self-hosted GitLab AI Gateway, and Microsoft Foundry model endpoints, enabling feature-level model selection for different Duo capabilities. This shifts responsibility for model deployment, capacity, networking, and credentials onto engineering teams, and requires verifying compatibility between Foundry's catalogue and GitLab's supported-model matrix. The move reflects a broader industry shift toward model-agnostic control layers, contrasting with GitHub Copilot's tighter integration with GitHub's managed service.
Questions this post answers
Does GitLab Duo Self-Hosted support models deployed through Microsoft Foundry?
Yes, GitLab Duo Self-Hosted now supports models deployed through Microsoft Foundry, letting organizations run GitLab's AI development features against models hosted in their own Azure environment. Supported model families include OpenAI GPT, Anthropic Claude, Meta Llama, and Mistral, giving enterprises control over model provider, deployment location, and data path. Teams weighing self-hosted AI options for coding tools can track integrations like this one on daily.dev.
What are the main components of a GitLab Duo Self-Hosted architecture using Microsoft Foundry?
The architecture has three parts: a self-managed GitLab instance, a self-hosted GitLab AI Gateway, and one or more model endpoints hosted through Microsoft Foundry. The gateway sits between GitLab Duo and the selected models, so individual Duo features aren't tied directly to one model provider, and different models can be assigned to different Duo capabilities. Architects planning AI infrastructure can compare setups like this one on daily.dev.
What are the trade-offs of using self-hosted AI models with GitLab Duo instead of GitLab-managed infrastructure?
Self-hosting shifts responsibility for model deployments, capacity, networking, credentials, availability, and model lifecycle onto engineering and platform teams, in addition to managing the GitLab environment itself. It also means model availability in a catalogue like Microsoft Foundry does not guarantee GitLab Duo compatibility, since Foundry's catalogue can change faster than GitLab's supported-model matrix. Engineers deciding between managed and self-hosted AI tooling can follow this trade-off on daily.dev.