GitLab Duo Self-Hosted Adds Microsoft Foundry Models for On-Prem AI Coding

GitLab has widened the model options for GitLab Duo Self-Hosted, adding support for AI models deployed through Microsoft Foundry. The change, reported by InfoQ, means customers can run GitLab's AI-assisted development features against models hosted inside their own Azure environment rather than sending requests to GitLab's cloud. For regulated teams that have balked at shipping source code to a third-party AI vendor, that is the difference between using Duo and not using it at all.

GitLab Duo Self-Hosted already allowed organizations to run the assistant against their own model infrastructure, but the supported paths were narrower. With Foundry in the mix, an enterprise can pick a model from the Azure catalog, deploy it in its own subscription, and wire GitLab to it without standing up separate serving hardware. The setup keeps prompts, completions and repository context inside the customer's network boundary, which is where most security reviews for AI coding tools stall.

The move lands in a crowded stretch for enterprise coding assistants. GitHub Copilot, Cursor and Amazon Q Developer have all pushed deeper into corporate accounts, and the differentiator increasingly is not the model but where it runs. GitLab's bet is that a meaningful slice of large buyers, particularly in finance, healthcare and government, will trade some model flexibility for control over data residency and audit trails. Microsoft Foundry, formerly Azure AI Foundry, gives those buyers a familiar procurement and identity layer to lean on.

There is a practical upside for platform teams as well. Running inference through an existing Azure agreement can fold AI spend into a cloud commitment the company has already negotiated, and it lets security teams reuse the same networking, logging and key-management controls they apply to other workloads. That matters because self-hosted AI deployments often fail on operations, not on model quality. The fewer new systems an admin has to patch and monitor, the more likely the rollout survives past a pilot.

What GitLab has not said is how much of Duo's feature set is fully supported on Foundry-hosted models, or whether every model in the Azure catalog qualifies. Model choice and feature parity have been the awkward seam in self-hosted AI assistants, since some capabilities depend on specific model behavior or context windows. The next thing to watch is the support matrix GitLab publishes, and whether customers report that code suggestions, chat and merge-request summaries all behave the same as they do on GitLab's own hosted models.