GitLab 19.0 expands self-hosted open source model support for its Duo Agent Platform, aiming to bridge the AI capability gap for regulated and isolated environments. This update caters to customers facing data residency, network isolation, and compliance challenges. These constraints often limit access to advanced AI models due to data security concerns. The new feature allows teams to deploy suitable models, even on their own GPUs, within isolated or air-gapped networks. GitLab's selection of open-source models addresses these challenges, enabling local inference and data privacy. The supported open source models include Mistral Devstral 2 123B, GLM-5.1, Kimi-K2.6, and MiniMax-M2.7. Customers can utilize on-premises hardware with vLLM or GPU-enabled virtual machines for self-managed inference. For air-gapped environments, on-premises open-source models are the primary solution. Hybrid deployments are also supported, enabling a mix of self-hosted and GitLab-managed models. Deployment choices depend on the specific environmental needs, such as network access restrictions. The update is accessible to both offline and online license holders, with options for combining self-hosted and GitLab-managed models. Contact GitLab sales for specific deployment needs and further details. This enhancement underscores GitLab's commitment to providing AI capabilities across diverse environments.