Solutions overview for GDC air-gapped

The Google Distributed Cloud (GDC) air-gapped solutions guides bridge the gap between out-of-the-box product capabilities and production-ready enterprise architectures. These guides equip platform administrators and application operators with conceptual frameworks, reference architectures, and practical implementation steps to run critical workloads in GDC air-gapped environments.

The solutions guides adhere to strict requirements for performance, security, and data sovereignty.

Document types

Solutions are documented across three complementary document types:

  • Reference architectures: Architectural blueprints that outline design principles, system topology, component interactions, and security controls for deploying a solution.
  • Reference implementations: Step-by-step guides for installing, configuring, and verifying the infrastructure, dependencies, and software deployment on GDC clusters.
  • User guides: Day-2 operational guides covering practical administration, service configuration, monitoring, scaling, and ongoing maintenance.

Available solutions

The following sections list the available solution guides for GDC air-gapped, organized by technical domain matching the Google Cloud Architecture Center.

AI and machine learning

  • AI Gateway
  • Elastic Stack
    • Reference architecture: Architecture for deploying the Elastic Stack and integrating it with an LLM service to enable Elastic's generative AI features.
    • Reference implementation: Deploy the Elastic Stack and integrate it with an LLM service to evaluate Elastic's generative AI features.
  • Open weight models
    • Reference architecture: Architecture for deploying and serving open weight large language models.
    • Ollama
      • Reference implementation: Deploy Ollama to serve open weight models on standard clusters.
      • Gemma models: Serve Gemma 4 models using Ollama; select the model and the GPU option in one guide.
    • vLLM
      • Reference implementation: Deploy vLLM to serve high-throughput model inference on standard clusters with GPUs.
      • Gemma models: Serve Gemma 4 models using vLLM; select the model and the GPU option in one guide.
      • Mistral models: Serve Mistral 7B, Mistral NeMo 12B, Mistral Small 3.2 24B, Ministral 8B, Mixtral 8x7B, or Magistral Small 1.2 using vLLM; select the model and the GPU option in one guide.
    • Open WebUI: Deploy and configure Open WebUI to interact with models served by Ollama and vLLM.

Databases

Networking

Security and IAM