Overview

Compute is how Abridge runs workloads: GKE namespaces for services, virtual machines for sandboxes, and GPU pools for ML. This guide covers what's available and how to request it.

What you can run

• GKE namespaces — the default home for long-running services, wired into the mesh and CI.
• Virtual machines — for sandboxes and workloads that don't fit a container.
• GPU pools — L4 for inference and serving, H100 for training, requested by the job.

Regions and capacity

• us-central1 is primary; us-east4 and us-west4 are available for latency or capacity.
• GPU capacity is pooled — provisioning may queue when a region is saturated.