AI Grid Network (AGN)

AI Grid Network (AGN) connects inference providers across Kubernetes clusters, cloud services, and external APIs. Its operator discovers participating sites, tracks provider availability, and publishes routing state for Praxis AI gateways. Gateways select a provider for each request from that local state.

AGN is a technical preview. Start with the scope and limitations before choosing a topology or planning an upgrade. You supply the clusters, inference backends, credentials, gateway deployments, and connectivity between sites.

Architecture

The operator watches Kubernetes custom resources and exchanges membership and provider state with peer operators over SWIM. For Praxis AI gateways, it publishes a routing overlay in a ConfigMap. The overlay-sync sidecar watches that ConfigMap and delivers a file that the Praxis AI gateway can reload without restarting.

For the project-owned grid-gateway, cross-site routing uses a separate serving ConfigMap. With praxisConfig.render.gridServing.enabled, the chart mounts this file and sets…