AI Platform
Two halves, one platform. Serving runs open-weights models on the cluster’s own GPUs behind an OpenAI-compatible gateway. Agents put models to work: a labelled issue becomes a sandboxed agent run that opens a pull request, under its own identity.
Smana/ai-native-ref and publish on
airef.ogenki.io. This platform pulls it as one
signed, version-pinned Flux artifact (ADR-0057).What runs where
The source ships suspended (flux/sources/ocirepo-ai-native-ref.yaml); each cluster decides what to resume.
The Agent Factory runs on aws-0, the one live cluster for this session: gcp-0 is torn down, and the ownership
fence .github/agent-factory.yaml (factory: aws-0) names the only factory allowed to act on this
repository.
| Umbrella | aws-0 | gcp-0 (torn down; Git state) |
|---|---|---|
OCIRepository ai-native-ref | live | live |
ai-gateway (CPU; the others depend on it) | live | live |
agent-platform (the Agent Factory) | live | live, fenced out |
llm-platform (GPU serving) | suspended | suspended |
To turn it on for a cluster, un-suspend the source with a patch in its
clusters/<cluster>/flux/flux-sources.yaml (as both clusters do), then set spec.suspend: false in
clusters/<cluster>/<umbrella>.yaml, ai-gateway first. A flux resume alone lasts only until the
next sync sets Git’s value back.
What this platform guarantees the AI platform (Flux names, substitution variables, Gateways, namespaces, OpenBao mounts) is the host contract.