Install Seldon Core 2 on Kubernetes using the provided Helm charts, which automatically deploys MLServer and Triton as backing servers
Create a Model resource manifest: specify apiVersion: mlops.seldon.io/v1alpha1, kind: Model, and in the spec provide storageUri pointing to the model artifact location and requirements listing the runtime (e.g., mlserver-sklearn)
Apply the manifest: kubectl apply -f model.yaml — Seldon Core 2 automatically selects an available server based on the declared requirements
Check model readiness: kubectl get model <name> and confirm READY is true
Send inference requests to the Seldon mesh ingress using the V2 inference protocol: POST /v2/models/<name>/infer with a JSON payload
Define a Pipeline resource to chain multiple models for composite inference workflows, referencing each model by name in the pipeline steps
Known gotchas
Seldon Core 2 uses multi-model serving by default — models share server pods rather than having dedicated pods, which improves density but means a misbehaving model can affect co-located models
The storageUri field requires the model artifact to be in a format the selected runtime understands; MLServer expects a model-settings.json file alongside the artifact for custom runtimes
Core 2 requires a Kafka broker for its internal messaging between the scheduler and servers — deploying without Kafka or with incorrect broker configuration causes models to remain in a loading state indefinitely
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp