Models and knowledge
Fleet model and fleet knowledge are versioned artifacts stored in a registry. They are not containers. A microservice binds them when the service needs those files on the Edgelet node.
A model is the weights the service runs. Inference and other Edge AI services load that artifact to produce results.
Knowledge is the documents or dataset the service retrieves. A service uses it for search, retrieval, or a local document bundle. It is not the weights.
Both objects follow the same path. Deploy the registry first, because the artifact record points at a registry id or at the remote or local alias. Then deploy the model or knowledge object with potctl deploy -f. A model and a knowledge artifact may share a name. They are separate lists.
Attach the artifact to the Edgelet node that will run the service. Attachment is how that node is allowed to hold a copy. Then bind it on the microservice. Models bind under spec.models. Knowledge binds under spec.knowledge. Use the artifact name, not a host path.
---
apiVersion: datasance.com/v3
kind: Microservice
metadata:
name: line-monitor/reader
spec:
models:
items:
- name: defect-weights
Distribution is the platform's job. Datasance PoT pulls the artifact from the registry and places it on the attached Edgelet node. You do not copy the files onto the device yourself.
Start the microservice after the artifact is on that Edgelet node. The catalog row says what the artifact is. spec.models or spec.knowledge says this container should use it.
Linking a node does not start a container. The attach step records which Edgelet nodes should hold the artifact. The microservice document is what schedules the service.