Agent Catalog
The AI agents of MX-PDK ship as blueprints: sets of agents and MCP
servers that work on one topic. Each blueprint is one CompositionModel with
the same name, and each of its agents and MCP servers is one mode of it.
Install a blueprint with brc install model <blueprint>, then list the
modes you want in an AIFabric. Lab 1 walks you through
it, and the labs' terminology explains
these words.
| Blueprint | What its agents do | Catalog |
|---|---|---|
supervisor | Receive your request, send it to the right agents of the fabric, and write one answer from theirs. | Supervisor |
observability | Answer questions about live networks: KPIs, logs, topology, alerts, configuration, packets, Kubernetes health and documentation, and find the root cause of a failure. | Observability |
automation | Design, deploy and update networks, UEs and RIC jobs from plain words, and change the cluster only after you confirm. | Automation |
The Observability and Automation pages end with evaluation results: how well their agents did on a live MX-PDK testbed with four LLMs. Lab 5 shows how to evaluate your own agent the same way.