Apstra manages devices using IP connectivity over the management network, which is a Layer 3 relationship. Whether you are using on-box agents or off-box agents, the controller (or cluster) communicates with the fabric devices using IP reachability (for example, to exchange management traffic, retrieve discovery state, collect telemetry, and push configuration). This is why Layer 2 adjacency is not required between the Apstra server and the managed switches; the essential requirement is routable IP connectivity and appropriate access (credentials/agent connectivity) to the device management interfaces.
From a platform perspective, Apstra does not need a dedicated physical NIC per managed device. Instead, the server/VM requires connectivity to the management network through a single network adapter, and that interface can route to all managed devices. In a typical data center deployment, the Apstra controller VM sits on a management VLAN/subnet and reaches the entire fabric through routed management. This scales operationally: adding devices does not require adding additional server NICs; it only requires IP reachability and capacity planning for telemetry and agent workloads. Thus, the correct statements are that Apstra uses Layer 3 to communicate with managed devices and that it requires a single network adapter for that management connectivity model.