Network Digital Twin (NDT) technology enables the replication of entire network infrastructures into fully synchronized digital replicas. This capability is especially timely for the safe integration of AI into critical networks, where model correctness and reliability are essential. In this work, we introduce GhostTwin, the first open-source framework for defining and operating high-fidelity NDTs in a self-contained environment. GhostTwin employs SmartNIC-based hardware emulation to realistically replicate programmable physical networks, capturing network performance metrics such as packet loss and residual bandwidth. The framework supports diverse telemetry sources, enabling scalable and accurate digital twin deployments in a programmable testbed.