This PoC demonstrates an LLM-based AI assistant that enables users to manage the F5G-A Network Fabric through a single, natural-language interface. The assistant is deployed on the F5G OpenLab testbed at Fraunhofer HHI, where it connects to live management and monitoring components of the network.
The system allows operators to query the network inventory, retrieve connectivity and topology information, perform configuration actions on devices via NETCONF, query telemetry data from the data lakes such as InfluxDB, generate Grafana dashboards for visualizing performance metrics, and access power-monitoring and energy-consumption information, all through conversational interaction. All LLM components run fully on premises within the OpenLab infrastructure to preserve data privacy and guarantee local processing without any cloud exposure.
The PoC is implemented on the F5G OpenLab testbed operated by Fraunhofer HHI. The testbed includes an optical access network with a single OLT and multiple ONUs, interconnected via a PON segment, and a management and analytics environment consisting of InfluxDB, Grafana, NetBox, and power-monitoring systems. The software stack of the assistant follows a modular, layered architecture comprising multiple components as shown in the figures below.
The user interface is a web-based chat application that allows operators to issue natural-language commands and receive responses in real time. The GUI features a conversation panel, agent status indicators showing which agent is currently handling a request (e.g., CONFIG, INVENTORY, DB, ANALYTICS, SUSTAINABILITY), a suggestion bar with example queries, and a debug panel for monitoring system-level events such as WebSocket connections and agent routing decisions.
