NTT DOCOMO put an agentic AI system into commercial use for mobile network fault management on February 4, 2026, across its mobile network services. At MWC 2026 the company published the system's architecture. It analyzes traffic and alarm data from more than one million network elements, from base stations to the core network, and is used for complex faults that previously needed in-depth manual analysis.
Data layer. An integrated data platform pulls operational data from DOCOMO's access, transport and core operations support systems, covering multiple vendors and network domains. Performance data (traffic), configuration data (topology) and fault data (alarms) are held in a graph database, a time-series database and a relational database. According to DOCOMO's head of operation support systems and the AWS team involved, the network topology is modelled as a graph in Amazon Neptune, while alarms and KPIs are federated across Amazon Timestream for InfluxDB and Amazon Aurora. Generative AI summarizes, normalizes and formats unstructured material such as past trouble tickets and methods of procedure into a knowledge base.
Agents. The agents are built with the Strands Agents framework and run on Amazon Bedrock AgentCore. An agent orchestrator coordinates three specialized agents: anomaly detection, root cause analysis and action recommendation. They reach the performance, configuration and fault data, the knowledge base of manuals and tickets, and construction information through Model Context Protocol (MCP) servers, and work on top of foundation models.
Root cause analysis. The topology graph serves as a digital twin of the network. Instead of fixed rules, the agents choose graph algorithms dynamically for each incident; an Amazon Neptune Analytics algorithm pipeline identifies the root cause in under five minutes, according to the AWS team. The recommendation agent draws on knowledge generated from past tickets to propose troubleshooting actions.
Operations workflow. Previously a tier-1 operator confirmed the alarm and a tier-2 operator gathered network element data, analyzed it, decided on actions and issued the report. Now, after alarm confirmation, the agentic AI carries out the analysis and issues the report, and the tier-1 operator escalates to tier 2 with it. DOCOMO says this cuts recovery time for these complex faults by more than 50%.
The system grew out of a network-digital-twin-as-a-graph prototype that DOCOMO and AWS validated on its transport and radio access networks in 2025.