Reference architecture
Governed multi-agent orchestration, from chat to cell site Every Copilot conversation flows down through a governed control plane to specialist agents and the network itself — and every finding flows back up with evidence. Nothing reaches the network without policy and, where required, human approval.
Experience
Copilot chat (this demo) NOC dashboards Teams & mobile APIs for existing OSS consoles Governance control plane — Agent365
Agent registry & identity Policy engine & guardrails Approval workflows (human-in-the-loop) Full audit trail & observability Multi-agent orchestration
Planner / router Alarm · KPI · RCA · Capacity agents Change · Knowledge · Remediation agents Comms & reporting agents Data & telemetry fabric
Alarms & events KPIs / PM counters Topology & inventory Tickets, runbooks & configs Network domains
RAN Transport / IP Core & cloud Fixed & edge Why the control plane matters Every agent action is authenticated, authorized and logged against policy before execution. A complete audit trail answers “what did the AI do, and why” for regulators and post-incident review. New agents onboard once into the registry and instantly inherit guardrails, telemetry and approvals.What a request looks like 1 Engineer asks about the Toronto latency spike in chat.2 Planner fans out to Alarm, KPI and RCA agents in parallel.3 Control plane checks each agent call against policy.4 Findings are fused, confidence-scored and cited.5 Remediation is proposed — executed only after approval. Try it in the Copilot