Agents that turn fault situations into ranked root causes and a recommended action. The NOC decides. We designed and built this for Nokia inside a Tier-1 Nordic NOC, measured it in the MVP, and the operator funded production on the results.
At one Tier-1 operator, about 1,000 fault situations a quarter each cost 2 to 4 engineer-hours of manual analysis. Senior engineers spend their day on triage, and the backlog never clears.
An agent that acts on the network with output nobody can reproduce will not be trusted, and should not be allowed. Operators need every step auditable before an agent gets near a ticket.
Public models are ruled out on confidentiality. Inference has to run inside the operator's own platform, with the same identity, tracing and access controls as everything else.
A deterministic four-step flow (retrieve, annotate, recommend, format) turns each fault situation into ranked root causes and a suggested action. Open-loop by design: the agent recommends, the engineer decides. Built on the 4G/5G Packet Core first, with more network functions following.
Agent tools over MCP and an A2A gateway for production. We designed the A2A integration with BMC HelixGPT's Best Action Plan: submit and poll, an agent card, an audit record and replay, with the REST to A2A transition tied to the platform release.
Open-weight models behind a LiteLLM gateway on the operator's own Kubernetes, with OpenTelemetry tracing end to end and OAuth2 at the gateway. For sites with no outside connection, we have sized an air-gapped GenAI platform down to the GPU plane.
A target architecture and a phased roadmap mapped to TM Forum Autonomous Networks levels, from single-agent delivery to a multi-vendor agent fabric. We shaped an agentic Centre of Excellence proposal for a vendor and its operator customer, and produced the figures used with product leadership.
The operator's NOC handles about 1,000 fault situations a quarter on its 4G/5G Packet Core, received through BMC Helix. Each one cost 2 to 4 engineer-hours of manual root-cause analysis, and public models were ruled out on confidentiality.
For Nokia, Atsky designed and built the multi-agent pipeline, ran the design workshops across vendor, operator and ISV, and wrote the solution description all three signed. Six Packet Core use cases, three on each network element type, in six months.
Measured in the MVP across 250 historical situations. The operator funded the move to production on these results. The work is part of a TM Forum Catalyst on multi-agent intelligence for packet core operations.
Read the case note →You see the architecture and the evidence before you commit to a build.
We agree with the NOC what will be measured and against what: alarm-noise suppression, time to root cause, acceptance of the agent's notes. The baseline comes from your historical situations, not from a slide.
Design workshops across vendor, operator and platform teams. Agent flow, tool boundaries, the ITSM integration, the inference platform and the identity model, with every open question on a register that has an owner.
A phased plan for the MVP with the first use case, the production path, and what it will cost to run. If the evidence says stop, we say stop.
Delivery with a small senior team, through your vendor or MSP partner or directly. At the reference operator the MVP took six months for six use cases, and production followed on the measured results.
We integrate with the ITSM, AIOps and network platforms you already run, and the agents live inside your tenancy.