Solution overview

Updated [08/18/2026] — Agentic Network Assistant 2.0

This lab has been rebuilt. Agents now detect injected faults on a live Cisco fabric and remediate them, supervised or fully autonomous. Same URL; existing links and bookmarks still work.

New in 2.0:

  • One-click fault injection from the Lab Buddy Break button (no more paste-into-chat)
  • Five fault scenarios: OSPF loopback withdrawal, OSPF area mismatch, BGP max-prefix shutdown, MTU mismatch, ACL service blocking
  • Full detect → diagnose → propose → approve → verify cycle with a human-in-the-loop approval gate



The WWT Agentic Network Assistant is a multi-agent AI system for network operations, built on LangGraph and running against a live Cisco NX-OS / IOS-XE leaf-spine fabric. It uses the Supervisor pattern: a Triage Supervisor classifies each event and routes it to a specialist agent, with a Human-in-the-Loop approval gate that pauses the workflow before any change reaches a device.

Rather than translating prompts into CLI, the assistant observes the network continuously and acts on what it finds:

Autonomous Drift Detection Detects configuration drift against a golden baseline by polling live device state and consuming syslog, with no prompting required.

Root-Cause Diagnostics A read-only Diagnostics Agent gathers telemetry through MCP tool servers and identifies the failing interface, protocol, and parameter.

Ansible Remediation Behind an Approval Gate A Remediation Agent compiles a formal plan naming a curated, idempotent playbook, then waits for operator sign-off before executing.

Visible Agent Reasoning An Agent Canvas renders the pipeline as it runs — triage, diagnosis, remediation — so the operator can audit how the conclusion was reached.

Tool Access via MCP Model Context Protocol servers expose device access, golden-config comparison, the Ansible runner, syslog/observability, and Cisco documentation RAG.

Lab diagram

Loading