The exact request
Monitoring reports oper-state down on clab-orch-demo-par-p ethernet-1/2. Which service and customer use it? Is this expected maintenance or a fault, and what is the impact?
The agent resolves a device and interface to a WFO subscription and customer, then checks maintenance flags against administrative and operational state.
Monitoring reports oper-state down on clab-orch-demo-par-p ethernet-1/2. Which service and customer use it? Is this expected maintenance or a fault, and what is the impact?
Hover, focus or tap a node to explore
A2A connects the agents; MCP exposes their tools. Network tools compare WFO intent over REST and GraphQL with SR Linux state over JSON-RPC. Select a node to follow its connections.
Start with a customer ticket, monitoring alert or change request. The planner combines service records, inventory and device evidence to investigate it.
The chat interface sends the request to an OpenAI-compatible endpoint. The A2A proxy bridges that request to the planner.
The proxy connects LibreChat to the agent stack. The incoming chat API and the agent-to-agent protocol are separate interfaces.
One planner delegates to three specialist agents over A2A, then combines their findings. MCP provides tools to the specialists; A2A connects the agents.
Looks up subscriptions, customers and stored service configuration through WFO’s MCP tools. An active subscription or enabled port in WFO does not establish live device health.
Resolves inventory context through netbox-mcp. NetBox supplies device and interface records, which are distinct from the live measurements collected by the network agent.
Compares WFO service intent with current SR Linux state and path measurements. The demo investigates and recommends follow-up; it does not execute repairs or maintenance changes.
The MCP server exposes inventory tools to the inventory agent and queries the NetBox API. It supplies inventory context rather than live network telemetry.
Reads stored service intent from WFO over REST and GraphQL, and queries SR Linux over JSON-RPC. These two evidence sources let the agent detect disagreement between configuration and device state.
Stores service intent, subscription status, customer mapping and maintenance flags. The WFO agent uses MCP; network-mcp also reads the service data over REST and GraphQL.
Provides the network inventory used in the investigation. Device and interface records help identify the affected equipment; live health is checked separately on SR Linux.
The emulated containerlab path connects Amsterdam, Paris and London with no alternate route. Interface state and packet-loss measurements are device evidence; individual samples can differ.
The agent investigates and recommends; people decide and act. Each step keeps a finding tied to evidence and keeps any change under operator control.
No. These are example tickets in an emulated three-node network. Real SR Linux instances carry the lab traffic. Static tickets stand in for a ticket system; automatic ticket-system integration is not demonstrated.
This demo is read-only. It reports findings and recommends next steps. An operator must authorize restoration or maintenance separately; no repair, port change or maintenance flag update occurs in the recording.
The silent English recording is about 70 seconds long. Model waiting has been compressed; this is an edited demonstration, not a benchmark. The final comparison and recap combine recorded WFO intent with independently captured device evidence.
Explore the agent and example stack. The network specialist and containerlab extend the base agent setup shown in the documentation.