Telepath AI Demo Guide

Welcome. This guide lets you explore a live, national-scale telecom operator estate through AI agents, using nothing but MCP tools. There is no software to install on your side beyond an MCP client (Claude Desktop, Le Chat, or any MCP-capable agent). You will connect to a running environment, ask questions in natural language, and watch the agent orchestrate governed, auditable actions across separate carrier systems.

Everything you touch is a simulation built for demonstration. Break things, inject faults, place orders. If you want a clean slate, email demo@akuig.com and we will reset it for you.

What you are looking at

Telepath AI is vendor-agnostic MCP middleware that lets AI agents take governed, auditable actions on real telecom OSS/BSS systems through TM Forum Open APIs. Carriers run their business on siloed systems: one for customers and billing, another for the physical network, another for trouble tickets, another for infrastructure. Those silos rarely talk to each other. Telepath exposes each of them through the Model Context Protocol so an agent can reason and act across all of them in a single conversation, linked only by customer ID, exactly as a real carrier estate works.

The environment you are connecting to is a complete operator network: a national core with residential and corporate accounts, live network topology, billing, SLAs, trouble ticketing, and sovereign GPU infrastructure. Every action the agent takes is logged to an audit trail.

The six systems

You connect to six independent MCP servers. Each represents a different carrier system, and they are linked only by customer ID, the way real estates are.

The three ITSM servers (ITSM Local, Jira, ServiceNow) implement the same trouble-ticket interface and are interchangeable. Your agent code and prompts do not change when you swap between them. For most scenarios, ITSM Local is the smoothest. Try the others when you want to see the same workflow land in a different enterprise ticketing backend.

Connecting

All servers use HTTP streaming transport. In Claude Desktop, add them to your claude_desktop_config.json under mcpServers:

{
  "mcpServers": {
    "bss":          { "url": "https://akuig-bss.ngrok.dev/mcp",          "name": "BSS" },
    "enm":          { "url": "https://akuig-enm.ngrok.dev/mcp",          "name": "ENM Network Manager" },
    "itsm-local":   { "url": "https://akuig-itsm-local.ngrok.dev/mcp",   "name": "Trouble Tickets (Local)" },
    "bare-metal":   { "url": "https://akuig-bare-metal.ngrok.dev/mcp",   "name": "Bare Metal & GPU" },
    "jira":         { "url": "https://akuig-jira.ngrok.dev/mcp",         "name": "Jira Tickets & KB" },
    "servicenow":   { "url": "https://akuig-servicenow.ngrok.dev/mcp",   "name": "ServiceNow Tickets" }
  }
}

Other MCP clients accept the same URLs. You do not need all six connected at once. BSS plus ENM plus one ITSM server is enough for most scenarios. Add Bare Metal for the GPU work.

Once connected, just talk to the agent in plain language. You do not need to know tool names or TMF numbers. The tool descriptions carry their own domain keywords so the agent finds the right one.

The network

The estate spans a national core across three main sites (Norvik, Springfield, Al Marsa) with three more fed from them (Shelbyville, Lishan, Eldby). It is a mix of residential fibre and cable customers and nine corporate accounts across enterprise, healthcare, utilities, retail, education, manufacturing, government, AI, and wholesale.

An interactive map of the topology, with pan, zoom, per-customer highlighting, and tooltips, is at:

https://www.akuig.com/network_topology.html

Open it in a browser alongside your agent conversation. When you inject a fault or trace a customer's tails, you can see exactly where in the network you are working.

Who is who

Residential customers sit in the 8452xxx range. Corporate accounts sit in the 9100xxx range, so you can tell segment at a glance. A few worth knowing:

Scenarios to try

Each of these is a short conversation. Paste the prompts in order, or paraphrase them. The agent will choose the right tools. Everything is reversible; email demo@akuig.com for a reset.

1. Customer 360 across every silo

The cross-silo pitch in miniature. Pick a customer and pull their whole footprint from four separate systems in one conversation.

"Give me a full picture of customer 8452934: their BSS profile and products, the network equipment serving them, any open trouble tickets, and any infrastructure they have allocated."

The agent reaches into BSS, ENM, your ITSM, and Bare Metal, and stitches them together on the customer ID. Try it again with 9100001 (Globex) to see a corporate footprint spanning multiple sites.

2. End-to-end service provisioning

One conversation, four systems, a service brought to life.

"I want to sell fibre to a new address. Qualify the location, look up the customer, place a product order, and activate the service."

Watch the agent qualify the address, check the customer, raise the order, then activate in the network. Every step is a distinct TMF API and every step is logged.

3. Fault injection and impact analysis

See how a network fault ripples into customers, SLAs, and tickets.

"Inject a fault on SPFLD-OLT-01. Then tell me which customers are affected, what their SLA tiers are, and open a prioritised trouble ticket for the worst-hit ones."

"Now clear the alarm and confirm we are back to a clean baseline."

The agent correlates the network alarm against BSS customer records to rank impact by SLA tier, then raises tickets in your chosen ITSM. This is the workflow carriers care about most.

4. The enterprise SLA outage (the money demo)

Globex Industries (9100001) is platinum, with diverse protected tails at HQ. This shows why protection matters in real money terms.

"Check path diversity for customer 9100001."

"Inject a fibre cut on the A tail at Globex HQ. What is the SLA credit exposure right now?"

Because the traffic survives on the backup tail, exposure is zero. Now break the other leg:

"Cut the B tail as well. What is the exposure now?"

With both legs down, the agent shows a live per-hour credit exposure against the platinum SLA.

"Open a priority trouble ticket and clear both faults."

5. Order-to-activate with live inventory

Barley & Sons (9100007) has an order installed but not yet switched on.

"What is the status of Barley & Sons' open order? What is still outstanding?"

The agent finds the order with everything done except activation, and the new site's demarcation point installed but locked.

"Activate it, then confirm the site is live and show me the first invoice."

Activation orchestrates across silos: BSS completes the order and raises the pro-rated first invoice while ENM unlocks and enables the equipment and creates the customer link.

6. Billing dispute and revenue assurance

QwikMart (9100004) is disputing an invoice.

"Show me QwikMart's disputed June invoice, then compare the charges against their actual usage for the month."

The agent surfaces the invoice and finds a discrepancy where two stores were double-billed for their backup connectivity.

"Resolve the dispute with a credit adjustment and confirm the invoice is settled."

7. Sovereign GPU and AI infrastructure

Aurora Cognition (9100005) runs AI workloads on sovereign infrastructure with strict data residency.

"Show Aurora Cognition's infrastructure footprint and how it connects to the network."

"They have an open expansion quote. Show it, then check whether the GPU capacity they want is actually available in the right country."

This is the data-sovereignty story: GPU capacity tied to specific national data centres, with residency treated as a first-class constraint rather than an afterthought. You can also provision fresh capacity:

"Check H100 availability across the sovereign data centres, provision a cluster for customer 8452938, and assign network and storage."

8. Swap the ticketing backend

Run scenario 3 or 4 with ITSM Local. Then disconnect it, connect Jira or ServiceNow instead, and run the exact same prompts. The workflow and your phrasing do not change; only the backend the ticket lands in does. This is the point of a standards-based interface.

Tips

• You do not need to know tool names or TMF numbers. Describe what you want; the agent picks the tool.
• Keep the topology map open in a browser tab. It makes fault and diversity scenarios far more vivid.
• If the agent seems to have gone off track or a server looks unresponsive, start a fresh conversation before assuming something is broken.
• Every mutation is reversible. Explore freely.
• For a full reset to a pristine baseline (alarms cleared, disputes and locked orders restored), email demo@akuig.com.

A note on what this is showing

The headline is not any single scenario. It is that one agent, in one conversation, took governed and auditable actions across a customer system, a network system, a ticketing system, and an infrastructure system that in a real carrier would never speak to each other. That cross-silo orchestration, with a full audit trail and standards-based interfaces, is what Telepath AI provides.

Questions, or want the environment reset? demo@akuig.com.

Appendix: reference data

Residential customers (8452xxx)

The cross-silo key throughout is the customer ID.

Corporate customers (9100xxx)

Corporate accounts use the 9100xxx range so segment is visible at a glance.

Product catalogue

Underlying service specifications: fibre internet (500M/1G/2G), cable internet, 5G wireless, TV, and mobile, plus the enterprise services carried by the corporate accounts.

Sovereign data centres

Server classes range from Standard Compute (64-core Xeon, 512GB) up to GPU Server H200 (128-core, 2TB, 8 GPUs).