CT-AI: Getting Started

CT-AI lets you ask plain-English questions about your calls, devices, and policies — and get data-grounded answers in seconds. This guide covers everything from first-time admin setup to your first query.

Feature Requirements

• Call Telemetry Server Appliance or Docker Deployment
• Call Telemetry Premium or Demo License
• CDR Integration with Cisco CallManager — CT-AI queries your CDR/CMR data. Verify recent call activity in Reporting > Overview before your first query.
• CURRI Policy Integration (optional) — enables AI-powered policy investigation, rule testing, and violation analysis
• CUBE XCC Integration (optional) — enables CUBE gateway health monitoring and SIP quality analytics
• An LLM provider API key — OpenAI, Anthropic, OpenRouter, Azure OpenAI, or a self-hosted Ollama instance
• Organization Administrator permissions to configure the AI provider in Settings > AI Settings

---

Admin Setup

Complete this once per organization in Settings > AI Settings.

Step 1 — Connect a Provider

Provider	Auth	Notes
OpenAI	API key	GPT-5.x, GPT-4.x, o-series reasoning models
Anthropic	API key	Claude Opus, Sonnet, Haiku
OpenRouter	API key	100+ models behind one key — browse openrouter.ai/models
Ollama	Endpoint URL	Self-hosted, on-premises. No data leaves your network
Azure OpenAI	API key + endpoint + deployment	Enterprise Azure-hosted models

Step 2 — Configure and Test

1. Enter credentials (and endpoint/deployment for Azure or Ollama).
2. Choose a model from the dropdown.
3. Click Test Connection — you should see a green success indicator.
4. Click Save Settings.

tip

CT-AI works great on Anthropic Claude Haiku 4.5 — a good balance of agentic tool use, cost, and model intelligence.

Step 3 — Verify

Open AI Chat and run:

Show blocked outbound calls from the last 24 hours.

If you get a data-grounded response with deep links, you're all set.

Runtime and Cost Controls

Setting	What it does
Temperature	Lower = more consistent answers. Higher = broader exploration
Max Tokens	Caps response length and per-query cost
Max Tool Calls per Turn	Prevents over-querying in a single response
Max Tokens per Day	Hard daily token budget per org
Daily Cost Limit (USD)	Hard org-wide spend cap

Privacy and Data Handling

CT-AI can mask phone numbers, caller names, and contact info before they reach any LLM provider. With Ollama, no data leaves your network at all. See Security & PII for redaction settings, audit logging, and retention policies.

---

5-Minute UI Quickstart

Step 1 — Open AI Chat

Click AI in the left sidebar.

Step 2 — Ask a Specific Question

Start with a concrete prompt that includes a time range:

Show me blocked outbound calls from the last 24 hours and why they were blocked.

You should see:

• A streamed response that builds in real time
• Tool activity cards showing which analytics tools CT-AI is using
• Deep links back into Call Telemetry reports and pages

Step 3 — Refine in the Same Thread

Follow up without starting over — CT-AI keeps the conversation context:

Now group that by site and call out the top 3 patterns.

Step 4 — Validate

Click a deep link in the response and confirm the destination page matches the AI summary. This is a good habit to build trust in the results.

---

Prompt Tips

The best prompts follow a simple pattern:

What you want + time range + scope (optional) + output format (optional)

Examples:

"Analyze call quality for SEP001122334455 over the last 7 days, show trend and top failure causes."

"Top 10 least-used phones in the past 60 days."

"Why were calls to extension 4500 failing yesterday?"

For the full list of capabilities, see the Feature Overview.

---

MCP Integration

The Model Context Protocol (MCP) lets AI assistants like Claude Desktop query your Call Telemetry data directly — without opening the web UI. CT-AI exposes all 16 capabilities as MCP tools through the @calltelemetry/ct-ai-mcp package.

Installation

npm install -g @calltelemetry/ct-analytics-mcp

Or run directly with npx (no install required):

npx @calltelemetry/ct-analytics-mcp

Environment Variables

Variable	Description	Example
CT_HOST	Your Call Telemetry instance URL	https://ct.example.com
CT_API_KEY	API key with CT-AI permissions	ct_api_xxxxx
CT_ORG_ID	Organization ID	1

Claude Desktop Configuration

Add to your Claude Desktop config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "calltelemetry": {
      "command": "npx",
      "args": ["@calltelemetry/ct-analytics-mcp"],
      "env": {
        "CT_HOST": "https://ct.example.com",
        "CT_API_KEY": "ct_api_xxxxx",
        "CT_ORG_ID": "1"
      }
    }
  }
}

After saving, restart Claude Desktop. You should see a "CallTelemetry" entry in the MCP tools panel.

Example Queries via MCP

Once configured, ask your AI assistant questions that reference your Call Telemetry data:

"Check my Call Telemetry system — why did the call from 5551234 to 5559876 fail yesterday?"

"Use Call Telemetry to test if rule R-100 would block a call to 19005551234."

The AI invokes CT-AI capabilities via MCP, retrieves live data, and returns results with the same PII redaction policies as the web UI. See the Feature Overview for all 16 available tools.

---

Next Steps

• Feature Overview — Full scope, capabilities, and template queries
• Security & PII — Redaction, retention, and compliance controls