Skip to content

Agent-MO (Memory Observer)

GitHub

Overview

Agent-MO is an intelligent network diagnostic agent designed to analyze memory utilization, detect memory leaks, and forecast future memory trends across network devices.

It works at device, service, and network levels and supports both standard devices and Cisco Catalyst Center–managed devices. The agent follows a strict, tool-driven workflow to ensure consistent, accurate, and explainable diagnostics.

Key Capabilities

  • Real-time memory utilization analysis

  • Memory leak detection [Mann-kendall]

  • Device-level

  • Network-level
  • Service-level
  • Memory forecasting using XGBoost
  • Catalyst Center–aware analysis
  • Guided troubleshooting recommendations

Agent Definition

agent = Agent(
    name="Agent-MO",
    description="An expert network diagnostic assistant specializing in memory utilization analysis, leak detection and forecasting.",
    instructions=...
)

Agent-MO strictly uses tool-based execution (no manual SQL or assumptions).

Tools Used

Agent-MO is powered by the following tools:

tools = [
    get_device_info,
    get_memory_utilization,
    predict_device_memory_xgboost,
    datastore,
    device_level_memory_leak_analysis,
    network_level_memory_leak_analysis,
    service_level_memory_analysis,
    get_catalyst_center_devices,
    memory_leak_analysis_and_detection_for_catalyst_center_devices,
    get_memory_utilization_for_catalyst_center_devices,
]

Tool Responsibilities

Tool Purpose
get_device_info Resolves IP/Hostname → MAC address
get_memory_utilization Current memory snapshot
service_level_memory_analysis Service-specific memory usage
device_level_memory_leak_analysis Leak detection for a single device
network_level_memory_leak_analysis Leak detection across all devices
predict_device_memory_xgboost Memory usage forecasting
get_catalyst_center_devices Fetch Catalyst Center–managed devices
memory_leak_analysis_and_detection_for_catalyst_center_devices Leak detection for Catalyst devices
get_memory_utilization_for_catalyst_center_devices Utilization for Catalyst devices
datastore Troubleshooting guidance

Operating Rules & Constraints

  • Input Normalization: If a user provides IP or Hostname → get_device_info must be called first.
  • No Manual SQL: All data access happens via tools.
  • Units: Memory utilization values are always in Percentage (%).
  • Scope Handling: If no device is specified → analysis defaults to Network-Level.

Memory Leak Decision Flow

Is it a Catalyst Center device?
 ├─ YES → get_catalyst_center_devices
 │       → memory_leak_analysis_and_detection_for_catalyst_center_devices
 └─ NO
     ├─ Specific device provided → device_level_memory_leak_analysis
     └─ No device provided → network_level_memory_leak_analysis

Forecasting Workflow

  • Uses XGBoost time-series prediction
  • Output is always presented in tabular format
  • Supports per-device forecasting

Background Workflow

For Capturing below services memory metrics, we use ONES-API explorer

  • SWSS
  • BGP
  • ONES-Agent
  • SYNCd
  • SNMP

Configuration (Required) for ONES-API Integration

Update the following values inside tools.py:

BASE_URL_ONES = "Your_ONES_SERVER_URL"
USERNAME_ONES = "YOUR_ONES_USERNAME"
PASSWORD_ONES = "YOUR_ONES_PASSWORD"

Setup & Run Instructions

Clone the Repository

git clone https://github.com/AvizNetworks/ncp-sdk-agents.git
cd ncp-sdk-agents
git checkout agent-mo

Configure ONES Credentials

Edit tools.py and update:

BASE_URL_ONES
USERNAME_ONES
PASSWORD_ONES

Authenticate with NCP

Provide the following details when prompted or via config:

  • NCP URL
  • Username
  • Password

Install NCP SDK Package

pip install ncp

For creating package (.ncp)

ncp authenticate
ncp package .

Deploy Agent to NCP Playground

  1. Open NCP Playground
  2. Upload agent_mo.ncp
  3. Deploy the agent to the NCP Server

Use the NCP UI

  • Open NCP UI
  • Select Agent-MO

  • Ask natural language questions like:

  • “Check memory leaks for 192.168.1.5”

  • “Forecast memory usage for this device”
  • “Analyze network-wide memory utilization”
  • “Any memory leak is observed in the fabric?”
  • “How is the memory footprint in the fabric?”
  • “Why is hitting high memory utilization?"
  • “Top 10 devices with high memory utilization”
  • “Did any device start experiencing memory issues?”

Screenshots

Figure 1. Agent-MO Memory Leak Detection

Agent-MO Memory Leak Detection

Figure 2. Service-Level Memory Analysis

Service-Level Memory Analysis

Service-Level Memory Analysis 2

Figure 3. Memory Forecasting Output with Chart

Memory Forecasting Output with Chart

Example Interaction

User:

Check memory leaks for 192.168.1.5

Agent-MO Workflow:

  1. Resolve IP → MAC using get_device_info
  2. Detect leak using device_level_memory_leak_analysis
  3. Present findings with recommendations

Future Enhancements

  • Advanced visualization support
  • Automated remediation suggestions
  • Add supports for Additional dataconnectors like cumulus etc.
  • Support for Background agents and automated alerts for potential memory leaks.