Network AI Agent

A powerful, AI-driven tool for seamless network automation and management.

A lightweight, "Human-in-the-Loop" AI agent that translates natural language into network configuration and show commands. Built with LangGraph, Nornir, and Chainlit.

🚀 Quick Start (Docker)

The easiest way to run the agent is using Docker. This ensures a consistent environment.

1. Setup

git clone https://github.com/yourusername/network-agent.git
cd network-agent

2. Configuration

Create your secrets and inventory files.

.env (Secrets):

GROQ_API_KEY=gsk_your_api_key_here

hosts.yaml (Inventory):

---
sw1:
  hostname: 192.168.1.10
  groups: [cisco]
sw2:
  hostname: 192.168.1.11
  groups: [cisco]

groups.yaml (Platform Definitions):

---
cisco:
  platform: cisco_ios
  username: admin
  password: admin

arista:
  platform: arista_eos
  username: admin
  password: admin

3. Run

docker compose up --build

Access the Web UI at http://localhost:8000

---

🏗️ Architecture

The project follows a flattened, KISS architecture leveraging Nornir for parallel execution:

Component	File	Purpose
Brain	agent/	LangGraph workflow, prompts, and decision routing.
Hands	tools/	Split into show.py (Read) and config.py (Write).
Engine	utils/devices.py	Nornir initialization and task execution engine.
UI	app.py	Chainlit Web Interface.
CLI	main.py	Terminal entry point.

Key Features

1. Parallel Execution:
   • Uses Nornir to execute read-only commands (show version) on hundreds of devices simultaneously.
2. Safety First:
   • Sequential Configs: Configuration changes are handled sequentially with a Human-in-the-Loop approval step for every batch.
   • The Agent cannot execute changes without your explicit "Yes".
3. Smart Input Sanitization:
   • Automatically cleans up LLM outputs (e.g., stripping markdown code blocks from config sets) before sending to devices.
4. Structured Output:
   • Uses TextFSM (via ntc-templates) to parse raw CLI output into structured JSON data for the LLM.

---

🛠️ Local Development

If you prefer running without Docker:

1. Install Dependencies:

   # Using uv (Recommended)
   uv sync
   # Or standard pip
   pip install -r requirements.txt

2. Run Web UI:

   chainlit run app.py -w

3. Run CLI:

   python main.py "Show ip int brief on sw1"

---

📚 Tools & Modules

• agent.nodes: Defines the graph state and interaction logic.
• utils.devices: Lazy-loads Nornir and injects secrets from environment variables.
• tools.show: Wraps nornir_netmiko.netmiko_send_command.
• tools.config: Wraps nornir_netmiko.netmiko_send_config.

🤝 Contributing

1. Fork the repo
2. Create a feature branch
3. Commit your changes
4. Push to the branch
5. Create a Pull Request