CLIXtract

CLIXtract is a deterministic CLI parsing engine for network devices. It converts raw CLI output into structured JSON using a local LLM (Ollama). The system is designed to extract only explicitly present fields and never invent data. Missing values are returned as "N/A".

---

Project Structure

Makefile                 # Build and utility commands
common/                  # Shared utilities and helpers
llm/                     # LLM clients and wrappers
main.py                  # FastAPI server entry point
prompts/                 # System prompts and instructions
postman/                 # Example Postman requests
requirements.txt         # Python dependencies
schema/                  # JSON schemas (optional)
settings.json            # Configuration for LLM, server, and models
TODO.txt                 # Notes and tasks
version.json             # Project version info

---

Prerequisites

• Python 3.12+
• Ollama installed and model downloaded (e.g., llama3.1:latest)
• macOS/Linux (Metal GPU recommended)
• pip for Python dependencies

---

Setup

1. Create virtual environment and install dependencies:

make prepare-venv

2. Start Ollama server:

ollama serve

Check that your model is available:

ollama list

---

Makefile Commands

• Format code:

make format

Formats Python code with black and isort.

• Clean project:

make clean

Removes temporary files and virtual environment.

• Prepare virtual environment:

make prepare-venv

• Run FastAPI server:

make run-server

Server port is picked from settings.json.

---

API

Endpoint: /parse-cli
Method: POST
Content-Type: application/json

Request Payload

Example for show version:

{
  "command_name": "show version",
  "command_output": "Cisco IOS Software, C3750X Software (C3750X-UNIVERSALK9-M), Version 15.2(6)E2, RELEASE SOFTWARE ...",
  "user_instruction": "Extract software version (sw), firmware version (fw), hardware type, IOS version, system serial number, system image. Return strictly valid JSON. Use 'N/A' if not available."
}

Example for show interface:

{
  "command_name": "show interface",
  "command_output": "GigabitEthernet1/0/1 is up, line protocol is up\n  Hardware is Gigabit Ethernet, address is 001a.2b3c.4d5e (bia 001a.2b3c.4d5e)\n  MTU 1500 bytes, BW 1000 Mbps, DLY 10 usec, ...",
  "user_instruction": "Extract interface name, status, protocol, hardware, MAC address, MTU, bandwidth. Return strictly valid JSON. Use 'N/A' if not available."
}

---

Response Example

{
  "sw": "C3750X Software (C3750X-UNIVERSALK9-M)",
  "fw": "N/A",
  "hardware": "WS-C3750X-24T-S",
  "ios_version": "15.2(6)E2",
  "serial_number": "FDO12345678",
  "system_image": "flash:c3750x-universalk9-mz.152-6.E2.bin"
}

---

Prompts

• System prompts are in prompts/system_prompt.txt.
• They define parsing rules and instructions for the LLM.
• Always returns a strict JSON array, or an empty array [] if no valid objects are found.

---

Notes

• Only strictly present fields in CLI output are included; missing fields return "N/A".
• LLM client settings are configured via settings.json.
• FastAPI server returns JSON for any network CLI command provided.

---

Postman

The postman/ folder contains example requests you can import to test the API.

---

TODO

See TODO.txt for ongoing tasks and improvements.

---