Edge & Infrastructure

Kite Agent

Cross-platform edge operations agent in Rust. AI-powered log analysis on 1GB RAM.

Lightweight Rust agent for edge server monitoring with local AI inference (via llama.cpp). Strict resource limits, cross-platform.

GitHub ↗
Rust 1.75+ MIT Linux + Windows

Features

  • Extremely lightweight — Designed for ≤1GB RAM and <10% CPU usage on edge hardware.
  • Cross-platform — Native support for Linux (systemd) and Windows (Service).
  • AI-powered local inference — Runs quantized GGUF models (TinyLLaMA-1.1B class) via llama.cpp for intelligent log analysis.
  • Fast regex rule engine — Pattern-match known failure signatures before AI inference for zero-latency alerting.
  • MQTT communication — Publish findings to a central MQTT server for aggregation.
  • Lazy model loading — AI model auto-unloads after idle periods to conserve memory.

Architecture

text 
Watchdog (resource monitor) + Collector (logs & metrics) + Rule Engine
       |
   Agent Core
       |
+------+------+
|             |
MQTT        Model Engine
Communicator (llama.cpp)

The Watchdog monitors resource usage, the Collector gathers logs and metrics, and the Rule Engine applies fast regex patterns. Results flow to Agent Core, which dispatches to the MQTT Communicator and optionally to the Model Engine for deeper AI analysis.

System Requirements

  • CPU: 1 core minimum
  • RAM: 1GB maximum budget
  • Storage: 100MB binary + 500MB model files
  • OS: Linux with systemd, or Windows Server 2016+

Quick Start

Linux

bash 
git clone https://github.com/Mi-Bee-Studio/kite-agent.git
cd kite-agent
sudo ./scripts/install_linux.sh
sudo systemctl start kite-agent

Tech Stack

  • Language: Rust 1.75+
  • AI inference: llama.cpp (GGUF quantized models)
  • Communication: MQTT
  • Platform: Linux (systemd), Windows Server 2016+
  • License: MIT

Documentation

Architecture overview, configuration reference, and deployment guides are in the GitHub repository.