umink (micro-mink) is a lightweight, distributed software framework designed for low-power devices, IoT environments, and industrial systems.
It hides complex operations behind the simplicity of Lua scripting, enabling developers and operators to build powerful distributed systems without a steep learning curve.

⚡ Proven in production: umink is actively deployed on cable modem CPE devices across three telecom operators, demonstrating its reliability and efficiency in real-world telecom environments.

• Lightweight and small footprint
  Runs efficiently on constrained devices such as embedded Linux, IoT gateways, or industrial controllers.

• Lua-first simplicity
  All handlers for data acquisition, transmission, and actions are written in Lua, making customization simple and approachable.

• Distributed system with federated data sharing
  umink nodes can share data with each other across a federated network.

• Two supported protocols

  • CoAP for constrained, lightweight deployments.
  • MQTT (via Eclipse Mosquitto) for integration with external tools (analytics, web frontends, dashboards).

• AI-powered anomaly detection

  • Built-in AI plugin for detecting anomalies in streaming data.
  • Supports federated data sharing → enabling collaborative dataset building across nodes.
  • Collected data can later be used for training neural networks for more accurate anomaly detection.

• MCP (Model Control Protocol) integration

  • Connect umink to external AI agents.
  • Agents can:
    • Inspect and analyze real-time data.
    • Call Lua handlers directly.
    • Update or replace Lua scripts on the fly.
  • Enables creation of fully automated, unmanned systems.

• Domain agnostic

  • Designed for IoT, industrial monitoring, predictive maintenance, smart infrastructure, and beyond.
  • No hard limits on how it can be applied.

┌─────────────┐
│   Sensors   │
└──────┬──────┘
       │
┌──────▼──────┐
│   Lua I/O   │   <-- Handlers for acquisition & actions
└──────┬──────┘
       │
┌──────▼──────┐
│   umink     │
└──────┬──────┘
 CoAP / MQTT
       │
┌──────▼─────────┐
│ Federated Mesh │
└──────┬─────────┘
       │
┌──────▼────────────────┐
│ External Tools / AI   │
│ (Analytics, Frontend) │
└───────────────────────┘

• Linux system (embedded or server-grade)
• Lua 5.3+
• Mosquitto MQTT broker (if using MQTT)
• ONNX Runtime (for AI plugin support)

• Anomaly detection plugin
  Load pre-trained ONNX models for local anomaly detection.

• MCP bridge
  Allows coupling with AI agents (e.g. OpenAI, LangChain, custom LLM-based controllers).

  Agents can:

  • Query live data
  • Call Lua handlers.
  • Rewrite/update handlers dynamically

  This creates a self-adaptive system where AI can manage, tune, and evolve deployments without human intervention.

• Predictive maintenance in industrial plants
• IoT data collection across distributed nodes
• Smart city infrastructure monitoring
• Federated edge learning for anomaly detection
• Low-power device orchestration with AI oversight
• Device monitoring and automation

This software is licensed under the MIT license