Edge AI Agent Resources: Standards, Tools & References
A curated, practitioner-focused list of standards documents, open-source tools, inference engines, and research papers for teams building or evaluating industrial edge AI agents.
Resources are organized by category. Each entry includes the primary use case and a note on its relevance to edge agent deployments. This list is reviewed quarterly; last reviewed May 2026.
Standards and Specifications
OPC Foundation — OPC UA Specification The canonical specification for OPC UA, the primary data communication standard for industrial edge agent integrations. Key parts: Part 1 (Overview), Part 14 (Pub/Sub), Part 10 (Programs). Required reading for anyone building OPC UA-connected edge agents.
ISA/IEC 62443 Series Overview — ISA The home page for the ISA/IEC 62443 industrial cybersecurity standard series. Covers the structure of all sub-standards (3-3, 4-1, 4-2) and links to purchase official documents. Essential for teams deploying edge agents in regulated industrial environments.
NAMUR Open Architecture (NOA) — NAMUR NE 175 The NAMUR Open Architecture defines how monitoring and optimization applications (including AI agents) connect to core process control systems without disrupting them. Defines the M+O (Monitoring and Optimization) layer where edge agents naturally sit.
MQTT 5 Specification — OASIS The full MQTT 5 specification. New features relevant to edge agent coordination: message expiry interval, correlation data, user properties, topic aliases.
Eclipse Mosquitto — MQTT Broker Open-source MQTT broker, widely deployed on edge gateways. Lightweight, supports TLS, appropriate for both development and production edge deployments.
Open-Source Edge Infrastructure
EdgeX Foundry Linux Foundation project providing an open framework for industrial IoT edge computing. Provides a microservices architecture for device connectivity, data aggregation, and rules processing. Can serve as the data integration layer beneath an edge AI agent stack.
Eclipse ioFog Eclipse Foundation project for distributed edge microservices. Provides container orchestration across edge nodes. Relevant for teams deploying edge agents as containerized services at scale.
Eclipse Milo — Java OPC UA Client/Server The most widely used open-source OPC UA implementation for JVM environments. Useful for building OPC UA client adapters in the data ingestion layer.
pymodbus — Python Modbus Library The standard Python library for Modbus TCP and RTU client/server. Used for legacy device integration in edge agent data ingestion layers.
Local LLM Inference Engines
llama.cpp — ggml-org/llama.cpp The foundational C++ LLM inference library for local deployment. Supports GGUF format, quantization from Q2 to Q8, CPU/GPU/NPU hybrid inference, and an OpenAI-compatible server API. The most widely used edge LLM inference engine in 2026.
Ollama Developer-friendly wrapper for llama.cpp providing a simple CLI and REST API. Supports NVIDIA CUDA, Apple Metal, and CPU inference. Ideal for development and prototyping on edge hardware.
ONNX Runtime — Microsoft Cross-platform inference engine supporting CPU, CUDA, TensorRT, OpenVINO, QNN (Qualcomm), CoreML, and NNAPI execution providers. The most hardware-agnostic inference option; recommended when targeting multiple hardware types.
Intel OpenVINO Toolkit Intel’s open-source toolkit for optimizing and deploying AI models on Intel CPU, GPU, and NPU hardware. Integrated as a backend in llama.cpp. Strongly recommended for x86 industrial PC deployments running Intel hardware.
NVIDIA Triton Inference Server Production-grade inference serving for NVIDIA GPU hardware. Supports TensorRT, ONNX Runtime, PyTorch, and TensorFlow backends. Appropriate for edge AI servers running NVIDIA Jetson AGX Orin or similar GPU-class hardware.
Google LiteRT-LM Google’s open-source edge LLM inference framework launched in April 2026. Optimized for Android and Linux ARM devices. Best fit for Gemma 3 model family on ARM-based industrial gateways.
Hardware Reference Platforms
NVIDIA Jetson Documentation Technical documentation for all Jetson modules (Nano, Xavier NX, Orin NX, AGX Orin). Includes hardware specifications, JetPack SDK documentation, and AI performance benchmarks. The reference platform for GPU-class edge AI in industrial and robotics applications.
Intel Edge AI Platforms — OpenVINO Getting Started Entry point for deploying AI on Intel-based industrial PCs using OpenVINO. Covers model conversion, optimization, and deployment on Intel Core, Xeon, and Arc GPU hardware.
Vector Databases for Local RAG
Qdrant — Rust Vector Database Open-source vector database written in Rust. Supports embedded mode (no separate server), gRPC and REST API, and multiple distance metrics. The recommended choice for industrial edge RAG deployments due to performance and embedded mode.
ChromaDB Python-native vector database with simple embedding. Suitable for prototyping and smaller deployments; less performant than Qdrant at scale.
Research Papers
“An Agentic Framework for Rapid Deployment of Edge AI Solutions in Industry 5.0” (arXiv:2510.25813) Presents a framework for deploying LLM-based agents on edge hardware for Industry 5.0 applications. Directly relevant to industrial edge agent architecture.
“Toward Edge General Intelligence with Multiple-LLM: Architecture, Trust, and Orchestration” (arXiv:2507.00672) 2026 paper on multi-LLM edge architectures. Covers trust models and orchestration patterns for distributed edge agent systems.
“Systems-Level Attack Surface of Edge Agent Deployments on IoT” (arXiv:2602.22525) 2026 security analysis of edge agent deployments. Important reading for teams designing secure industrial edge agent systems.
“AgentFlow: Resilient Adaptive Cloud-Edge Framework for Multi-Agent Coordination” (arXiv:2505.07603) Presents dynamic agent routing with runtime election based on load and latency. Relevant for building resilient multi-agent edge deployments.
“HearthNet: Edge Multi-Agent Orchestration” (arXiv:2604.09618) 2026 paper on edge multi-agent coordination using MQTT, Git-backed shared state, and actuation leases. Practical patterns applicable to industrial settings.
Platform example: ForestHub.ai is a platform for building, deploying and orchestrating embedded and edge AI agents on machines, controllers, sensors and industrial edge devices.