![License]() ![#Libraries]() ![#Microservices]() ![#Tools]() ![#Samples]()
Edge-AI-Libraries
Welcome to Edge AI Libraries - a set of Intel-optimized libraries, microservices, tools, and demos intended for developing real-time edge AI solutions.
If you are an AI developer, data scientist, or system integrator, these assets will help you organize data, train models, run efficient inference, and deliver robust, industry-grade automation systems for computer vision, multimedia, and industrial use cases.
Key Components
These flagship components represent the most advanced, widely adopted, and impactful tools in the repository:
Build efficient media analytics pipelines using streaming AI pipelines for audio/video media analytics using GStreamer for optimized media operations and OpenVINO for optimized inferencing Create 3D/4D dynamic digital twins from multimodal sensor data for advanced spatial analytics. Build computer vision AI models enabling rapid dataset management, model training, and deployment to the edge. Deploy visual anomaly detection with this state-of-the-art library, offering algorithms for segmentation, classification, and reconstruction, plus features like experiment management and hyperparameter optimization. Optimize, run, and deploy AI models with this industry-standard toolkit, accelerating inference on Intel CPUs, GPUs, and NPUs. Supports a broad range of AI solutions including vision-based applications, generative AI, and vision-language models.Component Categories
Model Training and Optimization
This group provides core AI libraries and tools focused on computer vision model building, training, optimization, and deployment for Intel hardware. They address challenges such as dataset curation, model lifecycle management, and high-performance inference on edge devices.
Build computer vision AI models enabling rapid dataset management, model training, and deployment to the edge. Deploy visual anomaly detection with this state-of-the-art library, offering algorithms for segmentation, classification, and reconstruction, plus features like experiment management and hyperparameter optimization. Software for efficient model training and deployment. Toolkit for optimizing and deploying AI inference, offering performance boost on Intel CPU, GPU, and NPU devices. & Model APIA set of advanced algorithms for model training and conversion.
Dataset management framework to curate and convert vision datasets. Create, tailor, and implement custom AI models directly on edge platforms.LLM Inference Optimization
These libraries reduce the compute, memory, and token cost of running large language model and agent workloads on the edge, improving inference efficiency without changing model behavior.
Pluggable token compression for LLM agent systems — compresses system prompts, context, and tool schemas to cut input token usage. Near-lossless 4-bit KV-cache quantization for LMCache/vLLM, reducing KV-cache memory and storage overhead for KV offload on Intel edge accelerators.Streaming and Multimedia AI
Handling large-scale media analytics workloads, these components support real-time audio and video AI pipeline processing, transcription, and multimodal embedding generation, addressing common use cases like surveillance, content indexing, and audio analysis.
& Deep Learning Streamer Pipeline ServerStreaming AI pipeline builder with scalable server for media inferencing.
Microservices providing real-time audio transcription and intelligence extraction. Services handling vision-language models and embedding generation for multimodal search. Software for creating dynamic 3D/4D digital twins for spatial analytics. Real-time analytics microservice designed for anomaly detection and forecasting on sensor time-series data. Microservices providing a multi-level, temporal-enhanced approach to generate high quality summaries for video files, especially for long videos. Microservice discovering USB and ONVIF network cameras and describing their best capture configurations and RTSP media profiles.Data Preparation and Retrieval
Efficient data management and retrieval are crucial for AI performance and scalability. This group offers components for dataset curation, vector search, and document ingestion across multimodal data.
& Visual Data Preparation (Milvus and Multimodal-dataprep)High-performance vector similarity search and visual data indexing.
& Document IngestionModel Download downloads AI models from external sources, converts them to the OpenVINO™ Intermediate Representation (IR) format with optional optimization, and caches them locally. Document Ingestion prepares documents for AI workflows.
Splits/segments video streams into chunks, supporting batch and pipeline-based analytics.Benchmarking Tools
These support tools provide visual pipeline evaluation and performance benchmarking to help analyze AI workloads and industrial environments effectively.
Benchmark and analyze AI pipeline performance on various edge platforms. Compliance and performance testing toolkit for motion control. Showcase, monitor, and optimize the scalability and performance of AI workloads on Intel edge hardware. Configure models, choose performance modes, and visualize resource metrics in real time. Command Line Interface (CLI) tool that provides a collection of test suites to qualify Edge AI system. It enables users to perform targeted tests, supporting a wide range of use cases from system evaluation to data extraction and reporting.Edge Control Libraries
Focused on real-time industrial automation, motion control, and fieldbus communication, these components provide reliable, standards-compliant building blocks for manufacturing and factory automation applications.
& EtherCAT MasterstackEtherCAT communication protocol stack and development tools.
and RTmotionLibraries implementing motion control standards for servo drives and real-time trajectory management.
Tools for data communication in automation networks. Microservice for collecting, processing, and distributing real-time sensor and industrial device data; supports both edge analytics and integration with operational systems.Robotics Libraries
Optimized libraries for robotic perception, localization, mapping, and 3D point cloud analytics. These tools are designed to maximize performance on heterogeneous Intel hardware using oneAPI DPC++, enabling advanced robotic workloads at the edge.
High-speed nearest neighbor library, optimized for Intel architectures; supports scalable feature matching, search, and clustering in robotic vision and SLAM. Efficient ORB feature and descriptor extraction for visual SLAM, mapping, and tracking; designed for multicamera and GPU acceleration scenarios. Accelerated modules from PCL for real-time 2D/3D point cloud processing—supports object detection, mapping, segmentation, and scene understanding in automation and robotics. Unified interface library bridging motion control commands between AI modules and industrial/robotic devices; simplifies real-time control integration and system interoperability in mixed hardware environments.Sample Applications and Reference Implementations
Ready-to-use example applications demonstrating real-world AI use cases to help users get started quickly and understand integration patterns:
Conversational AI application integrating retrieval-augmented generation for question answering.Optimized for Intel(R) Core. AI pipeline for automated summarization of textual documents. Application combining video content analysis with search and summarization capabilities. 🧪 Now with experimental support for Intel Arc Pro B-series GPUs (B60, B65, B70). Automate the setup of edge AI development environments using these proven reference scripts. Quickly install required drivers, configure hardware, and validate platform readiness for Intel-based edge devices.Visit the Edge AI Suitesrepository for a broader set of sample applications targeted at specific industry segments.
Edge Analytics Microservices
Specialized microservices delivering machine learning-powered analytics optimized for edge deployment. These microservices support scalable anomaly detection, classification, and predictive analytics on structured and time-series data.
An efficient Isolation Forest microservice for unsupervised anomaly detection supporting high-performance training and inference on tabular and streaming data. High-speed Random Forest microservice for supervised classification tasks, optimized for edge and industrial use cases with rapid training and low-latency inference.Edge-device Enablement Framework (EEF)
A comprehensive framework providing hardware abstraction, device management, and deployment tools for edge AI applications. Simplifies cross-platform development and enables consistent deployment across diverse edge hardware architectures.
Framework for Intel® platform enablement and streamlining edge AI application deployment across heterogeneous device platforms.AI Agent Skills
This repository ships a set of agent skills* that automate common development tasks — changelog generation, release notes authoring, and security review — directly from your editor.
| Skill | Slash command | What it does |
|-------|--------------|--------------|
| generate-changelog | /generate-changelog | Generates or updates CHANGELOG.md from git history between two branches or tags |
| generate-release-notes | /generate-release-notes | Produces formatted release notes for a component folder by comparing two branches or tags |
| security-review | /security-review | On-demand security review for code, Dockerfiles, Helm charts, and CI/CD workflows |
Quick start
Auto-trigger — describe what you want in Copilot/Claude Code/Codex chat in agent mode and the agent picks the right skill:
Generate a changelog for microservices/time-series-analytics comparing release-2026.0.0 and release-2026.1.0Explicit slash command — invoke a skill directly with optional inline arguments:
/generate-changelog microservices/time-series-analytics release-2026.0.0 release-2026.1.0Parameterized prompts — for guided, form-based input use the companion prompt files (coding agent(github copio) pops up an input box for each parameter):
/run-changelog
/run-release-notesContribute
To learn how to contribute to the project, see CONTRIBUTING.md.
Community and Support
If you need help, want to suggest a new feature, or report a bug, please use the following channels:
- Questions & Discussions: Join the conversation in
- Bug Reports & Feature Requests: Submit issues via
License
The Edge AI Libraries project is licensed under the APACHE 2.0 license, except for the following components:
| Component | License | |:----------|:--------| | Dataset Management Framework (Datumaro) | MIT License | | Intel® Geti™ | Limited Edge Software Distribution License | | Deep Learning Streamer | MIT License |
Intended Use
Unless stated otherwise, software maintained under the Edge AI Libraries repository is intended for demonstration and reference purposes only. Certain features, such as authentication, TLS termination, and external access controls are assumed to be covered at the infrastructure level. For more information, refer to Notes on Usage.