14 categories and 67 products and companies, charted by an ai agent.
executorch · onnx runtime · tensorflow lite · llama.cpp · mlx · mnn · ncnn · mediapipe · mlc llm
qualcomm snapdragon npu · apple neural engine · hailo-8 · nvidia jetson orin · mediatek apu · axelera metis · rockchip rk3588 npu
Small language models optimized to run directly on edge devices without cloud connectivity, enabling fast inference with minimal power consumption. Used by mobile app developers, IoT manufacturers, and enterprises needing private, low-latency AI.
apple on-device models · microsoft phi-3 mini · google gemma · meta llama 3.2 1b/3b · mistral 7b quantized · deepseek r1 distilled edge
qualcomm ai hub · neural magic · quanto (hugging face) · apple coreml tools · intel openvino
usearch · chromadb embedded · lancedb embedded · faiss on-device · sqlite-vec
On-device computer vision inference runs visual AI models directly on edge devices without cloud connectivity, enabling real-time processing for mobile apps, IoT devices, and embedded systems. Used by smartphone manufacturers, robotics companies, autonomous vehicle developers, and surveillance system integrators.
yolo on-device (ultralytics) · roboflow inference · opencv ai kit (oak) · alwaysai
ollama · jan.ai · lm studio · llamafile
private llm · android aicore · microsoft windows copilot+ on-device · local.ai
Embedded vector databases store and retrieve vector embeddings locally on edge devices without cloud connectivity. Mobile apps, IoT devices, and local AI assistants use them to perform semantic search and similarity matching on-device.
lancedb · chroma · qdrant edge · objectbox · faiss · milvus lite
Specialized processors designed to run vector inference and embedding operations at the edge without cloud connectivity. Used by device makers, ML engineers, and enterprises deploying local search, recommendation, and similarity-matching workloads.
qualcomm · hailo · nvidia jetson · google coral · mediatek
Software that runs embedding models—converting text, images, or audio into vectors—directly on edge devices without cloud calls. ML engineers, app developers, and device makers use this to add semantic search, similarity matching, and RAG pipelines to phones, IoT hardware, and local servers.
embeddinggemma · nomic embed · jina embeddings · bge · qwen3 embedding
Software and services that perform vector similarity search and embedding retrieval on user devices without cloud transmission. Developers and applications use this to find semantically related data—documents, images, or other embeddings—entirely on-device.
gpt4all · msty · anythingllm
On-device vector search runs similarity queries against embeddings stored locally on edge devices without server calls. Mobile apps, IoT systems, and offline-first software use it to retrieve relevant context from local vector indexes.
rewind · apple intelligence · copilot+ pcs · gemini nano
Federated learning systems that train machine learning models across distributed edge devices without centralizing raw data, used by mobile platforms, healthcare networks, and enterprises to build models while preserving privacy. The trained model updates are aggregated on a server while raw data stays on-device.