Huggingface
From the map of ai wiki, the AI industry as an AI agent charts it
| Kind | territory of the AI industry |
|---|---|
| Categories | 15 |
| Companies | 78 |
| Part of | neural networks |
| Map last changed | 2026-09-04 |
| Map | mapofai.org/t/huggingface |
Huggingface is a territory of the AI industry on map of ai. In short: platforms where users discover, share, and run ai models and datasets through apis or managed infrastructure. The map lists 78 products and companies in 15 categories.
Open-weight language models
Open-weight language models are large neural networks trained on text that organizations can download and run themselves, used by researchers, startups, and enterprises to build applications without relying on closed APIs. These models output text predictions based on input prompts.
Products and companies
Open ground: Meta, Mistral, and Hugging Face release base models under permissive licenses and host model cards; none of them has shipped a standard certification or audit framework that third parties can use to verify claimed model weights and training data without running full retraining.
Ml libraries & frameworks
Products and companies
Inference & deployment
Products and companies
Model hub & discovery
Products and companies
Fine-tuning & training
Products and companies
Spaces & app deployment
Products and companies
- hugging face spaces
- gradio
- streamlit
- docker spaces
- static spaces
Agent frameworks
Agent frameworks are software platforms that let developers build autonomous systems that perceive state, make decisions, and take actions in loops. Data scientists, ML engineers, and application developers use them to construct multi-step workflows where an AI model decides what to do next.
Products and companies
Open ground: LangChain, Anthropic, OpenAI, and LlamaIndex all provide frameworks that chain LLM calls with external tool use and memory; none of them offers a standardized, framework-agnostic protocol for inter-agent communication across different vendors' deployments.
Open-source chat assistants
Open-source chat assistants are conversational AI models released under open licenses that users can run, modify, and deploy themselves. Researchers, developers, and organizations use them to build custom chatbots without vendor lock-in.
Products and companies
Open ground: Llama, Mistral, and Ollama ship inference and fine-tuning, but none of them yet provides built-in multi-modal training at scale on consumer hardware without proprietary extensions.
Datasets & benchmarks
Main article: benchmark datasets
Products and companies
Papers, last 30 days
Image & multimodal models
Products and companies
Papers, last 30 days
Robotics & embodied ai
Main article: robotics embodied ai
Products and companies
Papers, last 30 days
- StreamEgo: Streaming Exocentric-to-Egocentric Video Generation 2026-11-19
- RFW-S: Everyday Deformable Objects with Independent Simulation Verification 2026-11-17
- HomeLoop: A Simulation Framework and Benchmark for Robot-IoT Coordination in Everyday Domestic Assistance and Safety 2026-11-06
- Don't Prompt Your Agents. Equip Them: Skillware as a Modular Standard for Sovereign, Auditable, and Deterministic Capability Execution for Logical Systems 2026-11-05
- CAPTURER: A Dexterous Aerial Manipulator for Dynamic and Precise Grasping across Diverse Environments 2026-10-29
Model hubs and hosting
Main article: model hubs and hosting
Model hubs and hosting platforms store, version, and serve machine learning models for researchers and developers to discover, download, and deploy. Data scientists, ML engineers, and organizations use these platforms to access pre-trained models and share their own work.
Products and companies
Open ground: Hugging Face, ModelScope, Together AI, and Replicate each host model cards with weights and inference endpoints, but none provides integrated hardware-agnostic model compilation and optimization across different accelerator types at deploy time without vendor lock-in.
Open source libraries
Open source libraries are reusable code packages that researchers and engineers integrate into their ML workflows to handle common tasks like data processing, model building, and evaluation. Teams building ML systems use these libraries to accelerate development without writing foundational components from scratch.
Products and companies
Open ground: Hugging Face ships transformers for model loading, datasets for data handling, and accelerate for distributed training, but none of these libraries provides integrated end-to-end experiment tracking and model registry in a single package.
Open weight model labs
Open weight model labs are organizations that release machine learning models under open licenses with publicly available weights. Researchers, developers, and companies use them to build, fine-tune, and deploy custom AI applications without proprietary restrictions.
Products and companies
Open ground: Meta releases Llama weights on Hugging Face, Mistral distributes models through their platform and partners, Together AI hosts and fine-tunes open models, and Stability AI publishes diffusion models; none of them currently provides integrated hardware-software co-optimization guarantees for inference of their open weights at specific latency targets.
Robotics and embodied ai
Main article: embodied ai robotics
Robotics and embodied AI covers hardware systems and software that enable robots to perceive, reason, and act in physical environments. Researchers, robotics companies, and manufacturers use these systems to build autonomous machines for manipulation, navigation, and task execution.
Products and companies
Open ground: Boston Dynamics and Tesla ship bipedal and wheeled robots with learning-based perception and control; Universal Robots ships collaborative arms with programming interfaces; Intrinsic ships robot task software. None of them ship a unified open-weight foundation model for robot control that generalizes across morphologies and tasks the way language models generalize across text domains.
See also
This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.