Cloudflare Workers AI

Run machine learning models, powered by serverless GPUs, on Cloudflare's global network. Workers AI allows you to run AI models in a serverless way, without having to worry about scaling, maintaining, or paying for unused infrastructure. You can invoke models running on GPUs on Cloudflare's network from your own code â€” from Workers, Pages, or anywhere via the Cloudflare API.

General
Author	Cloudflare, Inc.
Website	Cloudflare Workers AI
Documentation	https://developers.cloudflare.com/workers-ai/
Type	Serverless AI Inference Platform
Launch Year	2023
GPU Network	180+ cities globally

Features

50+ Open-Source Models

Text generation (Llama, Mistral, and more)
Text embeddings and classification
Image generation and classification
Automatic speech recognition
Translation models
Object detection capabilities

Serverless Infrastructure

Pay-for-what-you-use pricing model
Automatic scaling with demand
No infrastructure management required
Fast cold start times with V8 isolates

Global Edge Network

AI inference close to users for low latency
Models available in 180+ cities worldwide
Reduced network bottlenecks
Consistent performance globally

Developer Platform Integration

Seamless integration with Cloudflare Workers
Works with Pages for full-stack AI applications
REST API for platform-agnostic access
Integration with Vectorize (vector database)
AI Gateway for monitoring and control

Key Capabilities

Edge AI Computing: Run AI models at the network edge for minimal latency
Serverless GPU Access: Access powerful GPU infrastructure without provisioning
Model Catalog: Curated selection of popular open-source AI models
Real-time Inference: Low-latency AI processing for interactive applications
Global Deployment: Deploy once, run everywhere on Cloudflare's network
Integrated Ecosystem: Works with R2 storage, D1 database, and other Cloudflare services

Use Cases

Building AI-powered chatbots and conversational interfaces
Real-time content moderation and classification
Image and video processing at scale
Personalization and recommendation engines
Automated translation and localization
Voice recognition and text-to-speech applications
RAG (Retrieval-Augmented Generation) systems
AI-powered API endpoints and microservices

Supported Model Categories

Large Language Models: For text generation and chat applications
Embedding Models: For semantic search and similarity matching
Image Models: For generation, classification, and analysis
Speech Models: For transcription and synthesis
Vision Models: For object detection and recognition
Translation Models: For multilingual content processing

AI-Powered Emergency Room Digital Twin

# Uniq ER Digital Twin – AI-Powered Emergency Room Simulation Uniq ER Digital Twin is an AI-powered emergency room simulation developed for the AMD Developer Hackathon ACT III, under the Health & Wellbeing track. The project explores how artificial intelligence and digital twin technology can improve hospital efficiency, optimize resource allocation, and support critical decision-making in emergency departments. The platform simulates a dynamic emergency room environment with patients, doctors, nurses, ambulances, treatment rooms, and limited medical equipment. It features mass-casualty scenarios, cascading emergencies, and patient deterioration, creating realistic challenges in managing hospital resources such as CT scanners, ventilators, and specialized medical staff. Powered by AMD Developer Cloud through Evolus, the AI agent analyzes emergency room conditions and recommends patient assignments and resource allocation strategies. These AI-generated decisions are compared against the simulation's built-in allocation system under identical scenarios, allowing evaluation of waiting times, patient deterioration, resource utilization, and overall operational performance. The interactive 2D dashboard provides real-time visualization of hospital operations, patient movements, resource availability, and incident timelines. Users can control simulation speed, introduce disruptions, examine decisions, and generate detailed outcome reports. By combining AMD-powered AI, autonomous agents, and digital twin technology, Uniq demonstrates the potential of intelligent hospital coordination systems to reduce bottlenecks, improve emergency preparedness, and support healthcare professionals in high-pressure environments.

RoboProof: Motion-Learning Lab

RoboProof is a physical-AI research prototype for improving how a VEX V5 robot moves. Instead of asking a language model to control motors directly, it combines a Guiding Vector Field (GVF), which provides a structured path-following reference, with a TD3 reinforcement-learning policy that learns residual adjustments from simulated sensor feedback. The system uses the project’s original Simulator/iraLIB so candidate policies can be compared with the existing controller and an untrained policy under repeatable conditions. Our primary learned candidate is a 512-wide, 10-hidden-layer network. We are training it on an AMD Instinct MI300X and preserving run data, checkpoints, and evaluation evidence. The goal is to determine whether learning improves movement reliability and efficiency—not to assume that a larger network or more GPU utilization guarantees better results. RoboProof is still experimental. Training is underway, and the learned policy has not yet passed the independent qualification gate. We are looking for robotics, controls, reinforcement-learning, and simulator contributors to help valid

ATLAS

ATLAS is not a simple chatbot; it is a universal AI orchestration system designed to manage any organization or startup. It acts as the "operating system" of a company, capable of planning, delegating tasks to specialized AI agents, and enforcing strict security protocols (Zero Trust). The Demo: Orchestrating Cloud Infrastructure For this hackathon, we demonstrate ATLAS orchestrating a complex cloud infrastructure: ZTC Hosting. Managing web hosting requires juggling multiple APIs (cPanel/WHM, billing, DNS) and handling critical incidents. ATLAS acts as the central brain. It analyzes events, delegates tasks to specialized agents (Support, Network Operations, Provisioning), and uses secure, typed API tools to execute operations. Real-time 3D Digital Twin To make this complex AI activity understandable to humans, ATLAS projects the entire organization into a real-time 3D digital twin. Every AI agent has a 3D avatar. Their movements, statuses, and animations map directly to the actual infrastructure state. If an incident occurs, the NOC agent visibly investigates. Crucially, for high-risk actions, the agent waits for human approval (Human-in-the-loop). AMD Technology Integration To meet the hackathon's technical requirements, ATLAS's core reasoning engine (a high-performance Large Language Model) is hosted and served on AMD Developer Cloud using ROCm and vLLM. The heavy lifting of the AI inference is powered by AMD GPUs. This proves that ATLAS is a hardware-agnostic, production-ready architecture capable of running on AMD infrastructure without rewriting its core logic.

LiberaPack

LiberaPack è uno strumento per preparare gli ordini e scegliere come imballarli. Le tre sezioni Orders, Products e Boxes riuniscono gli ordini, i dati logistici dei prodotti e le scatole disponibili. Per ogni prodotto registra ingombro e peso; un algoritmo deterministico verifica orientamenti, spazio e vincoli di carico, poi produce un piano con il contenuto di ogni collo. Il magazzino può stampare riepiloghi di picking e documenti interni, mentre i dati mancanti o gli ordini cambiati vengono segnalati prima della preparazione. La sincronizzazione con Shopify importa catalogo e ordini nel flusso autorizzato; il collegamento ad altre piattaforme è una possibile estensione. Lerry, il copilota AI integrato con Evolus, aiuta a consultare il magazzino e a proporre regole, che il sistema valida prima di applicarle. Il workflow Evolus e il calcolo verificato sono già presenti nella demo. L’impiego di hardware AMD resta una fase futura da sperimentare e misurare

OPTEsc

In an emergency, people rely on the way they usually leave a building. When smoke or damage blocks that route, they hesitate, follow the crowd, or get trapped. OPTEsc lets safety teams see those moments before they happen. Start with a floor plan as an image, PDF or CAD file. OPTEsc reads the walls, doors, rooms, stairs and furniture and turns them into an interactive 3D building. You then run a fire drill or an earthquake and watch the scenario unfold: smoke spreads, walls crack, shelves tip over, doors jam and routes close. The simulated occupants are not all the same. They include older people, children with a parent, wheelchair users, people who can't see or hear well, fire wardens who help others, and people who panic or stop to grab their things. The results show how each group fared and where people got stuck. As conditions change, OPTEsc updates projected signs and gives short instructions drawn from real safety guidance, with their sources. An AI model running on AMD GPUs chooses from this vetted set of instructions. It never invents exits, because routing stays in deterministic code. Finally, OPTEsc exports a proposed projector and camera layout as a PDF, with a reason for every placement. It gives a site team something concrete to review. OPTEsc is a working prototype and does not replace a certified safety assessment.

CareCenter: AI Shift Handover on AMD

In disability and elderly care homes, the most dangerous moment of the day is the shift change. A night-shift carer may have recorded a fall at 5 a.m., a low blood sugar reading, or a resident who refused medication, but the incoming team has minutes to read dozens of scattered notes per resident. Important details get lost, and residents pay the price. CareCenter is an in-house care management system for disability and elderly care facilities. Staff record medications, vital signs, care tasks, observations and incidents for each resident, manage documents with expiry tracking, and administrators see every action in a plain-language audit log. The system is already in daily use at a real disability care facility. For this hackathon we added an AI Shift Handover Summary. With one click, a carer picks the last 12, 24 or 48 hours and receives a short, bullet-point handover: urgent items first, then medications given, vital signs, care performed and what the next shift must do. The summary is generated by an open model (DeepSeek-V4-Flash) running on AMD GPUs through AMD Token Factory, connected via an OpenAI-compatible provider layer. Because this is health data, privacy is built in. Before anything leaves the system, names, national ID numbers, phone numbers and dates are masked, and a final check blocks the request if any personal data remains. The model is instructed never to diagnose or invent information, staff notes are treated as data rather than instructions to resist prompt injection, every summary is logged without storing its content, and the UI clearly labels the output as AI-generated so carers verify it against the original records.