Product Overview

One-stop intelligent compute platform — compute, storage, APIs, and the LLM toolchain

Alaya NeW Cloud serves AI training, inference, agent development, and HPC (High-Performance Computing) workloads, providing a full-stack offering from low-level GPUs to the upper-layer model toolchain. This page lays out every product, what it does, and when to use it.

1. Product matrix

Core compute platforms

Product	What it is	When to use it
Cloud Container Instance (CCI)	Kubernetes-based serverless container service. Bring an image plus your business logic — no need to manage the underlying cluster.	Isolated, single-user GPU experimentation; rapid deployment and operations of inference services
Alaya Lab	Cloud-based one-stop AI development platform with VS Code and Jupyter as the entry points; integrates compute, storage, and training services.	Ready-to-use pre-integrated environment; unified development across local IDE, console, and logs
HyperTrain	Kubernetes-based serverless distributed training service. Submit a job and resources are allocated automatically; built-in intelligent scheduling and self-healing on failure.	Automated cluster and environment management; training stability and high availability
Virtual Kubernetes Service (VKS)	Unified scheduling for large-scale heterogeneous GPU compute, with cluster scheduling and automated operations.	Multi-node, multi-GPU training; full lifecycle management of training and inference
DKS (Dedicated Kubernetes Service)	Enterprise-grade, physically isolated Kubernetes clusters with full self-control and a graphical operations UI.	User-controlled physical resources; scenarios with strict data security and compliance requirements
DSC (Dedicated Slurm Cluster)	Standardized, highly available, elastically scaling Slurm-based compute service with integrated visual operations and multi-dimensional monitoring.	Standardized cloud HPC environments; large-scale parallel job scheduling

Development and toolchain

Product	What it is	When to use it
OpenClaw	Open-source AI Agent framework written in Node.js. Through a Gateway, it maps LLM inference into executable control over the local OS, file system, and APIs. Supports Feishu, DingTalk, WhatsApp, and other messaging tools as interaction front-ends; provides long-term memory and 24/7 background execution.	Upgrade AI from "conversational" to "actionable"; bridge AI with local OS-level permissions

Storage and image services

Product	What it is	When to use it
Mass storage	Cost-effective elastic shared file storage. Multi-protocol support, linear capacity scaling, distributed parallel architecture.	Multi-scenario integration; sharing and centralized management of storage resources across applications
High-performance storage	Parallel file storage with extreme throughput, global namespace, single-directory view, and sustained performance under massively concurrent access.	High-throughput training-data reads; enterprise-grade high-availability requirements
NAS storage	Elastic mounting of mass-storage or high-performance volumes with unified permission management; decouples compute from storage.	Pick mass storage (cost-effective) or high-performance storage (low latency) by scenario; unified shared storage with permissions
AI-accelerated storage	Cloud-native high-speed storage deeply optimized for LLM training. Intelligent caching plus multi-protocol access solves bottlenecks in massive file reads.	Multi-tier caching (memory / SSD / cluster) smoothly absorbs checkpoint write spikes
Object storage	Elastic object store for massive unstructured data. Flat structure with the standard S3 interface.	Data backup, disaster recovery, compliance; data sources for AI frameworks
Image registry	Fully managed container image management with shared access plus permission management; enterprise-grade encrypted storage and image isolation.	Sharing images within teams or across projects; rapid image-based deployment

API services

Product	What it is	When to use it
Open API	Wraps GPU compute, distributed scheduling, and container technology into standardized, programmable APIs.	Reduce AI training to API calls; serverless per-second billing
Model Service API	"Model as a Service" — wraps model deployment, operations, and scheduling as RESTful APIs.	Minute-level one-click deployment from model file to live inference
Image Registry API	Programmatic interface to the image registry.	Automated image-registry management

LLM development and inference

Product	What it is	When to use it
LM Lab	One-stop LLM development and training platform. Pre-integrates mainstream open-source models, distributed training environments, and visual tuning tools; supports multi-source data ingestion and full model lifecycle management.	Lower the barrier to LLM development; unified data and task management
Inference	Enterprise-grade model inference platform covering deployment, evaluation, release, and monitoring. Supports one-click deployment, version control, and elastic auto-scaling.	Bring inference services online in minutes; real-time monitoring and evaluation

2. Typical scenarios and product combinations

1. LLM training and fine-tuning

For iterative training workloads on large datasets (images, audio, text), use VKS or HyperTrain for Tensor Core accelerated cards; use LM Lab for data management, model training, task management, model management, and environment management — covering the full development pipeline.

2. AI model inference

For online inference workloads with strict requirements on response latency and stability, use VKS to receive inference requests in real time and balance load automatically; use Inference for deployment, compression, and inference of LLM, CV, and NLP applications. To call models directly via API, see AlayaCode (OpenAI / Anthropic compatible).

3. AI Agent development

Use Alaya Studio, which integrates LLMs, embedding models, and multi-modal model APIs. Fine-tune open-source models with your own data and publish agents as APIs. Common applications: customer service, medical diagnosis, financial services, education assistance, and smart home.

4. HPC (High-Performance Computing)

VKS elastic scalability matches the parallelism and data intensity of HPC workloads, adjusting compute resource allocation in real time. Common domains: image processing and computer vision, engineering and industrial applications, machine learning, etc.

3. Get started

Register: visit the Alaya NeW website, click Sign Up, and fill in the basic information. See Sign up for the full procedure.
Real-name verification: complete enterprise or individual real-name verification per the prompts to unlock full platform privileges.
Top up: top up the account to activate it and start using all services.