GPUStack
gpustack.ai
- Free plan
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What GPUStack is
GPUStack is an open-source GPU cluster manager for deploying and operating AI models across on-premise, cloud, or hybrid infrastructure. It provides models through OpenAI-compatible and Anthropic-compatible endpoints, and supports inference engines such as vLLM, SGLang, llama.cpp, TensorRT-LLM, and MindIE. Users can browse and download models from Hugging Face or ModelScope, or supply local files; the platform checks compatibility automatically. It can provision GPU instances with lifecycle controls, partitioning, SSH key injection, Jupyter access, and persistent S3 or NFS storage. Deployment options include on-premise, Kubernetes, and multi-cloud environments, and the maker says air-gapped deployments are supported. Listed accelerators span NVIDIA, AMD, Ascend, T-head, Hygon, MetaX, Moore Threads, Cambricon, and Iluvatar. Security controls include role-based access, multi-tenancy, SSO, scoped API keys, IP allowlisting, quotas, and usage analytics. GPUStack Community Edition is free. Worker nodes support Linux only; Windows users are directed to consider WSL2, and macOS is not supported for worker nodes.
Who it suits
GPUStack suits teams deploying AI models across on-premise, cloud, or hybrid GPU infrastructure. Its Linux-only worker requirement makes it a poor fit for teams that need macOS worker nodes.
Good to know, and what to watch
- Open-source Community Edition is free.
- Supports Kubernetes and multi-cloud deployments.
- Offers OpenAI- and Anthropic-compatible endpoints.
- Includes role-based access and usage analytics.
- Can provision GPU instances with persistent storage.
- Worker nodes support Linux only.
- macOS worker nodes are not supported.
- Enterprise Edition pricing requires contacting the company.
Plans
Every plan and priceGPUStack Community Edition
Free plan
Getting started
- Open the GPUStack website
- Choose the free GPUStack Community Edition or contact the company about Enterprise Edition
- Deploy GPUStack with Linux worker nodes
- Connect models from Hugging Face, ModelScope or local files
- Configure a supported inference engine and GPU deployment
- Expose models through the OpenAI-compatible or Anthropic-compatible endpoint
- Add listed integrations such as Kubernetes, Docker, OpenWebUI or LangChain
The Club's review
GPUStack is an open-source cluster manager for deploying AI models and allocating GPU resources across on-premise, cloud, and hybrid infrastructure. It suits organizations that need to serve models across varied accelerators and govern access to shared GPU capacity. The broad deployment and security controls are compelling, but Linux-only worker nodes rule out native Windows and macOS workers.
Overview
GPUStack brings model serving and GPU provisioning into a single management layer. Interactive scheduling, quota controls, utilization metrics, Kubernetes support, and cloud GPU support make it relevant to teams running private infrastructure as well as those combining it with cloud resources. Its hybrid deployment model can accommodate different environments, though operators should confirm that their worker-node setup meets the Linux requirement before committing.
Models can be exposed through OpenAI-compatible and Anthropic-compatible APIs. GPUStack supports vLLM, SGLang, llama.cpp, TensorRT-LLM, and MindIE, and can source models from Hugging Face or ModelScope or use local files. Automated compatibility checks are useful when working with different model and accelerator combinations.
Key features
The hardware list covers NVIDIA, AMD, Ascend, T-head, Hygon, MetaX, Moore Threads, Cambricon, and Iluvatar accelerators. This is a practical advantage for teams with mixed hardware or a need to avoid tying their cluster to one accelerator family. Integrations include OpenWebUI, LangChain, n8n, Dify, RAGFlow, Claude Code, OpenClaw, Docker, Podman, Kubernetes, Helm, Prometheus, and Grafana.
GPU services include lifecycle controls, GPU partitioning, SSH key injection, Jupyter access, and persistent S3 or NFS storage. Together these features give administrators tools to allocate GPU-backed environments while retaining control over access and persistence. Deployments can run on-premise, on Kubernetes, or across multiple clouds; air-gapped deployment is also supported, an important option for environments that cannot depend on external connectivity.
Access controls include role-based access control, multi-tenancy, single sign-on through OIDC, SAML, or AD-LDAP, scoped API keys, IP allowlisting, token quotas, and usage analytics. Enterprise adds immutable audit logs, CIDR-based allowlists and blocklists, and per-model and per-key access rules. Its control-plane and model-service high availability includes automatic failover and traffic rerouting, making Enterprise more relevant where service continuity and tighter governance are priorities.
Pricing
GPUStack Community Edition costs nothing and is an open-source GPU cluster manager. It is the natural starting point for teams that can operate the software themselves and need its core cluster-management capabilities without a license charge. The published plan information does not give a seat or usage cap, so teams should validate operational fit against their intended deployment.
GPUStack Enterprise has custom pricing, and licensing requires contacting the company. It is the option for organizations that need the additional audit, access-rule, and high-availability capabilities. The company offers personalized demos and architecture, deployment, and scaling guidance, with replies typically arriving within one business day.
Platforms
GPUStack is listed for API, Linux, macOS, self-hosted, web, and Windows use. That platform breadth does not mean every operating system can host a worker: worker nodes support Linux only. Windows users are directed to consider WSL2, while macOS is not supported for worker nodes.
Who it's for
GPUStack is best suited to teams that need centralized model serving and GPU allocation across on-premise, cloud, or hybrid infrastructure, particularly where hardware is mixed or access needs to be governed by quotas and roles. It is less suitable for organizations that require native Windows or macOS worker nodes, or that want a commercial Enterprise license with a published price.
Pros and cons
- Pro: Supports multiple inference engines and OpenAI- and Anthropic-compatible APIs, giving teams flexibility in how they serve models and connect applications.
- Pro: Broad accelerator support, GPU partitioning, and persistent storage options suit shared clusters with varied hardware and workloads.
- Pro: Air-gapped deployment and extensive access controls address needs in tightly governed environments.
- Con: Worker nodes require Linux, limiting native deployment choices for Windows and macOS teams.
- Con: Enterprise pricing and licensing require a company contact, so organizations cannot assess its commercial cost from a published rate.
Alternatives
Backend.AI is another option with a free, self-hosted Open-source plan for home users, plus a free Try-out plan; consider it when those plan choices better match how you want to start. dstack offers a free self-hosted OSS stack and a GPU marketplace billed by usage, making it worth considering if marketplace access is a priority.
ClearML has a free Community plan and describes its self-hosted version as fully open source, so it may suit teams considering that approach. Kueue is a free open-source Kubernetes project, an alternative for readers focused on that category. OpenPBS is a free open-source option under the AGPL 3.0 license, with community forum support that carries no guarantees. TORQUE Resource Manager is another free open-source project, subject to the PBS license. HAMi is free GPU virtualization middleware for AI workloads on Kubernetes. NVIDIA ShadowPlay is a free Windows option with a supported-GPU and driver requirement.
Verdict
Choose GPUStack if you need an open-source way to manage model serving and GPU resources across a varied, potentially hybrid infrastructure, and can run Linux workers. Its combination of engine choice, hardware coverage, provisioning controls, and access governance is the main reason to choose it. Look elsewhere if Linux workers do not fit your environment or if you need Enterprise pricing to be clear before contacting the company.
The verdict
GPUStack brings model serving, GPU provisioning, and access controls into an open-source cluster manager. Confirm that Linux worker nodes and your deployment setup fit your requirements; Enterprise licensing and pricing require contacting the company.
Questions people ask
Is GPUStack free?
GPUStack Community Edition is listed as free and described as an open-source GPU cluster manager. Enterprise Edition pricing requires contacting the company.
Which operating systems support GPUStack workers?
Worker nodes support Linux only. Windows users are directed to consider WSL2, and macOS is not supported for worker nodes.
Which inference engines are supported?
Listed engines include vLLM, SGLang, llama.cpp, TensorRT-LLM and MindIE.
Can GPUStack run in air-gapped environments?
Yes. The maker states that air-gapped deployments are supported, alongside on-premise, Kubernetes and multi-cloud deployments.
What model sources can users connect?
Users can browse and download models from Hugging Face and ModelScope or provide local model files, with automated compatibility checks.
What security controls are available?
Listed controls include RBAC, multi-tenancy, OIDC/SAML/AD-LDAP SSO, scoped API keys, IP allowlisting, token quotas and usage analytics. Enterprise adds audit logs and additional access rules.
What the maker publishes
- Deployment
- It supports on-premise, Kubernetes, and multi-cloud GPU deployments, and the maker states that air-gapped deployments are supported. source
- Enterprise availability
- GPUStack Enterprise describes control-plane and model-service high availability with automatic failover and traffic rerouting. source
- Enterprise licensing
- The contact page directs users to contact the company for pricing and Enterprise Edition licensing. source
- Enterprise security
- GPUStack Enterprise describes immutable audit logs, CIDR-based IP allowlists and blocklists, and per-model and per-API-key access rules. source
- GPU service
- GPUStack can provision GPU instances with lifecycle controls, GPU partitioning, SSH key injection, Jupyter access, and persistent S3 or NFS storage. source
- Headquarters
- Shenzhen, Guangdong, China source
- Inference APIs
- It exposes models through OpenAI-compatible and Anthropic-compatible endpoints. source
- Inference engines
- The platform supports inference engines including vLLM, SGLang, llama.cpp, TensorRT-LLM, and MindIE. source
- Integrations
- Listed integrations include OpenWebUI, LangChain, n8n, Dify, RAGFlow, Claude Code, OpenClaw, Docker, Podman, Kubernetes, Helm, Prometheus, and Grafana. source
- Linux worker requirement
- The Docker installation page says GPUStack worker nodes support Linux only; Windows users are directed to consider WSL2, and macOS is not supported for worker nodes. source
- Model sources
- Users can browse and download models from Hugging Face and ModelScope or provide local model files, with automated compatibility checks. source
- Security
- The site lists RBAC, multi-tenancy, OIDC/SAML/AD-LDAP SSO, scoped API keys, IP allowlisting, token quotas, and usage analytics. source
- Support
- The contact page offers personalized demos and architecture, deployment, and scaling guidance, and says replies typically arrive within one business day. source
- Supported hardware
- The maker lists NVIDIA, AMD, Ascend, T-head, Hygon, MetaX, Moore Threads, Cambricon, and Iluvatar accelerators. source
- What it does
- GPUStack is an open-source GPU cluster manager for deploying, governing, and scaling AI models across on-premise, cloud, or hybrid infrastructure. source
Checked against the maker's pages on 29 Sept 2026.