dstack
dstack.ai
- Free plan
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What dstack is
dstack is a free, open-source orchestration layer for AI workloads across GPU clouds, Kubernetes, virtual machines, and bare-metal clusters. YAML configurations define fleets, development environments, tasks, services, presets, and volumes, while dstack handles infrastructure provisioning and job scheduling, including auto-scaling, port forwarding, and ingress. Supported accelerators include NVIDIA, AMD, TPU, and Tenstorrent. Documented backends include AWS, Azure, GCP, Kubernetes, GPU cloud providers, remote SSH hosts, and experimental Slurm. Tasks can use frameworks such as accelerate, torchrun, Ray, and Spark. Users manage resources with the CLI or HTTP API, and the server can run on a laptop or another environment able to access the clusters in use. Services can publish inference endpoints through gateways with HTTPS, custom domains, auto-scaling, and rate limits. The open-source plan is self-hosted. Server data and project secrets are stored in plaintext by default unless an administrator configures AES-256-GCM encryption. TPU support is limited to single-host instances of up to eight cores.
Who it suits
dstack suits teams orchestrating AI workloads across cloud and on-premises infrastructure, including GPU clouds, Kubernetes, and bare-metal clusters. It is relevant to users comfortable managing a self-hosted stack through YAML, a CLI, or an HTTP API.
Good to know, and what to watch
- Free, open-source self-hosted orchestration stack
- Supports NVIDIA, AMD, TPU, and Tenstorrent accelerators
- Works with Kubernetes, VMs, GPU clouds, and bare metal
- CLI and HTTP API management options
- Inference gateways support HTTPS and custom domains
- Server data is plaintext by default
- Project secrets are plaintext by default
- TPU support is limited to single-host instances
- Single-host TPU instances are capped at eight cores
Plans
Every plan and pricedstack Sky GPU Marketplace
dstack OSS
Free plan
Getting started
- Open https://dstack.ai/
- Install or run the self-hosted dstack server on a laptop or reachable environment
- Connect AWS, Azure, GCP, Kubernetes, GPU cloud, SSH, or Slurm backends
- Define fleets, environments, tasks, services, presets, or volumes in YAML
- Manage resources with the dstack CLI or HTTP API
Limits to check
The free dstack OSS plan is self-hosted. TPU workloads are limited to single-host instances with a maximum of eight cores per TPU instance; GPU Marketplace usage uses prepaid credits and variable GPU-hour pricing.
The Club's review
Overview
dstack is an open-source orchestration layer for teams running AI workloads across cloud and on-premises infrastructure. It suits operators who need one workflow for varied accelerators and environments; its breadth is appealing, but plaintext defaults and the absence of quota controls call for deliberate administration.
The server can run on a laptop or another reachable environment, while workloads are defined in YAML as fleets, development environments, tasks, services, presets, or volumes. Teams can manage resources through the CLI or HTTP API. This is infrastructure software, not a general-purpose desktop app, despite platform support spanning API, Linux, macOS, self-hosted, web, and Windows.
Key features
Provisioning and scheduling
dstack provisions infrastructure and schedules jobs across GPU clouds, Kubernetes, virtual machines, and bare-metal clusters. Auto-scaling, port forwarding, and ingress extend the workflow to services as well as scheduled tasks. GPU utilization metrics help operators monitor compute, but the lack of quota controls leaves teams without an in-product way to enforce resource allocations.
Accelerators and frameworks
NVIDIA, AMD, TPU, and Tenstorrent accelerators are supported out of the box. The task guidance names Accelerate, torchrun, Ray, and Spark for distributed workloads, while the maker describes compatibility with any hardware, open-source tools, and frameworks. That range makes dstack relevant to mixed environments, though TPU support is limited to single-host instances of up to eight cores.
Services, interfaces, and integrations
Services can be exposed as inference endpoints through gateways with HTTPS, auto-scaling, custom domains, and rate limits. Users who need automation beyond the CLI can call the HTTP API directly. Documented backends include AWS, Azure, GCP, Kubernetes, multiple GPU cloud providers, and remote SSH hosts; Slurm support is experimental, so it is a less settled choice for Slurm-dependent operations.
Security and support
Server data and project-scoped secrets are stored in plaintext by default. Administrators can enable AES-256-GCM encryption for stored data; project admins manage secrets. This flexibility puts the security decision with the operator, but teams should configure encryption rather than assume sensitive values are protected by default. Documentation points users to GitHub for issues and Discord for questions.
Pricing
dstack OSS
The dstack OSS plan costs nothing and provides the open-source orchestration stack for self-hosting. It is the natural fit for teams that can operate their own control plane and infrastructure, with no paid tier required for the core orchestration stack.
dstack Sky GPU Marketplace
dstack Sky Marketplace has custom pricing: GPU use is pay-as-you-go, billed per GPU-hour, with rates varying by provider and GPU. Compute is prepaid through credits, and the console shows resource prices before provisioning. On-demand and spot GPU compute are offered. The BYOC mode currently carries no charge, while Marketplace use requires budgeting against changing provider rates rather than a fixed subscription.
dstack Factory
dstack Factory is the commercial extension, adding advanced multi-tenancy, usage metering, billing automation, and optimized inference presets for frontier open models. It has custom pricing and is a better fit for organizations that need those operational and commercial controls than for users seeking only open-source orchestration.
Platforms
dstack supports API, Linux, macOS, self-hosted, web, and Windows environments. Its hybrid deployment model lets the server reach the cloud and on-premises clusters being used; Kubernetes support and cloud GPU support broaden its infrastructure options.
Who it's for
dstack is best suited to AI teams and infrastructure operators coordinating workloads across different providers, accelerators, and on-premises systems. YAML configuration, CLI and API management, and support for both tasks and services make it useful from development through inference deployment. It is less suitable for organizations that need built-in quotas, default encrypted storage, or multi-host TPU support.
Pros and cons
- Pros: Broad infrastructure coverage brings GPU clouds, Kubernetes, VMs, and bare-metal clusters into one provisioning and scheduling layer.
- Pros: NVIDIA, AMD, TPU, and Tenstorrent support, plus named distributed frameworks, serve teams with mixed AI workloads.
- Pros: Free self-hosted OSS and uncharged BYOC mode give teams options to use the orchestration layer without a platform fee.
- Cons: Plaintext-by-default server data and secrets require administrators to configure encryption for stored data.
- Cons: No quota controls means teams must handle resource limits outside dstack.
- Cons: TPU support is restricted to single-host instances with at most eight cores.
- Cons: Marketplace GPU costs vary by provider and GPU, and prepaid credits require attention to usage before provisioning.
Alternatives
GPU Cluster Management Software is a useful category comparison for readers weighing cluster-management options.
- Backend.AI offers a free home-user open-source plan and a free trial, making it worth comparing for users who want to try a freemium platform.
- ClearML has a free Community plan and a self-hosted version its maker describes as fully open source; consider it when those terms align with your deployment preference.
- GPUStack is a free, open-source GPU cluster manager for readers seeking a focused cluster-management option.
- NVIDIA ShadowPlay is a free Windows app with a supported GeForce GPU and driver requirement, rather than a cross-infrastructure orchestration layer.
- Kueue is a free open-source Kubernetes project for readers whose needs center on Kubernetes.
- OpenPBS is a free open-source option under AGPL 3.0, with community forum support that carries no guarantees.
- TORQUE Resource Manager is a free open-source project whose use, modification, and distribution are subject to the PBS license.
- HAMi is free open-source GPU virtualization middleware for AI workloads on Kubernetes.
Verdict
Choose dstack if your team needs a free, flexible orchestration layer to provision and schedule AI workloads across heterogeneous cloud and on-premises infrastructure. Its accelerator range, YAML workload model, and API make it compelling for mixed environments. Look elsewhere if enforced quotas, encrypted storage by default, or multi-host TPU support are essential.
The verdict
dstack provides a free orchestration layer for AI workloads across heterogeneous infrastructure, with YAML configuration and CLI or API management. Administrators should account for its plaintext-by-default data and secrets, and for the stated single-host TPU limit.
Questions people ask
Is dstack free?
The dstack OSS plan costs nothing. dstack Sky GPU Marketplace usage is pay-as-you-go against prepaid credits, with prices shown before provisioning.
Which platforms does dstack support?
Listed platforms are API, Linux, macOS, self-hosted, web, and Windows.
Which infrastructure backends are documented?
Documented backends include AWS, Azure, GCP, Kubernetes, GPU cloud providers, remote SSH hosts, and an experimental Slurm backend.
How can resources be managed?
Users can manage resources with the dstack CLI or call the HTTP API.
Are stored secrets encrypted?
Server data and project secrets are stored in plaintext by default. Administrators can configure AES-256-GCM encryption for stored data.
What the maker publishes
- Accelerators
- dstack supports NVIDIA, AMD, TPU, and Tenstorrent accelerators out of the box. source
- API
- dstack offers an HTTP API for functionality not available in the CLI and for integrations that need to call the server directly. source
- Commercial offering
- dstack Factory extends the open-source product with advanced multi-tenancy, usage metering, billing automation, and optimized inference presets for frontier open models. source
- Company
- The terms identify dstack Inc. as a Delaware corporation with offices in Dover, Delaware, United States. source
- Cost model
- dstack Sky does not currently charge for BYOC mode; GPU Marketplace usage is prepaid and resource prices are shown in the console before provisioning. source
- Deployment
- The server can run on a laptop or another environment with access to the cloud and on-prem clusters being used. source
- Deployment limit
- The TPU guide says dstack currently supports single-host TPUs only, with a maximum of eight cores per TPU instance. source
- Framework compatibility
- The maker describes dstack as compatible with any hardware, open-source tools, and frameworks. source
- Frameworks
- The tasks guide names accelerate, torchrun, Ray, and Spark as distributed frameworks that work with dstack. source
- Hosted pricing
- dstack Sky Marketplace pricing is dynamic by provider, shown in the console before provisioning, and billed against prepaid credits. source
- Inference
- Services can deploy model inference as endpoints, and gateways support HTTPS, auto-scaling, custom domains, and rate limits. source
- Integrations
- Documented backends include AWS, Azure, GCP, Kubernetes, multiple GPU cloud providers, remote SSH hosts, and an experimental Slurm backend. source
- Integrations
- Listed backend types include AWS, Azure, GCP, Kubernetes, Slurm, Runpod, Nebius, Lambda, and other cloud providers. source
- Interfaces
- Users can manage resources with the dstack CLI or call its HTTP API. source
- Platforms
- The CLI runs on Linux, macOS, and Windows; the server can be installed on those systems, with Windows using WSL 2. source
- Provisioning
- dstack manages infrastructure provisioning and job scheduling, including auto-scaling, port forwarding, and ingress. source
- Purpose
- dstack is an open-source orchestration layer for AI workloads on heterogeneous accelerators, including GPU clouds, Kubernetes, VMs, and bare-metal clusters. source
- Secrets
- Secrets are project-scoped, managed by project admins, and stored in plaintext by default unless server encryption is configured. source
- Security
- Server data is stored in plaintext by default; administrators can configure AES-256-GCM encryption for stored data. source
- Security
- The service configuration supports authorization, which is enabled by default for services. source
- Service endpoints
- Services can be published with HTTPS, custom domains, auto-scaling, and rate limits through gateways. source
- Support
- The documentation directs users to report issues on GitHub and ask questions in the dstack Discord server. source
- Support
- The dstack Sky page directs users with questions or needing help to Discord or to contact the company directly. source
- What it does
- dstack is an open-source orchestration layer for AI workloads across GPU clouds, Kubernetes, VMs, and bare-metal clusters. source
Checked against the maker's pages on 30 Sept 2026.