Repolode
Submit
dstackai

dstackai/dstack

An open-source control plane that provisions and orchestrates GPU compute for training and inference across GPU clouds, Kubernetes, VMs, and on-prem clusters.

View on GitHubWebsite
Stars
2,284
Forks
270
Open issues
63
Last commit
10 hours ago

What is it?

dstack is a unified control plane for GPU provisioning and orchestration. It lets teams run development environments, training jobs, and inference services on GPU clouds, Kubernetes clusters, VMs, or bare-metal machines through one set of configurations, instead of handling each platform separately.

Workloads are defined as YAML files in a repository and applied with the dstack CLI or a programmatic API. The system handles provisioning, job queuing, auto-scaling, networking, volumes, port forwarding, and recovery from run failures and out-of-capacity errors. Accelerators from NVIDIA, AMD, Google TPU, and Tenstorrent are supported, and on-prem machines can be added through SSH fleets without configuring a cloud backend.

The project also ships agent skills so AI coding agents can create fleets and submit workloads through the CLI, and it integrates with tools such as NVIDIA Dynamo and Slurm.

Key features

  • 01Provisions and manages GPU fleets across clouds, Kubernetes, and on-prem machines.
  • 02Runs dev environments, training tasks, and inference services from YAML configurations.
  • 03Supports NVIDIA, AMD, Google TPU, and Tenstorrent accelerators.
  • 04Handles job queuing, auto-scaling, volumes, and port forwarding automatically.
  • 05Provides agent skills so AI coding agents can submit workloads via the CLI.

Use cases

  • Sharing a pool of GPUs across a research team on mixed cloud and on-prem hardware.
  • Deploying LLM inference as scalable endpoints on Kubernetes.
  • Running multi-node fine-tuning or training jobs without writing platform-specific scripts.
  • Letting AI agents provision compute and launch experiments on a team's behalf.

Quick start

  1. 1

    Install and start the server

    uv tool install "dstack[all]" -U
  2. 2

    Run the server

    dstack server
  3. 3

    Install the CLI (only if not installed with the server)

    uv tool install dstack -U
  4. 4

    Point the CLI to the server by adding a project

    dstack project add --name main --url http://127.0.0.1:3000 --token bbae0f28-d3dd-4820-bf61-8f4bb40815da

Things to know

  • Orchestrating cloud or Kubernetes compute requires configuring backends first.
  • The server needs Git and OpenSSH and runs on Linux, macOS, or Windows via WSL 2.

Topics

#gpu#orchestration#machine-learning#kubernetes#inference#training#llms#cloud#agent-skills#agentic-orchestration

FAQ

What is dstack?
An open-source control plane that provisions and orchestrates GPU compute for training and inference across GPU clouds, Kubernetes, VMs, and on-prem clusters. dstack is a unified control plane for GPU provisioning and orchestration. It lets teams run development environments, training jobs, and inference services on GPU clouds, Kubernetes clusters, VMs, or bare-metal machines through one set of configurations, instead of handling each platform separately.
Is dstack free to use? What license does it have?
dstack is open source under the MPL-2.0 license. Check the license terms before using it in commercial products.
Is dstack still maintained?
Actively maintained. The last commit to dstack was on Oct 20, 2026.
What can I use dstack for?
Sharing a pool of GPUs across a research team on mixed cloud and on-prem hardware. Deploying LLM inference as scalable endpoints on Kubernetes. Running multi-node fine-tuning or training jobs without writing platform-specific scripts. Letting AI agents provision compute and launch experiments on a team's behalf.
How do I install dstack?
Install and start the server: uv tool install "dstack[all]" -U Run the server: dstack server Install the CLI (only if not installed with the server): uv tool install dstack -U Point the CLI to the server by adding a project: dstack project add --name main --url http://127.0.0.1:3000 --token bbae0f28-d3dd-4820-bf61-8f4bb40815da
Where can I get dstack?
The source code and installation instructions are on GitHub: https://github.com/dstackai/dstack

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects