dstackai/dstack

Health v2 · maintenance only

Adoption not part of the score

2221 stars · 251 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

dstack is an open-source, vendor-agnostic control plane for GPU provisioning and orchestration that works across GPU clouds, Kubernetes, and on-prem clusters via SSH. It supports NVIDIA, AMD, Google TPU, and Tenstorrent accelerators and streamlines dev environments, training tasks, and inference services through declarative YAML configurations applied via CLI or API.

Use cases

• provision GPU instances across multiple cloud providers from one control plane
• run distributed LLM fine-tuning jobs on H100 or MI300X clusters
• deploy model inference endpoints with auto-scaling and ingress
• spin up cloud dev environments accessible from VS Code or Cursor via SSH
• manage on-prem GPU servers as SSH fleets without Kubernetes
• orchestrate agentic workloads and batch jobs on spot GPU instances
• benchmark and optimize LLM inference configurations with agent-driven presets

When to choose

• you need multi-cloud or hybrid GPU orchestration without vendor lock-in
• you want declarative YAML-based provisioning for training and inference workloads
• you mix cloud providers, Kubernetes, and bare-metal SSH servers
• you need support for diverse accelerators including AMD, TPU, and Tenstorrent

When to avoid

• you only run CPU-only workloads with no GPU needs
• you need a fully managed service and prefer not to self-host a control plane server
• your orchestration needs are simple single-machine Docker deployments

Facets

framework · maturity active

container-orchestration deployment scheduling llm-inference machine-learning gpu-computing cli api-framework machine-learning deep-learning large-language-models cloud-computing infrastructure-as-code gpu-computing windows python cli self-hosted cloud gpu-orchestration gpu-cloud training fine-tuning inference-endpoints fleets ssh-fleets slurm tpu tenstorrent amd nvidia control-plane dev-environments agent-skills containers devops ai-agents linux macos docker kubernetes

Member repositories

For agents

markdown · JSON · MCP: product_card(name="dstackai/dstack")