Does amicu replace SkyPilot?

No. SkyPilot covers a much wider range of AI workloads and infrastructure. amicu focuses on declaratively deploying LLM servers.

Which providers does amicu support compared with SkyPilot?

amicu runs on this machine and deploys to Modal and RunPod today. Hetzner, OVHcloud, Scaleway and SSH hosts are in preview, and AWS, Azure, GCP and Nebius are planned. SkyPilot lists many more clouds, plus Kubernetes and Slurm.

Yes. They are independent tools. Use SkyPilot for jobs and clusters and amicu where you want a declarative LLM deployment.

amicu vs SkyPilot

SkyPilot is a broad system for running AI workloads (clusters, jobs and services) across many clouds, Kubernetes and Slurm. amicu is narrower: a declarative CLI that deploys vLLM or llama.cpp servers from HCL with a plan you review, recorded state and pre-flight checks. If you need training, batch jobs or the widest choice of infrastructure, SkyPilot covers far more ground today. If you want a reviewable, Terraform-style workflow for open-weight model endpoints on your machine or on GPU providers such as Modal and RunPod, amicu is built for that.

Quickstart All comparisons

Side by side

amicu vs SkyPilot
	SkyPilot	amicu
Scope	AI workloads in general: clusters, jobs and services.	LLM inference servers only.
Infrastructure	Many clouds, Kubernetes, Slurm and existing machines over SSH.	This machine, plus Modal and RunPod. Other providers are preview or planned.
Interface	CLI and Python API; tasks can be described in a YAML file.	CLI driven by declarative HCL with validate, plan, apply and destroy.
Plan before change	Shows the chosen resources and asks for confirmation on launch; no diff against recorded state.	plan shows a diff against recorded state; apply asks for approval.
Serving	Services with replicas across regions, clouds, clusters, pricing models or GPU types.	Deployments of vLLM or llama.cpp exposing an OpenAI-compatible endpoint.
Model fit checks	You choose the accelerator and your own serving setup.	validate checks that weights plus KV cache fit the accelerator or host memory.
Cost control	Clusters can autostop or tear down when idle; SkyServe replicas can scale to zero.	RunPod serverless endpoints scale to zero by default; budgets are recorded but not enforced.
Training and batch jobs	Supported through jobs and clusters.	Not a goal.
Maturity	Broader provider coverage and a larger scope.	Early and not stable; breaking changes are expected.
License	Apache-2.0.	AGPL-3.0-only.

Sources for the SkyPilot column: SkyPilot overview, SkyPilot installation. Details can change; check the official documentation.

When to choose SkyPilot

You need training, fine-tuning or batch jobs alongside serving.
You want to spread work across many clouds, Kubernetes clusters or Slurm.
You need capabilities amicu does not have, such as multi-cloud failover.

When to choose amicu

You only need open-weight LLM endpoints and want a plan-and-apply workflow for them.
You want policy checks (residency, provider allowlists) before anything starts and an endpoint auth check after deploy.
You want one small CLI whose state records how to stop each deployment, so destroy knows what to stop.

Questions

Does amicu replace SkyPilot?: No. SkyPilot covers a much wider range of AI workloads and infrastructure. amicu focuses on declaratively deploying LLM servers.
Which providers does amicu support compared with SkyPilot?: amicu runs on this machine and deploys to Modal and RunPod today. Hetzner, OVHcloud, Scaleway and SSH hosts are in preview, and AWS, Azure, GCP and Nebius are planned. SkyPilot lists many more clouds, plus Kubernetes and Slurm.
Can I use both?: Yes. They are independent tools. Use SkyPilot for jobs and clusters and amicu where you want a declarative LLM deployment.

Try amicu

Install the CLI and deploy a model on your machine or a GPU provider.

Quickstart Providers