VESSL Cloud

Stay in your editor,
run on cloud GPUs

Persistent GPU instances, one-command jobs, and the same environment from first experiment to production.

Get startedTalk to sales

See it in action

Watch VESSL Cloud work

Claude Code drives vesslctl on a real account: checking GPU availability, launching a job on an 8×H100 node, and verifying it in seconds.

University of California, Berkeley
“VESSL Cloud meaningfully reduces the time I spend on job wrangling (resource requests, environment quirks, monitoring) and shifts that time back into experiment design and analysis. In particular, reliable compute availability of VESSL Cloud allowed me to significantly reduce monitoring efforts with fire-and-forget.”
Joseph Suh, Researcher at Berkeley AI Research (BAIR)
NXN Labs
“We got an H100 cluster on VESSL Cloud through self-serve, without the usual procurement back-and-forth. The preconfigured Docker environments removed OS and driver conflicts, so experiments stopped dying to infrastructure issues. We now run more than twice as many every month, at nearly half the infrastructure cost.”
AI Research Engineer, NXN Labs

The GPU cloud with a workflow on top

Develop in a workspace, submit the same environment as a job, and scale to multi-node clusters when one node isn't enough.

Container Compute

Workspace

Spin up a single-node workspace with up to 8 GPUs, your image, and your data preloaded. Connect over SSH from your own editor, and pick up exactly where you left off. Close your laptop; the environment persists.

  • Up to 8 GPUs on one node (A100, H100, L40S)
  • SSH remote into your own IDE
  • Persistent environment, image, and volumes
Read the docs

Container Compute

Job

Submit a training run, evaluation, or preprocessing pipeline and let it run to completion. Jobs are non-interactive: bring your image, mount volumes, and they queue for capacity, stream logs and metrics, then free the GPU when done.

  • Training, fine-tuning, batch inference, and sweeps
  • Live logs and metrics; queues for GPU capacity
  • Billed only while running, then frees the GPU
Read the docs

Container Compute

From Workspace to Job, no rework

Prototype interactively in a workspace, then submit the exact same environment as a job to run to completion. No rebuild, no reconfiguration. The environment you debugged is the environment that runs.

Read the docs

Command line interface

vesslctl

vesslctl is the command line for VESSL Cloud. Run jobs and workspaces straight from your terminal. Then install the skill, and a coding agent like Claude Code drives GPU compute with the very same commands.

  • Run jobs and workspace from your terminal
  • Agent drive with vesslctl as GPU compute tool
  • Built to pair with AI coding tools like Claude Code
Read the docs
terminal

# Download and install vesslctl (macOS / Linux)

$ curl -fsSL https://api.cloud.vessl.ai/cli/install.sh | bash

# Teach your agent how to use it

$ vesslctl skill install

Need more?

Outgrown a single node?

Dedicated multi-node GPU clusters with InfiniBand and root SSH, for large-scale distributed training.

Explore VM Cluster

Storage that keeps up

Cluster and object storage attached to your workloads: fast where you train, cheap where you keep.

Explore Storage

From first experiment to full scale

01

Pick a GPU

Grab a GPU across clouds and start in minutes, billed by the second with no quota tickets to file.

02

Prototype in a Workspace

SSH from your own IDE and iterate live on a persistent GPU box. Close your laptop; the environment is waiting when you're back.

03

Hand off as a Job

Once the code works, submit the same environment as a batch job. Fire-and-forget, with logs and metrics captured for you.

04

Scale to a Cluster

Outgrown one node? Move to a dedicated multi-node cluster over InfiniBand for large distributed training.

Frequently Asked Questions

What is VESSL Cloud?

A GPU cloud that sources capacity across multiple clouds and puts a real workflow on top: persistent single-node workspaces, one-command jobs, and fast shared storage, with dedicated multi-node clusters available when you outgrow a single node. You pick a GPU and start in minutes, instead of managing quotas and per-cloud lock-in.

Which GPUs can I use self-serve right now?

H100, A100, and L40S are available self-serve. H200, B200, B300, and GB300, plus reserved capacity, are available by talking to our team.

Can I run inference on VESSL Cloud?

Yes. Run workspaces and jobs for serving as well, interactive or batch. For fully managed, optimized production endpoints, see Inference Endpoints.

How is VESSL Cloud different from RunPod or Lambda?

VESSL Cloud adds a workflow on top of the GPUs: persistent environments, interactive-to-batch with no rework, one-command jobs, and a CLI that agents can drive. You spend time on experiments, not infrastructure.

Can a single workspace or job use multiple GPUs?

Yes, within a single node. A workspace or job can use up to 8 GPUs on one node (for example, 8×H100). For training that spans multiple nodes over InfiniBand, use a dedicated VM Cluster.

Can I drive VESSL Cloud from an AI agent?

Yes. Install vesslctl, then run vesslctl skill install to give a coding agent like Claude Code the skill to provision GPUs and run jobs autonomously, using the same commands you do.

Do you offer managed Kubernetes or Slurm?

Both are on the roadmap. For dedicated multi-node capacity today, you can order a VM Cluster (beta).

Is my data secure?

VESSL AI is SOC 2 Type II certified. See our Trust Center at trust.vessl.ai for current controls and reports.

How does pricing work?

On-demand is billed as you go; reserved gives guaranteed capacity at a lower rate for committed use. See the pricing page.

Where AI models
get their GPUs.

Get StartedTalk to Sales
  • Start in minutes
  • Scale to multi-node clusters
  • Capacity reserved to your timeline
  • Dedicated support