Skip to content
Get packs

Guide · Agent skill

How to use the Lambda Labs GPU Cloud agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The Lambda Labs GPU Cloud skill teaches an AI agent how to rent and manage GPU instances on Lambda Labs, a paid cloud service that bills by the minute for graphics processors used in machine learning training and inference. This guide is for people who have never installed an agent skill before. By the end you will know what the skill lets the agent do, how to install it, how to ask for things in plain language, and what to check before the agent spends money on your behalf.

Lambda Labs GPU CloudOrchestra-Research/AI-Research-SKILLs · MIT

Quick start

  1. 1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, upload it under Settings, Capabilities, Skills instead.
  2. 2Create your Lambda Labs account. Sign up, add a payment method, and generate an API key from the dashboard.
  3. 3Set the API key and register an SSH key. Store the key in the LAMBDA_API_KEY environment variable and add an SSH key to your Lambda account before launching anything.
  4. 4Ask for what you need in plain language. Describe the GPU, region and purpose, for example a single H100 for a training run with your filesystem attached.
  5. 5Confirm before launching or terminating. Check the agent's proposed action, including the instance ID for any termination, before it runs, since billing is per minute.
  6. 6Check running instances. List your instances at the end of a session to make sure nothing is still running and charging.

What the skill makes the agent do

Once installed, the agent knows which GPU types Lambda offers and what each one costs per GPU per hour, how to create an account and register an SSH key, and how to launch, list and terminate instances. It can make these calls either through the lambda-cloud-client Python package or through curl commands against Lambda's API.

The skill also covers persistent filesystems, which keep data across instance restarts, and Lambda Stack, the pre-installed set of PyTorch, CUDA, NCCL and related tools on each instance. The skill does not rent or pay for anything itself. It gives the agent the know-how to call Lambda's API with your own credentials.

  • Lists GPU types, VRAM and per-hour prices from the skill's reference table
  • Launches, lists and terminates instances via the Python client or curl
  • Explains SSH access, SSH tunnelling for Jupyter and TensorBoard, and filesystem mounts
  • Covers multi-node distributed training patterns for larger clusters

Install the skill in Claude

Download the skill's ZIP file from this page. For Claude Code, unzip it into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill to apply only to that project.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there. Either way, the folder should contain SKILL.md along with the references folder (advanced-usage.md and troubleshooting.md).

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your project
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up your Lambda Labs account before you start

The skill assumes you already have a Lambda Labs account with a payment method added. Generate an API key from the dashboard and store it in an environment variable named LAMBDA_API_KEY, since both the Python and curl examples read it from there.

You also need an SSH key registered with Lambda, because the console and the API require one before an instance can be launched. The skill shows how to generate one locally with ssh-keygen and add the public key to your account.

  • Create an account and add a payment method
  • Generate an API key and set it as LAMBDA_API_KEY
  • Register an SSH key with Lambda (required before launching)
  • Install the lambda-cloud-client Python package if you want the Python examples
  • Have curl and jq available for the CLI examples
ssh-keygen -t ed25519 -f ~/.ssh/lambda_key

Ask the agent for GPU work in plain language

You do not need to type commands yourself. The skill's description says it applies when you need dedicated GPU instances with simple SSH access, persistent filesystems, or multi-node clusters for large-scale training. Requests that match those needs should lead the agent to use it.

Be specific about the GPU, region, and what should happen to the instance afterwards. Asking the agent to check first is a good habit, because launching bills per minute and an instance keeps charging until it is terminated.

  • Show me the GPU types Lambda has right now and their hourly prices.
  • Launch a single H100 instance in a US region with my SSH key and attach my training filesystem.
  • List my running instances and tell me their IPs and status.
  • Terminate instance <INSTANCE-ID>, but first show me its name and status so I can confirm.
  • Set up a 4-GPU instance for a long fine-tuning run and give me the SSH command.
curl -u $LAMBDA_API_KEY: https://cloud.lambdalabs.com/api/v1/instance-types | jq

Use persistent storage so checkpoints survive restarts

Filesystems keep data across instance restarts, and they are mounted at /lambda/nfs/ followed by the filesystem name. Store datasets, checkpoints and final models there. The local disk under /home/ubuntu is faster but temporary.

A filesystem must be attached when the instance is launched, and it must be in the same region as the instance. If you forget to attach one, you cannot add it later through this skill's examples, so check this before launching.

  • Create the filesystem under Storage in the Lambda console, in the same region as the instance
  • Attach it at launch, either in the console or with file_system_names in the API request
  • Save checkpoints to /lambda/nfs/<FILESYSTEM_NAME>/checkpoints

Stop instances safely and avoid surprise charges

Terminating an instance is the step that stops the billing, so the agent should always confirm the instance ID and name before running it. Lambda's API terminates instances through an operations endpoint that takes a list of instance IDs.

The skill's own guidance is to check the agent before it creates or terminates resources. Ask it to list running instances at the end of each session to confirm nothing is left running.

  • Ask the agent to list running instances before and after any launch or termination
  • Confirm the instance ID and name in your request
  • Keep training outputs on a persistent filesystem before terminating
curl -u $LAMBDA_API_KEY: -X POST https://cloud.lambdalabs.com/api/v1/instance-operations/terminate -H "Content-Type: application/json" -d '{"instance_ids": ["<INSTANCE-ID>"]}' | jq

Know what the skill needs and what it does on your machine

The skill makes network calls to Lambda's API using your key, and it needs curl, jq or the Python client depending on the method the agent uses. According to its metadata, it does not run scripts on your machine. Its content is documentation and examples, licensed under MIT.

This file does not cover Lambda's security or data policies. If you plan to handle sensitive data on shared infrastructure, read Lambda's own policies first and review every API call the agent proposes before it runs.

  • Network access: yes, to Lambda's API with your LAMBDA_API_KEY
  • Runs scripts: no, according to the skill metadata
  • Keep your API key out of shared project files and chat logs
  • Review launch and terminate requests before approving them

What people use it for

  • Launching a single H100 or A100 instance for a multi-hour fine-tuning run and connecting over SSH
  • Saving model checkpoints to a persistent filesystem so training can resume after a restart
  • Scripting the launch of several instances and terminating them when a batch job finishes
  • Forwarding Jupyter or TensorBoard ports over an SSH tunnel to a remote GPU
  • Starting a multi-GPU or multi-node training job with PyTorch DDP on a pre-configured Lambda Stack

Frequently asked questions

How do I install the Lambda Labs GPU Cloud skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. If you use the Claude apps, upload the ZIP under Settings, Capabilities, Skills.

Is the Lambda Labs GPU Cloud skill free?

The skill is MIT licensed, but Lambda Labs itself is a paid service that bills per minute for GPU instances. You need a payment method on your Lambda account, and the skill does not pay for anything on your behalf.

What do I need before the skill works?

You need a Lambda Labs account with a payment method, an API key stored in LAMBDA_API_KEY, and an SSH key registered with Lambda. The Python examples need the lambda-cloud-client package, and the CLI examples need curl and jq.

What if Lambda has no capacity for the GPU I want?

The troubleshooting reference suggests checking availability through the instance-types API, trying other regions, or trying a different GPU type such as an A100 in place of an H100. If an instance stays in the booting state much longer than the expected launch time, the guide suggests terminating it and trying another region or type.

Is Lambda Labs cheaper than RunPod or Vast.ai?

The skill does not claim that. It points to Modal for serverless, auto-scaling workloads, to RunPod for cheaper spot instances and serverless endpoints, and to Vast.ai for a GPU marketplace with the lowest prices. Compare current prices on each provider before choosing.

Can the agent accidentally leave an instance running?

It can if no one checks. Launching bills per minute and an instance keeps charging until terminated, so ask the agent to list running instances and confirm each termination by instance ID.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Choose this skill if you need dedicated GPUs with full SSH access for long training runs and want an agent to handle the launch and shutdown steps. Pick a serverless or spot option like Modal, RunPod or Vast.ai if you need per-second serverless billing or the lowest prices, and review Lambda's security policies before using it with sensitive data.

Download the skill