Meshive Python SDK & CLI

Rent GPUs on Meshive from your terminal or from Python. Launch pods, run serverless tasks, serve models and manage your files: the same things you can do in the console, scriptable.

Documentation · Console · GPU pricing · Use it from AI agents

Install

pip install meshive

Requires Python 3.10 or newer (tested on 3.10 to 3.13). This installs the meshive command and the meshive Python package. Check it with meshive --version.

In a requirements file, pin the major version so you get fixes but nothing that breaks your code:

meshive>=1,<2

Upgrading from a pre-1.0 release

1.0.0 works with code written for the 0.1 releases; no changes are needed. Change your pin to meshive>=1,<2 and reinstall, and regenerate your lock file if you use one. Most pre-1.0 releases are being removed from PyPI (see the changelog for dates), so a pin to one of them stops installing after its removal date. Environments where it is already installed keep working.

Quickstart

1. Get an API key. In the console, open your workspace's Settings → API keys. Pick Read & write so you can create pods.

2. Log in.

meshive login          # paste the key; it is saved for later commands
meshive workspaces     # note the ID of your workspace

3. Pick a GPU and a template.

meshive gpus                       # GPUs you can rent right now, with hourly prices
meshive templates --name jupyter   # note the template ID

4. Launch a pod.

meshive pod-create <workspace> my-first-pod --template <template-id> --gpu "RTX 3060" --wait running

The CLI shows the hourly price and asks before anything is billed. With --wait running it returns once the pod is up and prints its ID.

5. Open it.

meshive pod <workspace> <pod>      # the pod's URLs (Jupyter, ...) and its login, if it has one

6. Stop or delete it when you're done.

meshive pod-stop <workspace> <pod>     # compute billing stops; attached storage is still billed
meshive pod-delete <workspace> <pod>

Use it from Python

from meshive import Meshive, format_hourly

with Meshive() as client:  # uses the key saved by `meshive login`, or MESHIVE_API_KEY
    workspace = client.list_workspaces()[0].namespace_name

    for gpu in client.list_gpus():
        print(gpu.gpu_model, f"{gpu.vram} GB", format_hourly(gpu.price_per_hour), gpu.available_gpus)

    template = next(t for t in client.list_templates() if t.name == "Jupyter")
    estimate = client.estimate_pod("my-first-pod", template.template_id, workspace=workspace, gpu_model="RTX 3060")
    print("estimate:", format_hourly(estimate.price_per_hour))

    client.create_pod("my-first-pod", template.template_id, workspace=workspace, gpu_model="RTX 3060",
                      max_price_per_hour=0.10)  # never pay more than $0.10/hr for compute

    pod = client.wait_for_new_pod("my-first-pod", workspace, until="running")
    for endpoint in pod.endpoints:
        print(endpoint.name, endpoint.external_url)

    client.stop_pod(pod.pod_name, workspace)

AsyncMeshive has the same methods for async code, and the package ships type hints. Prices come back as the exact numbers the server sent; format_hourly and format_usd round them the way the console shows them.

Errors are exceptions you can catch by kind. All of them derive from MeshiveError:

from meshive import InsufficientCreditError, PodCreationFailedError, WaitTimeoutError

try:
    pod = client.wait_for_new_pod("my-first-pod", workspace, until="running")
except PodCreationFailedError as err:
    print("the pod could not start:", err)
except WaitTimeoutError:
    print("still not running; `meshive transactions <workspace>` shows what it is waiting for")

What else you can do

Every command has --help. The CLI reference lists them all.

Good to know

• Nothing is billed without asking. Commands that spend credit or delete something show the price or a summary first and ask you to confirm. Add --estimate to see the price only, or --yes to skip the question in scripts.
• Two kinds of keys. A Read only key can see everything. A Read & write key is needed to create, change or delete, and it expires (after 30 days by default).
• IDs and names. Lists show the NAME you chose and an ID. Commands take the ID.
• Scripting. -o json prints the full response and -o name prints only IDs, one per line.
• Retries are safe. Rate limits and network errors are retried automatically. Every create or change carries an idempotency key, so a retried request is never applied twice.

Configuration

The API key is looked up in this order: the --api-key option (or Meshive(api_key=...)), the MESHIVE_API_KEY environment variable, then the file saved by meshive login.

meshive logout deletes the saved key.

Troubleshooting

Exit codes

Versioning

This package follows semantic versioning. Within 1.x, upgrades don't break your code: new features arrive in minor versions and fixes in patch versions. Anything slated for removal first warns with a DeprecationWarning and is removed only in the next major version. Every change is listed in the changelog.

Use Meshive from an AI agent

The Meshive MCP server lets Claude Code, Codex, Cursor and other agents do all of this in plain language.

Learn more

• CLI reference: every command and option
• SDK reference: methods, data models, errors, retries
• Authentication: keys, environment variables, config location
• Changelog
• Contributing: DEVELOPMENT.md

License

Apache License 2.0