Introduction

A decorator on your Python function names the packages, CPU, memory, GPU, and storage it needs, so whoever writes the function, you or an agent, also writes where it runs. LazyCloud builds the container, runs it on cloud CPUs and GPUs, and stops it when the work is done. You pay per second while it runs. A coding agent can do all of this from your machine or from a devbox.

from lazycloud import App, Image
from pydantic import BaseModel

app = App("reports")


class Summary(BaseModel):
    total: float
    mean: float


@app.function(image=Image(python_version="3.12"), cpu=2, memory="4Gi")
def summarize(amounts: list[float]) -> Summary:
    return Summary(total=sum(amounts), mean=sum(amounts) / len(amounts))

Develop locally, test in the cloud, deploy

The same file runs at every stage:

uv run pytest                                           # tests call summarize.local(...)
uv run lazycloud run reports:summarize '[12.5, 30, 8]'   # run once in the cloud
uv run lazycloud deploy reports                         # keep it running in the cloud

No step opens a browser or prompts. With --json the result is one JSON document on stdout, logs go to stderr, and a failure exits nonzero, so an agent can run the loop and read the outcome.

What you can ship

APIs: a function as an HTTPS endpoint, or a full FastAPI app.
Jobs: tests, evals, data processing, and GPU work fanned out across containers.
Background work and crons: tasks with retries, cancellation, and live logs.
Agent sandboxes: isolated containers for running your agent’s code and tests.
Devboxes: SSH machines whose disk keeps your files.

Next steps

Quickstart: develop, test, and deploy a first workload.
Apps and workloads: the workload kinds and how an app groups them.
Examples: GPU inference, training, document processing, and a sandboxed coding agent.

Getting started

Core concepts

CLI reference

Examples

Account

Develop locally, test in the cloud, deploy

What you can ship

Next steps

​Develop locally, test in the cloud, deploy

​What you can ship

​Next steps

Develop locally, test in the cloud, deploy

What you can ship

Next steps