workpaper

Cloud computers for AI agents, from Python. Each computer is a real Linux machine with its own kernel, ready in under a second. Stop it and it keeps its disk and memory; fork it while it runs. No dependencies.

pip install workpaper

Make an API key once with the CLI (curl -fsSL https://workpaper.co/install | sh, then workpaper auth login):

workpaper keys create my-app      # prints wpk_…; set it as WORKPAPER_API_KEY

Quickstart

from workpaper import Workpaper

wp = Workpaper()  # reads WORKPAPER_API_KEY

c = wp.run("my-agent")  # create it, or wake it if it exists
r = c.exec("python3 --version")
print(r.exit_code, r.stdout)

c.write("/home/user/app/main.py", "print('hi')")
c.exec("python3 main.py", cwd="/home/user/app", on_stdout=print)

c.stop()   # disk and memory saved, billing stops
c.start()  # running programs carry on from the same moment

Fork a running computer

base = wp.run("base")
base.exec("git clone https://github.com/you/app && cd app && pip install -r requirements.txt")

# Each copy starts with the files, memory and running programs of `base`.
tries = [base.fork(n) for n in ("try-a", "try-b")]

API

Errors are WorkpaperError with status and, when the service is full, reason == "at_capacity".

Every computer comes with Node 24, Python 3.12, uv, git, Claude Code, Codex and a headless browser. Limits and pricing: docs.workpaper.co.