qbox-sh 1.4.0

Python SDK for qbox.sh — self-hosted Firecracker code sandboxes

qbox — Python SDK

The official Python SDK for qbox.sh, the self-hosted Firecracker code-sandbox platform. Mirrors the e2b Sandbox shape so migration is mostly a one-line import change.

pip install qbox-sh

import qbox

with qbox.Sandbox.create(template="python-ml") as sandbox:
    result = sandbox.run_code("import pandas as pd; pd.DataFrame({'a':[1,2,3]}).describe()")
    print(result.text)

Authentication

The SDK reads QBOX_API_KEY and QBOX_BASE_URL from the environment, or you can pass them explicitly / via qbox.configure(...):

export QBOX_API_KEY=qbox_...
export QBOX_BASE_URL=https://qbox.your-company.internal   # self-hosted

qbox.configure(api_key="qbox_...", base_url="https://qbox.acme.internal")

What's included

• Sandbox — create / connect / list / get_info / set_timeout / kill, context manager
• sandbox.commands — run (foreground + background=True), run_argv, list, connect, kill; CommandHandle streams output
• sandbox.files — read / write / append / list / stat / exists / remove / rename / mkdir / download / upload
• sandbox.run_code / run_code_stream / code_contexts — the code interpreter (rich outputs: text/html/png/json)
• sandbox.shell() — interactive PTY (install qbox[shell])
• qbox.Templates — read-only list / get
• Typed errors for every path, env/global/explicit config, transient-only retry, py.typed

import qbox

with qbox.Sandbox.create(template="python-ml") as sandbox:
    sandbox.files.write("/tmp/data.csv", "a,b\n1,2\n")
    out = sandbox.run_code("import pandas as pd; pd.read_csv('/tmp/data.csv').sum().to_dict()")
    print(out.text)
    proc = sandbox.commands.run("python --version")
    print(proc.stdout)

Async (qbox.AsyncSandbox) is planned for v1.1.

CLI

pip install qbox-sh also installs the qbox command (operators may prefer pipx install qbox-sh). It wraps the SDK and adds multi-profile config, rich output, and operator commands.

qbox auth login                         # set base URL + API key (saved to ~/.config/qbox/config.yaml)
qbox doctor                             # end-to-end deployment check
qbox sandboxes create python-ml         # → sb_018f… (prints the id; scriptable)
qbox sandboxes run-code sb_018f -e "print(6*7)"
qbox sandboxes exec sb_018f -- pytest -q
qbox sandboxes files cp ./script.py sb_018f:/tmp/script.py
qbox sandboxes ssh sb_018f              # interactive PTY (needs qbox[shell])
qbox sandboxes kill sb_018f -y

Operator surface: qbox hosts list|drain|cordon|uncordon, qbox api-keys create|list|revoke|rotate, qbox audit logs, qbox templates create|list|delete.

Output: human tables/prose in a TTY, JSON when piped; force with -o json|yaml|table|prose. Multi-profile via --profile/QBOX_PROFILE. Exit codes are documented (0 ok, 2 usage, 3 auth, 4 not-found, 5 transient, 6 conflict, 10 declined, 124 timeout).

Development

pip install -e ".[dev]"
make test        # unit tests
make typecheck   # mypy --strict
make lint        # ruff