quixand 0.1.5

Local-first, adapter-pluggable sandboxes for AI code execution (E2B-like API)

quixand — Local & Pluggable Sandboxes for AI Code Execution

Quixand is a local-first sandbox and code interpreter library. It mirrors the developer-facing API of E2B’s Sandbox and AsyncSandbox while running locally by default via Docker/Podman, with a clean adapter interface to plug in other backends (your infra, remote HTTP, etc.).

Why quixand?

• Familiar: mirrors E2B’s concepts and method names where practical
• Local-first: default execution in local containers (Docker/Podman)
• Pluggable: unified Adapter protocol for custom backends
• Async & streaming ready; simple, secure, fast
• CLI parity for everyday workflows

Requirements

• Python 3.10+
• Docker or Podman installed and in PATH

Install

Editable install with Docker extras:

python3 -m venv .venv
. .venv/bin/activate
pip install -U pip
pip install -e '.[docker]'

Configuration

Quixand works out of the box, but you can configure via environment variables or code.

Environment variables:

• QS_ADAPTER: local-docker (default) | chutes | remote-http
• QS_TIMEOUT_DEFAULT: default timeout seconds (default 300)
• QS_IMAGE: default container image (default python:3.11-slim)
• QS_RUNTIME: docker | podman (auto-detects if unset)
• QS_ROOT: host directory for state/template cache (default ~/.quixand)
• QS_METADATA: JSON string to tag sandboxes

Programmatic config:

import quixand as qs
cfg = qs.Config(timeout=600, image="python:3.11-slim")

Python API

Top-level imports:

from quixand import Sandbox, AsyncSandbox, connect, Templates

Sync Sandbox:

import quixand as qs

sbx = qs.Sandbox(template="python:3.11-slim", timeout=600, metadata={"user":"alice"})
sbx.files.write("hello.txt", "hi!")
print([f.path for f in sbx.files.ls(".")])
res = sbx.run(["python","-c","print(2+2)"])
print(res.text)  # "4\n"
execn = sbx.run_code("x=1\nx+=1\nprint(x)")
print(execn.text.strip())  # "2"
sbx.shutdown()

Async Sandbox:

import asyncio, quixand as qs

async def main():
    sbx = await qs.AsyncSandbox.create(template="python:3.11-slim")
    await sbx.files.put("data.csv", "/workspace/data.csv")
    res = await sbx.run(["python","-c","print(42)"])
    print(res.text)
    await sbx.shutdown()

asyncio.run(main())

Connect to running sandbox:

sbx1 = qs.Sandbox(template="python:3.11-slim")
sid = sbx1.id

sbx2 = qs.connect(sid)
print(sbx2.status())

Filesystem API:

sbx.files.write("/workspace/notes.txt", "hello")         # text
sbx.files.write("/workspace/blob.bin", b"\x00\x01", mode="binary")
print(sbx.files.read("/workspace/notes.txt"))             # returns str
print(sbx.files.read("/workspace/blob.bin", mode="binary"))  # returns bytes
sbx.files.mkdir("/workspace/data", parents=True)
sbx.files.put("./local.txt", "/workspace/local.txt")
sbx.files.get("/workspace/local.txt", "./local_copy.txt")
sbx.files.rm("/workspace/local.txt")
paths = sbx.files.glob("/workspace/*.txt")

Process execution:

res = sbx.run("echo $USER && uname -a")
print(res.exit_code, res.text)

Python code convenience:

sbx.install_pkg("pandas==2.2.2")
execn = sbx.run_code("import pandas as pd; print(pd.__version__)")
print(execn.ok, execn.text)

Networking and ports (adapter-dependent):

binding = sbx.expose(port=8000, host_port=18000, proto="tcp")
print(binding)

Timeouts and lifecycle:

st = sbx.status()                     # created_at, last_active_at, timeout_at
sbx.refresh_timeout(900)              # extend timeout
sbx.shutdown()                        # stop container and clean state

CLI

The qs CLI mirrors E2B’s verbs.

qs --help

# Sandbox lifecycle
qs sandbox create --template python:3.11-slim --timeout 900 --env FOO=bar --env BAZ=qux
qs sandbox ls
qs sandbox connect <id>
qs sandbox exec <id> -- echo hello
qs sandbox run-code <id> --code "x=1; print(x+1)"
qs sandbox refresh-timeout <id> --seconds 600
qs sandbox kill <id>

# Files
qs files ls <id> /workspace
qs files put <id> ./local.txt /workspace/local.txt
qs files get <id> /workspace/local.txt ./local_copy.txt
qs files mkdir <id> /workspace/data --parents
qs files rm <id> /workspace/local.txt --recursive

# Templates
qs templates build ./path --name py311-tools
qs templates ls
qs templates rm py311-tools

Templates

Quixand supports building images from e2b.Dockerfile or Dockerfile and caching references locally under ~/.quixand/templates.

qs templates build ./examples --name py311-tools
qs templates ls

In code:

from quixand import Templates, Sandbox
img = Templates.build("./examples", name="py311-tools")
sbx = Sandbox(template=img)

Adapters

Adapters implement a common protocol in quixand.adapters.base.Adapter. Included:

• LocalDockerAdapter (default) — local Docker/Podman containers
• ChutesAdapter (skeleton) — wire to your infra
• RemoteHTTPAdapter (skeleton) — call a remote service

Select adapter via QS_ADAPTER or passing an instance to Sandbox(adapter=...).

State and Timeout Enforcement

Quixand keeps a local registry at ~/.quixand/state.json. A lightweight watchdog process enforces idle timeouts and cleans up containers and state when deadlines are exceeded. Default timeout is 300s; refresh with refresh_timeout() or CLI.

Security Defaults

• Non-root user inside container (image-dependent)
• No privileged flags
• Bridge network by default (configurable); none supported
• CPU/memory/pids limits supported by the adapter
• Ephemeral writable layer

Troubleshooting

• Docker/Podman not found: ensure one is installed and in PATH
• Permission errors: add your user to the docker group (Linux) or run Docker Desktop
• Hanging exec or timeouts: increase timeout in run() or QS_TIMEOUT_DEFAULT
• Unicode output: CommandResult.text decodes UTF-8 with replacement; use stdout bytes for raw data

Examples

See examples/:

• minimal.py
• async_minimal.py
• template_build.py
• streaming.py (PTY placeholder; streaming to be implemented)

Roadmap

• Full PTY streaming
• Richer RemoteHTTP/Chutes adapters
• Test suite and CI

License

MIT