openagentworld-sandbox 0.1.0

A lightweight, framework-agnostic safe code execution library for AI agents

openagentworld-sandbox

A lightweight, framework-agnostic safe code execution library for AI agents.

No lock-in. Swap backends in one line. Works with AutoGen, LangChain, or standalone.

---

The Problem

Every AI agent framework needs to execute code. But:

• AutoGen's executor only works with AutoGen
• LangChain has no safe execution built in
• E2B is cloud-only and paid
• exec() is dangerous with no isolation

openagentworld-sandbox is the missing abstraction layer.

---

Install

pip install openagentworld-sandbox

With Docker support:

pip install openagentworld-sandbox[docker]

With all backends:

pip install openagentworld-sandbox[all]

---

Quick Start

from openagentworld_sandbox import SafeExecutor

# Local subprocess (dev/testing)
executor = SafeExecutor(backend="local")
result = executor.run("print('hello from sandbox')")
print(result.output)   # hello from sandbox
print(result.exit_code)  # 0

# Docker (isolated, production-safe)
executor = SafeExecutor(backend="docker")
result = executor.run("print(2 ** 32)")
print(result.output)   # 4294967296

# E2B cloud
executor = SafeExecutor(backend="e2b")
result = executor.run("print('running in cloud')")

---

Security Profiles

from openagentworld_sandbox import SafeExecutor, SecurityProfile

# STRICT: blocks dangerous imports, no network, 128MB RAM cap, wipes on exit
executor = SafeExecutor(backend="docker", security=SecurityProfile.STRICT)

# DEFAULT: sane defaults, nothing blocked
executor = SafeExecutor(backend="local", security=SecurityProfile.DEFAULT)

# PERMISSIVE: minimal restrictions for trusted code
executor = SafeExecutor(backend="local", security=SecurityProfile.PERMISSIVE)

---

AutoGen Integration

from openagentworld_sandbox.integrations.autogen import AutoGenSandbox
from autogen_agentchat.agents import AssistantAgent

executor = AutoGenSandbox(backend="docker", security=SecurityProfile.STRICT)

agent = AssistantAgent(
    name="coder",
    code_executor=executor
)

---

LangChain Integration

from openagentworld_sandbox.integrations.langchain import LangChainSandbox
from langchain.agents import initialize_agent

tool = LangChainSandbox(backend="docker")

agent = initialize_agent(
    tools=[tool],
    llm=llm,
    agent="zero-shot-react-description"
)

---

ExecutionResult

Every .run() call returns:

@dataclass
class ExecutionResult:
    output: str         # stdout
    exit_code: int      # 0 = success
    is_timeout: bool    # True if killed by timeout
    error: str | None   # stderr if any
    backend_used: str   # "local" | "docker" | "e2b"

---

Backends

Backend	Isolation	Cost	Requires
local	Low (subprocess)	Free	Nothing
docker	High (container)	Free	Docker Desktop
e2b	High (cloud VM)	Paid tiers	E2B API key

---

Roadmap

• Firecracker backend
• Multi-language support (JS, bash)
• Session persistence across .run() calls
• CrewAI integration
• OpenManus integration

---

Contributing

Issues and PRs welcome at openagentworld/openagentworld-sandbox

---

License

MIT