This example demonstrates how to use the Cloud Code SDK together with LangGraph to build an AI coding agent that can create, read, search, and edit files in a sandbox environment. The agent uses a ReAct-style loop and exposes tools for file management and code editing.

​
Example: sandbox.py

import os
import json
import sys
from typing import Any, Dict, List, Optional, Annotated, Sequence
from typing_extensions import TypedDict

from langchain.chat_models import init_chat_model
from langchain_core.messages import BaseMessage, SystemMessage, ToolMessage
from langchain_core.runnables import RunnableConfig
from langgraph.graph import StateGraph, END
from langgraph.graph.message import add_messages

# Import Cloud Code Sandbox SDK
from cloudcode import SandboxSDK

# Define the agent state
class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], add_messages]
    sandbox_id: Optional[str]
    working_dir: str
    sdk: Optional[Any]

DEFAULT_WORKING_DIR = "/home/user"
DEFAULT_SANDBOX_TIMEOUT = 300  # 5 minutes

def initialize_sandbox(state: AgentState) -> AgentState:
    sdk = SandboxSDK(
        model="gpt-4.1",
        api_key=os.getenv("CLOUD_CODE_API_KEY"),
        sandbox_timeout=DEFAULT_SANDBOX_TIMEOUT,
        user_id="react-agent-user"
    )
    sandbox_info = sdk.get_sandbox_info()
    state["sdk"] = sdk
    state["sandbox_id"] = sandbox_info["sandbox_id"]
    state["working_dir"] = sdk.working_dir
    sdk.create_file(
        "hello.py",
        """def greet(name):
    return f\"Hello, {name}!\"

if __name__ == "__main__":
    print(greet("World"))
"""
    )
    return state

def cleanup_sandbox(state: AgentState) -> Dict:
    if state.get("sdk"):
        return state["sdk"].kill_sandbox()
    return {"success": False, "message": "No sandbox to terminate"}

# Tool functions for the agent
# ... (see full code for details)

# Define the tools list and create a lookup dictionary by name
# ...

# Initialize the language model and bind tools
# ...

# Define the tool node for the react agent
# ...

# Define the node that calls the model
# ...

# Define the conditional edge that determines whether to continue or not
# ...

# Create the react agent graph
# ...

# Create the agent
# ...

if __name__ == "__main__":
    # Initialize state
    state = {
        "messages": [],
        "working_dir": DEFAULT_WORKING_DIR,
        "sandbox_id": None,
        "sdk": None
    }
    try:
        # Initialize the sandbox
        try:
            state = initialize_sandbox(state)
            print(f"🚀 LangGraph Sandbox App initialized with:")
            print(f"   - Sandbox ID: {state['sandbox_id']}")
            print(f"   - Working directory: {state['working_dir']}")
            print(f"   - Model: gpt-4.1")
        except Exception as e:
            print(f"❌ Failed to initialize sandbox: {str(e)}")
            print("Please check your CLOUD_CODE_API_KEY environment variable.")
            sys.exit(1)
        print("\nEnter your coding request (or 'quit' to exit):")
        while True:
            user_input = input("\n> ")
            if user_input.lower() in ("quit", "exit", "q"):
                break
            messages = [{"role": "user", "content": user_input}]
            print("\nProcessing your request...\n")
            # Stream the agent's response
            for step in agent.stream(
                {"messages": messages, **state},
                stream_mode=["messages"],
                config={"configurable": {"thread_id": 1}},
            ):
                if "messages" in step:
                    content = step["messages"][-1].content
                    if content:
                        print(content, end="")
    finally:
        # Clean up sandbox
        if state.get("sdk"):
            print("\nCleaning up sandbox...")
            result = cleanup_sandbox(state)
            print(f"Sandbox termination: {result['message']}")
            print(f"Termination success: {result['success']}")

Note:

  • This example demonstrates how to build a ReAct-style AI coding agent using LangGraph and the Cloud Code SDK in a sandbox environment.
  • You must set your CLOUD_CODE_API_KEY environment variable before running.
  • The agent exposes tools for file creation, reading, searching, code editing, and running shell commands.