AgentCube Python SDK

The official Python SDK for AgentCube, enabling programmatic interaction with secure, isolated Code Interpreter environments.

This SDK creates a seamless bridge between your application and the AgentCube runtime, handling the complexity of:

• Session Management: Automatically creating and destroying isolated environments.
• Execution: Running shell commands and code (Python, Bash) remotely.
• File Management: Uploading and downloading files to/from the sandbox.

Features

• Simple API: Pythonic context managers (with statement) for automatic resource cleanup.
• Flexible: Supports both short-lived (ephemeral) and long-running sessions.
• Kubernetes Native: Automatically authenticates using Service Account tokens when running in-cluster.

Installation

git clone https://github.com/volcano-sh/agentcube.git
cd agentcube/sdk-python
pip install .

Quick Start

from agentcube import CodeInterpreterClient

with CodeInterpreterClient() as client:
    output = client.run_code("python", "print('Hello, AgentCube!')")
    print(output)
# Session automatically deleted on exit

Manual Lifecycle Management

For long-running applications, you can manually control the lifecycle:

# Create a session with a 1-hour timeout
client = CodeInterpreterClient(ttl=3600)

try:
    client.run_code("python", "print('Session started')")
    # ... perform operations ...
    # File system state persists within session
    client.write_file("42", "value.txt")
    client.run_code("python", "print(open('value.txt').read())")
finally:
    client.stop()  # CRITICAL: Ensure resources are released

File Operations

with CodeInterpreterClient() as client:
    client.upload_file("./data.csv", "/workspace/data.csv")
    client.run_code("python", """
import pandas as pd
df = pd.read_csv('/workspace/data.csv')
df.describe().to_csv('/workspace/summary.csv')
""")
    client.download_file("/workspace/summary.csv", "./summary.csv")

API Reference

Configuration

CodeInterpreterClient(
    name="custom-template",        # CodeInterpreter CRD template name
    namespace="agentcube",         # Kubernetes namespace
    ttl=7200,                      # Session TTL (seconds)
    session_id="existing-id",      # Optional: reuse existing session
    verbose=True                   # Enable debug logging
)

Environment Variables:

• WORKLOAD_MANAGER_URL: Control Plane URL
• ROUTER_URL: Data Plane URL

Advanced: Session Reuse

For workflows requiring file system state persistence across multiple client instances:

Note: Each run_code call spawns a new process. Python variables do NOT persist. Only file system state is preserved.

# Step 1: Create session and save state to file
client1 = CodeInterpreterClient()
session_id = client1.session_id  # Save for reuse
client1.write_file("42", "value.txt")
# Don't call stop() - let session persist

# Step 2: Reuse session with new client
client2 = CodeInterpreterClient(session_id=session_id)
client2.run_code("python", "print(open('value.txt').read())")  # File persists
client2.stop()  # Cleanup when done

Development

make build-python-sdk
python3 -m pytest sdk-python/tests/