Python Executor

Execute Python code with API interception

Overview

The PythonExecutorProxy intercepts Python HTTP libraries (requests, urllib) and routes API calls to isolated environments.

Configuration

The executor automatically loads credentials from environment variables:

export AGENT_DIFF_API_KEY="ad_live_sk_..."
export AGENT_DIFF_BASE_URL="https://api.agentdiff.dev"

Basic Usage

from agent_diff import PythonExecutorProxy

# Create executor - auto-loads base_url and api_key from env vars
executor = PythonExecutorProxy(environment_id="env-123")

# Execute code
result = executor.execute("""
import requests

response = requests.post('https://slack.com/api/chat.postMessage', json={
    'channel': 'C01GENERAL99',
    'text': 'Hello from Python!'
})
print(response.json())
""")

print(result["stdout"])  # API response
print(result["status"])  # "success" or "error"

Creating Framework Tools

OpenAI Agents SDK

from agent_diff import PythonExecutorProxy, create_openai_tool
from agents import Agent

# Uses env vars automatically
executor = PythonExecutorProxy(env.environmentId)
tool = create_openai_tool(executor)

agent = Agent(
    name="Assistant",
    tools=[tool],
    instructions="Use execute_python to interact with Slack API..."
)

LangChain

from agent_diff import PythonExecutorProxy, create_langchain_tool

executor = PythonExecutorProxy(env.environmentId)
tool = create_langchain_tool(executor)

# Use with LangChain agent

smolagents

from agent_diff import PythonExecutorProxy, create_smolagents_tool

executor = PythonExecutorProxy(env.environmentId)
tool = create_smolagents_tool(executor)

# Use with smolagents CodeAgent

Supported Libraries

The Python executor intercepts:

• requests: All HTTP methods
• urllib.request: urlopen, Request
• httpx: (Coming soon)
• aiohttp: (Coming soon)

Execution Result

result = executor.execute(code)

# Result structure:
{
    "status": "success" | "error",
    "stdout": "...",     # Captured print() output
    "stderr": "...",     # Error output if any
    "exit_code": 0       # Process exit code
}

Error Handling

result = executor.execute(code)

if result["status"] == "error":
    print(f"Execution failed: {result['stderr']}")
else:
    print(f"Output: {result['stdout']}")

Example: Slack API Calls

result = executor.execute("""
import requests
import json

# List channels
channels = requests.get('https://slack.com/api/conversations.list').json()
print("Channels:", [c['name'] for c in channels['channels']])

# Post message
response = requests.post('https://slack.com/api/chat.postMessage', json={
    'channel': 'C01GENERAL99',
    'text': 'Hello!'
}).json()

print("Message sent:", response['ok'])
print("Timestamp:", response['ts'])
""")

Example: Linear GraphQL

result = executor.execute("""
import requests

query = '''
mutation {
  issueCreate(input: {
    title: "Fix bug"
    teamId: "team-123"
  }) {
    success
    issue {
      id
      title
    }
  }
}
'''

response = requests.post('https://api.linear.app/graphql', json={
    'query': query
})

print(response.json())
""")

Configuration Options

# Option 1: Environment variables (recommended)
# export AGENT_DIFF_API_KEY="ad_live_sk_..."
# export AGENT_DIFF_BASE_URL="https://api.agentdiff.dev"
executor = PythonExecutorProxy(environment_id="env-123")

# Option 2: Explicit configuration
executor = PythonExecutorProxy(
    environment_id="env-123",
    base_url="https://api.agentdiff.dev",
    api_key="ad_live_sk_..."  # Passed as Authorization header
)

Next Steps

TypeScript Executor

Execute TypeScript/JavaScript code

Bash Executor

Execute shell commands