Code Execution Tool

ChimerAI's Code Execution Tool runs Python code in an isolated sandbox and returns the output — ideal for data analysis, calculations, and dynamic code generation workflows.

What you get

Python execution — Run arbitrary Python in a sandboxed environment
Isolated per-request — Each execution uses a fresh subprocess
Timeout protection — Configurable execution time limit (default 30s)
Output capture — stdout, stderr, return value
Package access — NumPy, pandas, matplotlib, statistics, math pre-installed
Plot output — matplotlib figures returned as base64 PNG
Security — No file system access, no network, no subprocess.run

Quick setup

npx chimerai add ai-tools --only code

Scaffolds:

app/api/tools/code/execute/route.ts  ← Execution endpoint
services/ai/tools/code_tools.py      ← Python sandbox runner

Usage

const res = await fetch('/api/tools/code/execute', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    code: `
import statistics
data = [10, 20, 30, 15, 25]
print("Mean:", statistics.mean(data))
print("Stdev:", round(statistics.stdev(data), 2))
    `.trim(),
    timeout: 15,
  }),
});

const { output, error, executionTime } = await res.json();
// output: "Mean: 20\nStdev: 7.91"

Code + AI workflow (code generation)

// 1. Ask AI to generate code
const generated = await openai.chat.completions.create({
  model: 'gpt-4o',
  messages: [
    { role: 'system', content: 'Write Python code. Return only code, no explanations.' },
    { role: 'user', content: 'Calculate the first 10 Fibonacci numbers' },
  ],
});

const code = generated.choices[0].message.content ?? '';

// 2. Execute the generated code
const result = await fetch('/api/tools/code/execute', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ code }),
});

const { output } = await result.json();

Python sandbox implementation

import subprocess, sys, json, textwrap

def execute_code(code: str, timeout: int = 30) -> dict:
    # Wrap code to capture output
    wrapped = textwrap.dedent(f"""
import sys, io
_stdout = io.StringIO()
sys.stdout = _stdout
try:
{textwrap.indent(code, '    ')}
except Exception as e:
    print(f"Error: {{e}}", file=sys.stderr)
finally:
    sys.stdout = sys.__stdout__
    print(_stdout.getvalue(), end='')
""")
    result = subprocess.run(
        [sys.executable, "-c", wrapped],
        capture_output=True, text=True, timeout=timeout,
        # Security: no inherit env, no shell
        env={"PATH": "/usr/bin:/bin"},
    )
    return {
        "output": result.stdout,
        "error": result.stderr or None,
        "returnCode": result.returncode,
    }

Code Execution Tool

What you get

Quick setup

Usage

Code + AI workflow (code generation)

Python sandbox implementation

Further reading