Execute Python Scripts and Capture Output

The Code Interpreter Tool lets a LlamaIndex agent run Python scripts and capture the resulting stdout and stderr. CodeInterpreterToolSpec exposes a single function, code_interpreter, which evaluates a Python script the agent writes and returns its output.

The source carries an explicit warning: this tool gives the agent access to the subprocess.run command, meaning arbitrary code execution is possible on the machine running it. The source states plainly that it is not recommended for production use, and would require heavy sandboxing or virtual machines to be used safely.

Use it only in development, testing, or fully isolated environments where you understand and accept that the agent can execute arbitrary code on the host machine - for example prototyping an agent that writes and runs small calculations or scripts. Do not use it in production, or on any machine with access to sensitive data or systems, without first adding real sandboxing or running it inside a disposable virtual machine; the tool itself provides no isolation.

code_interpreter evaluates a Python script the agent supplies and returns its stdout and stderr output, with no built-in execution isolation.

from llama_index.tools.code_interpreter import CodeInterpreterToolSpec
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.openai import OpenAI

code_spec = CodeInterpreterToolSpec()

agent = FunctionAgent(
    tools=code_spec.to_tool_list(), llm=OpenAI(model="gpt-4.1")
)

Developers prototyping agents that need to write and run Python code, in development or fully sandboxed environments only - not for production or unisolated use given the arbitrary code execution risk.