Skip to main content
Version: 2.31

Monty

haystack_integrations.tools.monty.python_tool​

MontyPythonTool​

Bases: Tool

A Haystack Tool that lets an Agent run Python code in a Monty sandbox.

Monty is a minimal Python interpreter written in Rust. The tool keeps a pool of Monty worker processes and runs every call in a fresh interpreter, so nothing leaks between calls, users, or concurrent tool invocations. The tool returns what the code printed, the repr() of its last expression, and any error as a traceback, so the LLM can read the result and fix its code.

Security model​

Monty is a language-level sandbox: its interpreter implements no operation that reaches the host, so code has no access to files, the network, environment variables, or subprocesses. It runs in worker subprocesses started with an empty environment, so a crash never takes down the host process. The tool mounts no directories and exposes no host functions to the sandbox. Execution time and heap memory are capped by resource_limits.

Usage example​

python
from haystack.components.agents import Agent
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.dataclasses import ChatMessage
from haystack_integrations.tools.monty import MontyPythonTool

# Requires the OPENAI_API_KEY environment variable
agent = Agent(chat_generator=OpenAIChatGenerator(), tools=[MontyPythonTool()])
result = agent.run(messages=[ChatMessage.from_user("What is the sum of the first 100 prime numbers?")])
print(result["last_message"].text)
# >> The sum of the first 100 prime numbers is 24133.

init​

python
__init__(
    *,
    name: str = "run_python",
    description: str | None = None,
    resource_limits: ResourceLimits | None = None,
    type_check: bool = False,
    max_output_chars: int = 20000
) -> None

Create a MontyPythonTool.

Parameters:

  • name (str) – Tool name exposed to the LLM.
  • description (str | None) – Tool description exposed to the LLM. If None, a description of the sandbox and the supported Python subset is used.
  • resource_limits (ResourceLimits | None) – Monty resource limits for each call, merged over the defaults of 30 seconds of execution time (max_feed_duration_secs) and 256 MiB of heap memory (max_memory). Set a key to None to disable that limit. See pydantic_monty.ResourceLimits for the available keys.
  • type_check (bool) – If True, type-check the code with Monty's bundled type checker before running it, and return type errors to the LLM instead of executing the code.
  • max_output_chars (int) – Maximum number of characters kept from each of the printed output, the result, and the error before they are returned to the LLM. The output and the result keep their beginning, the error keeps its end, where the exception is.

Raises:

  • ValueError – If resource_limits contains a key that Monty doesn't support, or max_output_chars is less than 1.

warm_up​

python
warm_up() -> None

Start the pool of Monty worker processes. Called by Agent.warm_up(); safe to call more than once.

close​

python
close() -> None

Shut down the pool of Monty worker processes.

Safe to call more than once. The tool starts a new pool if it is invoked again.

to_dict​

python
to_dict() -> dict[str, Any]

Serialize the tool to a dictionary.

Returns:

  • dict[str, Any] – Dictionary with serialized data.

from_dict​

python
from_dict(data: dict[str, Any]) -> MontyPythonTool

Deserialize the tool from a dictionary.

Parameters:

  • data (dict[str, Any]) – Dictionary to deserialize from.

Returns:

  • MontyPythonTool – Deserialized tool.