Running Pydantic's Monty Rust sandboxed Python subset in WebAssembly

2026-02-06 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Pydantic has released Monty, a custom Python-like language subset implemented in Rust, designed for safely running LLM-generated code in a sandbox environment. Monty aims to reduce the overhead of full container-based sandboxes, offering startup times in microseconds and blocking host environment access (filesystem, environment variables, network) unless explicitly allowed via external function calls. The author successfully compiled Monty into WebAssembly, creating both a standalone `.wasm` module callable from JavaScript and a `pydantic_monty-0.0.3-cp313-cp313-emscripten_4_0_9_wasm32.whl` wheel file for use with Pyodide. This dual WebAssembly compilation enables Monty to run directly in a browser or within a Pyodide environment, demonstrating its portability and strict resource controls for memory, CPU, disk, and network access.

Key takeaway

For AI Engineers integrating LLM-generated code, consider Monty as a lightweight, high-performance sandboxing solution. Its WebAssembly compatibility allows for secure, client-side execution in browsers via JavaScript or Pyodide, significantly reducing latency and complexity compared to traditional container-based approaches. Evaluate Monty for scenarios requiring strict control over untrusted code's access to host resources and rapid execution.

Key insights

Monty provides a fast, sandboxed Python subset in Rust for safely executing LLM-generated code.

Principles

• Isolate untrusted code execution.
• Minimize sandbox overhead.
• LLMs can iterate on error messages.

Method

Compile a Rust-based Python subset (Monty) into WebAssembly, generating both a standalone WASM module for JavaScript and a Python wheel for Pyodide integration, enabling browser-based sandboxed execution.

In practice

• Use `uv run --with pydantic-monty python -m asyncio` to test Monty.
• Compile Rust to WASM for browser/Pyodide deployment.
• Leverage LLMs for iterative code generation against sandbox errors.

Topics

• Pydantic Monty
• WebAssembly
• Code Sandboxing
• LLM Agents
• Pyodide

Code references

• pydantic/monty
• astral-sh/uv
• simonw/research

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Why Rust is the Ideal Language for Vibe-Coding — Daniel Szoke, Sentry — Rust's strict compiler constraints, though challenging for LLMs initially, enhance production code safety by preventing common errors.
• Publishing WASM wheels to PyPI for use with Pyodide — PyPI now directly supports WebAssembly (WASM) wheels for Pyodide, simplifying package distribution and reducing maintainer burden.
• Building AI Agents in Rust — part 3 — Using traits and a registry simplifies AI agent tool dispatch and schema management in Rust.
• Running Python code in a sandbox with MicroPython and WASM — Combining MicroPython with WebAssembly provides a robust, embeddable sandbox for untrusted Python code execution.
• Implementing programmatic tool calling on Amazon Bedrock — PTC enables LLMs to orchestrate multi-step tool interactions via generated code in a sandbox, drastically cutting tokens and boosting accuracy.
• Every Python Concept a Generative AI Developer Actually Needs to Know — Mastering Python's concurrency, memory, and structural patterns is vital for building scalable and robust GenAI applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.