Pydantic Creator Open-Sources Monty, a Python Sandbox with 1ms Startup

Samuel Colvin, creator of Pydantic, open-sourced Monty v1 on September 25, a Python sandbox designed for AI agent code execution. Its headline feature is extremely low startup overhead: 1 millisecond per launch versus roughly 1.5 seconds for typical cloud sandboxes, so agents are no longer bogged down when frequently writing and running short scripts.

Confirmed

• Monty v1 was open-sourced by Samuel Colvin on 09-25
• Startup time is just 1 ms, compared with 1.5 s for cloud sandboxes
• The author's benchmark ran 10,000 sandboxed scripts in 674 ms total, while a cloud sandbox would take over 3 hours for the same workload

Why it matters

• Agent workflows involve many short-lived scripts, and the 1.5-second per-run overhead of traditional sandboxes becomes a major bottleneck at scale
• By compressing startup overhead to the millisecond level, Monty makes high-frequency, concurrent sandboxed code execution locally feasible, offering new infrastructure for the agent toolchain

2026-09-25 ~ 2026-09-26 · 5 related posts

• [source] Pydantic creator open-sources Monty: a 1ms-startup Python sandbox that ran 10k scripts in 674ms — samuelcolvin · 2026-09-25

4 near-duplicate retellings: samuelcolvin · samuelcolvin · samuelcolvin · samuelcolvin