modal-labs/devlooper
A program synthesis agent that autonomously writes and fixes code by running tests in a sandbox environment.

Devlooper is an autonomous coding agent that generates code and iteratively fixes it based on test results. It extends smol developer with a Modal sandbox where it runs tests, captures errors, and uses an LLM (GPT-4) to diagnose failures and generate a debug plan. The debug plan can patch files, install packages, or run commands, and this loop repeats until all tests pass. It supports React/Jest, Python, and Rust templates but can be extended to any language with a test harness.