Hi HN, we recently built and open sourced Sweep, an AI powered junior developer. Kevin, Luke, and I were tired of handling tech debt when we knew LLMs could do this for us.<p>Sweep solves small features, bugs, and missing documentation by writing pull requests. This allows you to review and validate the code before pushing it to your codebase.<p>Features:<p>- Turns issue descriptions directly into pull requests (without an IDE)<p>- Addresses developer replies & comments<p>- Fixes the code based on failing tests & linters from GitHub Actions<p>Examples: <a href="https://docs.sweep.dev/examples">https://docs.sweep.dev/examples</a><p>Check out our Github repo and install Sweep at: <a href="https://github.com/sweepai/sweep#-getting-started">https://github.com/sweepai/sweep#-getting-started</a>!
Nice UX.<p>George Hotz said recently that he sees this exact idea as the future of how hw would do things (maybe email him!).<p>The async nature means a 1 hour auto-gpt style cycle is no issue for a human waiting on it.<p>However this so depends on how good a job GPT4 with the correct stage hypnotism can do at this task.<p>I think I like it for doco changes definitely. Just having it take on the PR prep hassle for that is worth it. The human mind is great at verifying that in review.<p>But code changes: LLMs make great looking code that subtly doesn’t work. I would only trust it if there are extensive tests. But I guess I can use this to make the tests.<p>I presume it is open source as well as hosted model here?