Hey HN!
I've been working on a project called Darwin that I'm thrilled to share with you.<p>Darwin is essentially your GitHub agent powered by large language models (LLMs). It checks out your projects, understands them through natural language prompts, and automates tasks such as fixing issues, documenting code, reviewing pull requests, and more.<p>What drove me to create Darwin was a desire to harness the power of LLMs in a way that's seamlessly integrated with the tools I use daily. The motivation came from my curiosity about what could be possible when writing code that understands code.
Darwin stands out because it's designed for developers who want to leverage AI without needing deep expertise in LLMs or prompt engineering. It offers:<p>- hands-off approach to automate routine development tasks.<p>- Novel and creative ways of making LLMs work for you<p>- A unique API for each project, allowing for customized automation tools.<p>Currently, Darwin is in alpha. It's functional, with users able to connect their repositories, define tools, and run tasks. I'm especially interested in feedback at this stage — everything from output quality to user experience. Every project starts with a $5 free budget to try it out, and while payment isn't implemented yet, I'm keen to hear your thoughts.<p>The vision for Darwin is not just about automation but creating a more productive, creative, and enjoyable development experience. I believe we're just scratching the surface of what's possible with AI in software development, and I'm excited to see where we can take this.<p>For those interested, I'm looking for alpha testers and feedback. If you're curious about automating your GitHub workflow or want to push the limits of what AI can do for development, Darwin might be for you.
Check it out and let me know what you think!
Do you have an open and public PR that shows off Darwin's PR review capabilities? I'm reading a lot of words about what it can do, but I'd much rather see an actual example.
It asks for access to my private repos. I don't want to give a brand-new AI tool access to my private repos. But if it looking at my private repos was optional, I would try it.
I like it! I built a command line AI coding assistant for my projects (<a href="https://codemancer.codes" rel="nofollow">https://codemancer.codes</a>) when GPT-4 first came out, but still think this space is wayyy under explored given the possible capabilities.
As someone who’s also solo-employed on AI development, I love your work based on the copy - will try when I get home. Please, tho, do some thinking about safety if you haven’t. There’s a lot that can go wrong between “nothing” and “paperclip apocalypse”, and I think the autonomy you’re giving machines here should only be done with great care.<p><a href="https://www.lesswrong.com/posts/kpPnReyBC54KESiSn/optimality-is-the-tiger-and-agents-are-its-teeth" rel="nofollow">https://www.lesswrong.com/posts/kpPnReyBC54KESiSn/optimality...</a>
Let's say I want to have a somewhat genereric CRUD-app: A database, some application logic, a login system, logging/telemetry and a UI.<p>What kind of stack should I use so that tools like Darwin (or just plain old copilot) can be most effective?
I would be more willing to try this if it was possible to host it myself. I get that you want to monetize this and that’s fair, but I am also not going to trust my code to a random online tool.<p>Maybe I’ll give a bash at making something like this for my own use
UPDATE<p>First of all - big thank you to everyone who gave feedback and tried out the project!<p>I hear your concerns about trust and invasive permissions. Here is what I will do tonight:<p>- Remove all data of everyone who has signed up so far<p>- Revoke all granted privileges<p>- Deactivate signup<p>- Put a detailed note on the front page for anyone who's already signed up<p>Your feedback has been invaluable and I will start working on the next iteration soon.
I think this is a cool idea. I'm curious why it could't work like dependabot does though. Dependabot can scan my public repos and even create PRs from its own ephemeral forks without the overstepping of permissions that this needs.
Is it self-hostable / usable with your own Github App?<p>There's no way I'm granting a third party app access to all my public and private repos.
Might be getting a big hug from our friends here.<p>> Error: Server Error<p>> The server encountered a temporary error and could not complete your request.<p>> Please try again in 30 seconds.
Things like this are clearly the future. Having seen how some open source projects are run, and how we delegate some tasks to juniors, LLMs can play a major role. Imagine not having have to spell out a response to a bug report or a feature request, but the LLM gives you three options and you just pick one. Super simple and that alone would be an insane boost for productivity.<p>For Darwin, one improvement that comes to mind is to make a catalog of blueprints with common actions, so users have less to configure.<p>Good luck with the product!