Hi HN! I've never posted anything on here before, but this project was too fun to pass up. I've been learning a lot about agents lately, and wanted to see if I could replicate some of the results of Windsurf and Cursor. Generally I love these tools, but when you're stuck behind an "enterprise" firewall, they don't work as well. This sad state of affairs motivated me to understand SWE agents more.

Does it work? Yes. Is it as good as Cursor and Windsurf? Not quite, but it's pretty darn close, depending on the model you use. I'm lacking a few features like semantic indexing, MCP support, but I'll add them soon.

Is it redundant with codex? Yeah, probably, but it was a super fun to build, and since I wrote it I have full control over any aspect of it to adapt to weird silly corporate environment network goofiness. Also, Codex doesn't give you a puppy emoji nor does it have "Always obey the Zen of Python, even when you're not writing Python." hardcoded into the prompt.

Cute / nifty fun fact:

At a certain point while I was building code_puppy, I swapped from Windsurf to using code_puppy to build code_puppy. The agent realized that it was working on itself, and made note of it while invoking the "thinking" tool.

  The puppy is unleashed!!!!