I’ve been using a lot of Gitea.
Or perhaps I’ve been rediscovering the repos I’ve starred, readme-patched (I’m SORRY!) and probably otherwise drive-by-commited to over the years.
Woah. Things have changed a bit.
Entire whatever model and harness you use these days and point them at a codebase, they can generally do a pretty good job of explaining the ins and outs of code, and it feels pretty good to be able to think about code and then check your assumptions immediately.
The tea command comes into play when you want to quickly file an issue to come back to – instead of an agent memory, a project to-do. Sometimes… often over-zealous but if told to leave a bullet list of conclusions about the codebase, that’s something you can use elsewhere.
You don’t need something the size of a microwave, and it doesn’t need to be an M4 Mac mini, though they are certainly handy.
I’ve been using Raspberry Pis, the 4 and up, specifically (the 3b+ has a use! a topic for a later NetBSD post).
Tiny little nodes that can run a harness, Ollama, and point to the cloud, giving me evals, experiments, traces and spans that let me then test the code – what a concept! – and deploy it to a tiny fake “gas town” fleet to run on.
Of course the first thing I reached for was IRC. This is Colloquy on a Mac, asking one of the agents about temps. You don’t need a cl4w for this, and though I’m not recommending vibe coding anything to do with irssi in public, for an offline ergo server and a few clients, it’s pretty low-overhead.