aricode started as a side project by one developer. It still is.
I'm Alex. I'd been using various AI coding services — tried a few of them, liked parts of each — but I kept wanting to just build my own thing. Not because the other tools were bad, but because it sounded fun.
I was already using Ollama's cloud API for other stuff and noticed how little usage it actually consumed compared to what I was paying elsewhere. So I started tinkering. Could I wire up a coding agent that talks to any backend? Turns out, yeah.
It was supposed to be a weekend project. Then it got a knowledge graph. Then dreaming. Then a safety system. Then an SDK. At some point it stopped being a toy and started being something other people might actually want to use.
"This whole thing was vibecoded. I didn't hand-write most of the code — I directed AI to build it. Which is kind of the best demo I could ask for."
— AlexHonestly? Because it's more fun when other people use it too. And because I think developers should have options. You shouldn't need a specific subscription to get a good AI coding experience.
aricode works with whatever backend you want — Ollama cloud, a local model, OpenAI, Anthropic, DeepSeek, self-hosted endpoints. The agent doesn't care where the model lives, and that flexibility is the whole point.
The full source code will be released at v1 once the foundations are solid — MIT licensed, no strings attached. For now, the binary is free to use and there's no business model behind it. Just a project I enjoy working on.
Any model, any backend, any hardware. aricode is the agent, not the provider.
This started as a side project and still feels like one. That's a feature, not a bug.
Full source drops when the foundations are solid. MIT licensed, no strings.
aricode was built by directing AI. If the tool can build itself, that says something.