Show HN: Fork of Claude-code working with local and other LLM providers (opens in new tab)

> Isn't Claude-code a proprietary version of https://aider.chat/ in the first place?

It might be; but aider is Apache 2 licensed and I don't think Anthropic would be able to change the license. But either way, these tools are simple enough that it's not hard to replicate what they do. So simple that it's kind of pointless for Anthropic to get proprietary and protective about the whole thing. The moat they have is with the model quality.

I saw a very similar tool demoed at the local Kotlin meetup in Berlin just one day after Anthropic announced claude code that also uses the Claude API.

https://github.com/xemantic/claudine

Kotlin might not be everybody's cup of tea but actually seems to be an effective tool for the job. The level of simplicity of the code base is what I found the most shocking. There's almost nothing to it. It's tiny bits of code needed to read/write files and execute commands and a very simple loop where it interacts with claude via a simple REST API. You can read through the code base (a couple of hundred lines of Kotlin) in a few minutes and get a feel for what it does. We had some fun trying to get the tool to improve itself during the meetup. It worked pretty well.

Probably, the most complex thing of the whole code base is the system prompt used to orchestrate the whole thing. That looks like some thought went into it and there are probably a few gotchas that are yet to be addressed. But even that is doable. A determined person could probably replicate what this does in a day or so from scratch. Probably a lot faster if you use one of these tools to build the next one. This isn't rocket science.

I'm guessing we'll be seeing a lot of tools like this in the next few months. It will be interesting to see what refinements people come up with. I think most of the innovation is going to be in giving these agents better/smarter tools to work with. For example, there's a lot of stuff encapsulated in IDEs that it probably a lot easier to use than trying to do the same thing with random cli commands.

Claude Code works similarly to Aider in that they both run in the terminal and write code using LLMs, but it shares no code with Aider as far as I know. Aider is written in Python and Claude Code in Javascript, among other reasons to think that it is not derived from Aider.

The tools also work very differently when you're actually using them, with Claude Code doing more of an agent loop and having very different methods of putting together the context to pass to the model.

Similar concept but entirely different implementation.

(aider is Python, Claude Coder is TypeScript, for one)

If you want one that was shipped 3 months earlier, is true FOSS (Apache 2.0,), and can even work reasonably with qwen-32b-coder-instruct, check https://github.com/ai-christianson/RA.Aid

Disclaimer: am maintainer

If you try it and have feedback, I'm very curious to hear it.

Cross fingers :)

https://github.com/anthropics/claude-code/issues/249#issueco...

I'm holding off on using it until the IDFK license is OSI approved.

Yeah, hopefully they'd let it slide

Yeah, "Terminal-based AI coding retard" is unnecessary.

Thanks! Love to hear people are still using it, even though it needs a flag toggle now.

Are there examples of LLMs fixing spaghetti code in the wild?

The fundamental reason is that RAG kind of sucks and requires a ton of effort/optimization to reach a high degree of reliability for most applications. RAG solutions are not Anthropic's core product. Just reading all the relevant files is more expensive but is more effective and efficient from a dev time perspective.

Why spend Anthropic developer time on charging you less when an MVP impl can charge you $1 for indexing a simple small Helm chart?

As an aider devotee I am now all in on Claude Coder.

Was fun while it lasted.

It's not much different than your proxy idea. It's implemented as a transformation between the internal message structure (close to Anthropic API's) to OpenAI message spec and vice-versa. Then, it's calling all the other models using the openai-node client, as pretty much everyone supports that now (openrouter, ollama, etc)