What do you recommand to get a Claude-code-like experience in the open-source + local llm ecosystem?
There is nothing at the moment that I would recommend. However I'm quite convinced that we will see this soon. First of all I quite like where SST's OpenCode is going. The upcoming UX looks really good. Secondly because having that in place, will make it quite easy to put local models in when they get better. The issue really is that there are just not enough good models for tool usage yet. Sonnet is so shockingly good because it was trained for excellent tool usage. Even Gemini does not come close yet.
This is all just a question of time though.
The amusing thing is people normally recommend using aider to save money. With Aider, you can control the size of the context window, and selectively add/drop files from the window. I typically aim for under 25K tokens at a time. With Gemini, that's about 3 cents per prompt (and often much less when I have only, say, 10 tokens). So for me, I'd need to do well over 3000 coding prompts a month to get to $100. I simply don't use it that much.
Also, at work, I have Copilot, and one can use Aider with that. So I only pay for my personal coding at home.
Getting to the original question - Aider probably lags Claude Code significantly at this point. It's a fantastic tool and I still use it - primarily because it is editor agnostic. But some of the other tools out there do a lot more with agents.
To give you an idea - my combined AI use - including for non-code purposes - is well under $20/mo. Under $10 for most months. I simply don't have that much time to do coding in my free time - even with an AI doing it!
I find agents do a lot of derpy shit for hard problems but when you've got fairly straightforward things to build it's nice to just spin them up, let them rip and walk away.
Aider feels more like pair programming with an agent, it can kind of be spun up and let rip, but mostly it tries to keep a tighter feedback loop with the user and stay more user directed, which is really powerful when working on challenging things. For stuff like codebase refactors, documentation passes, etc that tight loop feels like overkill though.
Correct me if I'm wrong, but Aider still doesn't do proper tool calling? Last time I tried it, they did it the "old school" way of parsing out unix shell commands from the output text and ran it once the response finished streaming, instead of the sort of tool call/response stuff we have today.
Single-file download, fuss-free and install-less that runs on mac, windows and linux (+ docker of course.) It can run any model that talks to openai (which is nearly all of them), so it'll work with the big guys' models and of course other ones like ones you run privately or on localhost.
Unlike Claude Code, which is very good, this one runs in your browser with a local app server to do the heavy lifting. A console app could be written to use this self-same server, too, of course (but that's not priority #1) but you do get a lot of nice benefits that you get for free from a browser.
One other advantage, vis-a-vis Armin's blog post, is that this one can "peek" into terminals that you _explicitly_ start through the service.
It's presently in closed alpha, but I want to open it up to more people to use. If you're interested, you and anyone else who is interested can ping me by email -- see my profile.
What does that mean? I've never seen any locally run model talk to OpenAI, how and why would they? Do you mean running an inference server that provides an OpenAI-compatible API?
So, if your model inference server understands the REST API spec that OpenAI created way back, you can use a huge range of libraries that in theory only "work" with OpenAI.
Worth clarifying that what the ecosystem/vendors have adopted is the "ChatCompletion" endpoint, which most models are under. But newer models (like codex) are only available under the Responses API, which the ecosystem/vendors haven't adopted as widely, AFAIK.
https://www.app.build/ was just launched by the Neon -- err, Databricks -- team and looks promising.
Yes it's not a standalone CLI tool, but IMHO I'd rather have a full editor available at all times, especially one that's so hackable and lightweight.