Hacker News Reader

Comment by umvi | Hacker News Reader

Can't wait until we get really well designed harnesses written in a compiled language that doesn't bloat my computer. Will probably have to be written by a company other than OpenAI because codex is like the exact opposite of everything I just listed

View on HN

Found this one: https://usehax.dev/

Anyone experiences with it?

Yeah, using it a bunch with local LLMs. It's fast (small system prompt), focused on the basics, made around simple choices, and possibility to call it in scripts.

So far I have very positive experiences, the results are generally better (and not just faster) than in other harnesses with the same models. I've done the same (user) prompts for tasks in a range of them, and compare the outcome to eval between them. Simple (harness) beats complicated one many times. I do wish it could use MCP though, but not as often as I think, and CLI tool usage often compensates on this front with little extra effort)

That is now my usuall go-to harness for things usually. Still figuring out the edges though (when we get off the happy path, like models reaching their max context, or such...)

It's very light, recommed you try it out, and see if it fits your toolbox.

Just tried. And wow!! I opened my mac's activity monitor to see how much load it was putting on my system, and i had to scroll all the way down ... unlike those current harnesses that sit right on top (sorted by used cpu, descending).

And it worked. I'm switching to this unless i find a reason not to.

it's okay.

it's minimal and somewhat opinionated, not integrated well with memory subsystems (although that's kind of the point)

my only real criticism is that last I played with it it deals with slow provider responses very poorly , so lots of locally hosted llms require a lot of tuning to get it to handle streaming/non-streaming/timers well.

way less minimal (but still pretty minimal) I prefer Pi; but different markets I think.

hax fit my openbsd laptop well, espescially since Pi and related derivatives are kind of un-spokenly incompatible with the BSDs.

(author passing by) Could you please report the issues you've experienced on the repo? I did quite a bit of testing myself on local models (Qwen, Ornith, Gemma, Muse Glimmer) with llama.cpp, and they all worked fine. Details would be helpful.

Also one brief note for Codex users specifically. hax can work out-of-the-box as codex-cli drop-in replacement (since it can use the already authenticated codex token), so it's super easy to try it - basically it will be a zero-config experience.

Firstly thanks for writing it and making it available. The problem I had when I tried it is it would consistently blow out the small context window of my local qwen and although there was seemingly a config knob I could use to prevent that, seemingly that didn’t do anything.

Thanks for making it :)

That landing page is hard to look at, let alone read. Have LLMs really not yet learned to write webpages that read the sytem theme and look that way? Must be hard.

Codex is written in rust, fyi... Not that I disagree.

Oh, I haven't been keeping up. I assumed it was still TS since it prompts me to run npm all the time.

They just use npm as a binary delivery mechanism, for whatever reason (I guess they expect all their users to have npm?)

It's pretty darn ubiquitous, yeah. Plus enterprises tend to run npm mirrors, so delivery integrates into existing IT compliance frameworks.

Which generates unnecessarily huge binaries.

It's because they statically linked it, and probably rely on a trillion pointless crates (as is standard in Rust development). A good statically linked harness can easily be ~5MB MAX

5MB is still huge.

There's https://3code.capocasa.dev/ which is written in Nim and focuses on token efficiency.

does it achieve token efficiency? vs something like maki? It's not at all clear what the method of that "efficiency" is.

my main issues with maki are that the line editor (what you use to enter your prompts) is very barebones, copy-paste is broken (no middle-click paste, or copy to PRIMARY; implicit copying to CLIPBOARD that can only be used with a mouse), and scrollback is not native. So it's not a real cli, but some sort of weblike tui.

I tried 3code; it does not mess with selection, but code printed in the conversation has | markers in every line, and is thus uncopyable in practice. Don't know how to use that.

I've only dabbled with 3code. The author of 3code is pretty active, so he may take suggestions. I'd agree that copy/paste is important for a good agent workflow.

Thanks for that. Looks like exactly the agent I might have been looking for.

Just write your own harness. Ironically Codex/Claude are both abysmal.

How do you even write a harness that can use a Claude subscription? I thought it was not allowed by Anthropic's ToS and can get you banned?

I don't use Claude, but it appears it is permitted to use Claude as a stdin/stdout -communicated tool using JSON, or use Claude SDK, with the monthly plans. Monthly Max/Team plans also include credits that could be used directly via the API: https://support.claude.com/en/articles/17154008-monthly-api-... .

E.g. OpenCode can be used with all three methods. Looks like it needs a plugin for the two Claude methods, though.

It's trivial to bypass this by tweaking the system prompt in other harnesses. They use string matching and a blacklist.

https://x.com/FlorianKluge/status/2041855675295318039

(Whether that's against the ToS is not clear.)

Amp recently achieved it. Apparently it's permitted as long as you use the Claude Agent SDK. https://ampcode.com/news/use-your-claude-plan

This is bait and switch..

You can’t just use your Claude sub on a third party harness without getting blocked even if you use the Agent SDK. Apparently, Anthropic gives “credits” depending on your sub and you can use those _credits_ when you use third party harnesses or Claude -p mode. That usage is not the same Claude sub one would assume it is… It is all quite deceiving. Don’t get a Claude sub unless you’re okay with being locked into Claude Code…

If you're using the sub, only OpenAI (via the OAuth token) is viable. Claude is API_KEY only

That’s a different discussion.

I suspect a lot of people wouldn't use AI if they had to pay per token at current rates. Yes, it's technically a different discussion, but "just write your own harness" isn't really an answer for the vast majority of us riding on the investor-subsidized subscription plans.

It's allowed, but you get bumped to per token pricing.

Not true; They rolled back some of the changes you're thinking about IIRC.

That defeats the purpose of using the Claude subscription.

I did, and the weird things you have to do to make work at least decent. I’ve been doing evals and I got some things better than opencode but not as good as pi. Though my goal is to optimize for a browser and browser based llm so it can create stuff in a web container or at least generate ui by requesting json schema (a2ui style).

Omp

They're not made of fairy dust; you can build your own.

Reply on news.ycombinator.com