Can someone who has done this, simplify and say what specs we need on a `local computer` to run and test this, with a reasonable speed?

Excluding MBP M5 128GB.

It is slow but usable via opencode on a mbp m3 max 48 gb. So I guess hosted is still the better option for most people.

The local models are considerably better relative to the hosted ones compared to 6 months ago. Bench maxing or not - stuff is happening in this area for sure.

I've got the unsloth q4_K_XL 35b running in llama.cpp on an i9/64G/4090 machine doing double-digit tokens per second with a 90k+ token context window available. The model's completely in VRAM.

The replies in the Issue from the maintainers were clear. At some point in the future, they will probably accept PR submissions from LLMs, but the current policy is the way it is because of the reasons stated.

Honestly, they recognized the gravity of this first bot collision with their policy and they handled it well.

What policy are you referring to? Is there a document?

Bot is not a person.

Someone, who is a person, has decided to run an unsolicited experiment on other people's repos.

OR

Someone just pretends to do that for attention.

In either case a ban is justied.

Yep, there's nothing wrong about walled gardens. They might risk to become walled museums, but it's their choice.

Moderation is needed exactly because it's not a walled garden, but an open community. We need rules to protect communities.

Humans are no longer the only entities that produce code. If you want to build community, fine.

Generated code is not a new thing. It's the first time we are expected (by some) to treat code generators as humans though.

Imagine if you built a bot that would crawl github, run a linter and create PRs on random repos for the changes proposed by a linter - you'd be banned pretty soon on most of them and maybe on Github itself. That's the same thing in my opinion.

Many open source contributions are unsolicited, which makes a clear contribution policy and code of conduct all the more important.

And given that, I think "must not use LLM assistance" will age significantly worse than an actually useful description of desirable and undesirable behavior (which might very reasonably include things like "must not make your bot's slop our core contributor's problem").

There is a common agreement in the open source community that unsolicited contributions from humans are expected and desireable if made in good faith. Letting your agent loose on github is neither good faith nor LLM assisted programming, it's just an experiment with other people's code which we have also seen (and banned) before the age of LLMs.

I think some things are just obviously wrong and don't need to be written down. I also think having common rules for bots and people is not a good idea, because, point one, bots are not people and we shouldn't pretend they are

It doesn't address the maintainer's argument which is that the issue exists to attract new human contributors. It's not clear that attracting an OpenClawd instance as contributor would be as valuable. It might just be shut down in a few months.

> The bot (allegedly) did a better performance improvement than the maintainer.

But on a different issue. That comparison seems odd

The ends almost never justify the means. The issue was intended for a human.

Do the means justify the ends?

You missed the rest of it.

> “So you keep pushing on making it constitutional. I’m totally supportive of that,” Karp said.

Which in context means, you keep pushing to make the military be sure they are operating within the bounds of the constitution.

When he says "push to make it constitutional" what he means is push to make them comply with complex rules.