Nicely done! I've been enjoying using it. Thanks for sharing.

Personally, I disable pre-commit hooks because they’re annoying and slow me down. Pre-push hook would make more sense. Secondly, i’d ideally want to use my codex/ claude subs for this, not an api key

I agree, pre-commit is way too annoying. We're working to add a pre-push option and make it the default.

You made a great point about the codex/claude subs. I'll create an issue for this and we'll try to implement it. One challenge that comes to mind is that the output will be more unstable compared to the API so it might be more challenging to parse the findings.

Very interesting ideas. Thanks for trying out the tool!

I use OMP. Was on Pi but switched over because I like OMP's subagents implementation. Use 5.5/4.8 for coming up with a very detailed plan, ds4pro/mimo for implementation of easy features, 5.5/4.6 for larger/harder features, and 5.5/4.6 for a review. I use them in onorca.dev(settled here after trying just about any "agent ade". i also sometimes use Zed as well. I'm curious to know if folks use any meta-harnesses?

How do you a) share context/chat etc with each other? plugin? have them make a file? Why did you opt for either of them instead of something like opencode that supports those models?

Currently we're mostly working on separate issues or aspects, so not much context sharing needed. We have some skills that we share in a shared network directory.

As for why using both instead of OpenCode or similar, well, subscriptions take you a long way for not very much. Also security is a thing here now so sec guys feel more comfortable with big names rather than "random" open source.

Opencode and other harnesses like Pi do support codex and claude(for now) subscriptions, but the point about security makes sense.

I wonder what if anything has changed ever since this article. Is llm-based compression more mainstream?

Fabrice Bellard did something with neural nets and a transformer model [1] that was very successful.

I suspect that LLMs wouldn't be ideal to use as compressors, because they are large, consume a lot of resources, and are constantly changing. You need the model to produce exactly the same output at encoding and decoding time, or else you get gibberish.

[1] https://bellard.org/nncp/

This post offers virtually no data to back up their objections and reads as LLM-generated for the most part. I

Incredibly cool

I hope they bring it to their apis, especially v4flash. I find myself using mimo 2.5 more since it supports vision and makes it cheap for doing e2e tests with playwright or similar

They have been recently scaling their team maybe we will updates sooner

Have you tried glm 5.2? It’s better than ds4pro and very close to opus4.6. Alot more expensive than ds4pro/mimo but significantly chesper than opus 4.6 for 90% of the quality

I haven't yet, no.

I alternate between Opus 4.8 and DS 4. If I remove my Anthropic subscription I will get GLM 5.2 in place of it.

You can BYOK with openrouter and pay nothing up until 1mil requests/month iirc which is pretty generous for one person