vp

@JustJoshLife Nice, followed and reviewed for a tiny bit of boost.

Our Podcast is on Spotify! open.spotify.com/show/4Xlju1Sq2… If you enjoy listening to podcasts on your way to work, at the gym, or anywhere else, here's the link

Open sourced htmui: a typed HTML component library for Python @htmx_org apps (for those tired of Jinja) It has most of the @basecoatui components (shadcn for non-React) plus some utilities. Credits to @hunvreus Also serves as a holm + htmy demo. htmui.vercel.app

@championswimmer Models with way less world knowledge but more intelligence, augmented by good tooling and tool use would be a positive direction imo. Distilled intelligence instead of knowledge, better security, cheaper, possibly less hallucination, easier customization through the harness...

The latest Chinese labs open model trio (GLM 5, Kimi K2.5 and Minimax M2.5) are extremely good for agentic use and tool calling showing a lot of post-training that went into that. But their pre-training is very poor (world knowledge and scientific reasoning)

@dhh K2.5 was the first open weight model that was pleasant to work with. GLM feels quite similar. Big upgrade for those who want to avoid using OpenAI/Anthropic. Minimax 2.5 is okay for simple tasks, but not in the same league.

Kevin: "I just raced Claude and Kimi K2.5 against that bug that Ryan was talking about. K2.5 fixed it in 21s. Claude took just over a minute to make the plan, then about 2 minutes to execute on it. Both had the same fix, though." (K2.5 is now my main driver. Opus just backup.)

@JustJoshLife Do you plan have the episodes on Spotify?

The State of Laptops in 2026: youtu.be/MOEmuQKMPjg

@buggyflex777 There are extensions for this I think. And you see your balance in Zen. That's it right now as far as I know. Surely the team is planning something, it's kind of essential. There may be higher priority tasks though, they seem to be working very hard.

Is there no way to count (or estimate) token usages and costs? Surely an app idea in this?

@opencode Love OpenCode and Zen, but this is so misleading... There's also no built-in way to track token usage. 3rd party plugins are not a solution.

we're dropping our kimi K2.5 prices by 20% 3.75x cheaper than sonnet 6.25x cheaper than opus absolutely the lowest price you can get anywhere

@thdxr this would be the responsible version of openclaw... the direction opencode and zen is heading is great, i keep recommending it to everyone

finally got around to setting up an always on opencode server so i can run sessions on any device from anywhere takes a few minutes - showed it off here

@htmx_org This is quite nice! Wondering how I could add it for juniors and mediors in my projects, besides the actual agents.md.

just posted a video for my students explaining why they *have* to write the sort of code that, yes, can be generated by an LLM I am giving them this AGENTS.md file: gist.github.com/1cg/a6c6f2276a… and telling them "Look, here are the guard rails. I can't make you use the guardrails. If you want to be a good programmer though, use the guardrails." I'm moving all my quizzes to in-person on-paper no digital devices, and that's what their grades will be mostly based on because there's simply no way to tell if a successful assignment was done by the student or an LLM. adapt. improvise. re-adapt.

@thdxr join mastodon, it's pretty good

this website was pretty great because it was the only place where you had a shot at someone big reading your reply this sometimes would change your life now no one is reading replies because it's all ai reply crap

@thdxr So true. This is one of the things I really like about open source, as the maintainer, you can just do this, delay and ship the right stuff. It's rarely possible at "work", business or management usually gets in the way.

sometimes there's a clear feature users would benefit from but the only implementation i can think of feels off sometimes i won't ship it until i can figure out what's going on - holding this line is what results in long term quality LLMs make it harder to hold this line

@hunvreus holm (volfpeter.github.io/holm/) may be what you're looking for. Currently it's like FastHTML on steroids thanks to FastAPI and htmy. I plan a feature that lets you use HTML and MD everywhere. The primitives are ready in htmy. Wish I had a week to get it done already...

I really wish I had the time to build a FastAPI + HTMX framework. I find working with Python more enjoyable than JavaScript (especially with React and RSC), but I find myself having to write a lot of boilerplate over and over to make things work with HTMX. I've looked at FastHX and FastHTML, but I don't like the idea of abstracting away HTML. HTML is actually great, especially when extended with HTMX. I don't want to be forced to work within the constraints of Python to generate my front-end. Maybe I'll just try and list the patterns I've used so far and see if I can get Claude/Codex to come up with something...

@klusterai Will you stop continuously spamming users of other code review services??? GDPR time, delete my personal information from your system.

@harjotsgill @coderabbitai I haven't tried it. I don't plan to, to be honest. If I want an AI to address an issue, I'd just trigger it locally with the tooling I normally use.

Thanks for the feedback. Our settings UI is currently auto-generated from the settings JSON schema. Our team is currently working on shipping a simpler version of it, which is curated by hand. I am curious, did you get to try the planning feature on issues yet? That was the reason we were being overreaching to post those comments (which are now switched off by default).

@coderabbitai It's incredibly frustrating that the automatic issue commenting feature is opt out and not opt in. It's spam! And it can't even be disabled on the dashboard, I need to commit a config file to each repo to stop it. 👎

FastHX (Python server-side rendering and @htmx_org library for @FastAPI ) now supports async HTML streaming: volfpeter.github.io/fasthx/ This means holm also supports this, although it's not yet the default: volfpeter.github.io/holm/ Happy new year!

@harjotsgill @coderabbitai Thanks for the info. The settings UI seems a bit overcomplicated. Less would be more in this case. FWIW: "Enable automatic issue enrichment" was already off and I still got the comments. Now I turned off everything on that tab, but I don't dare to remove the config file.

@volfpeter @coderabbitai Hi! Sorry about the issue. The UI should have an option to turn it off under the “Issue Enrichment” tab.

@ChaiKamNamak @vboykis I don't get your question, I said the exact opposite... They're great, I'm using both wherever I can, and I'm pretty sure I'll switch from mypy to ty for static type checking quite soon.

@volfpeter @vboykis What’s wrong with ruff and uv?

started using Ty lsp in both PyCharm and VSCode on Friday, literally like a light went on in my codebase. This is Civilization

@vboykis Yeah, that's a more straightforward switch. Planning the same in @zeddotdev - I used pyright as the LSP until now (it's okay) even though they shipped ty for quite a while.

@volfpeter Mypy, just lsp for now