it’s definitely a real thing you can do using docker, but don’t have the larger context of how it came to this :)

it is pretty snappy

there seem to be a whole a bunch of these projects now

oh I didn’t even consider the last bit

Steam Deck is a really under appreciated Linux machine in my opinion.

ATLAS has some benchmarks in the repo, and it’s comparable to opus 4.6, you don’t actually even need 128gb model for that. An 8 bit quantized model will run with around 32gb and still perform quite well.

Yeah that’s true, depending on the electricity costs, you could be better on a subscription. Especially with DeepSeek, which is incredibly cheap now.

You should be able to get very decent performance with 128gb vram running Qwen 3.6 with something like https://github.com/itigges22/ATLAS especially if you run MTP https://huggingface.co/unsloth/Qwen3.6-27B-MTP-GGUF

A friend of mine gets something like 50 tokens a second with it, and output quality is quite decent.

Agreed, all the mass layoffs are really helping build class consciousness among the labor aristocrats.

and?

Yeah, as I recall it was just codex repackaged, but there is also the aspect of the tooling itself that the model interacts with which they massively fumbled. People will still use codex on its own, but nobody wants to ever use copilot.

Oh that’s the magic of tools like opencode, you run it in a folder and it acts as a harness for the model where it can interact with the filesystem. You could do the same with an IDE as well, making your own agentic harness is actually pretty straight forward. So you could make a plugin that talks to, say, ollama https://ampcode.com/notes/how-to-build-an-agent

I find these kinds of projects are neat, but if I’m being honest, I tend to just keep plain markdown files as well. The only thing I find that’s missing with that is searchability. Once you get enough files, it can get unwieldy. Although, I’ve been playing around with just using a local model lately as the interface. You can throw opencode at a folder with the files, and even a small model can find stuff fairly competently there.

kinda of hilarious that MS has a stake in both of the worst commercial LLM offerings

ooh neat

yeah logseq is great, but does need a bit of upfront investment

yup

I feel very validated :)

🤣

if somebody already has access to your machine, but doesn’t have root privileges