Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

As a recent M6 Mac mini owner, HELL TO THE YEAHHH BABYYYYYYY

I use oMLX btw, on 32GB URAM, running Hermes with Qwen 3.6 @ 128k context window.

I realize I am a smalltimer but for general day to day sensitive agentic stuff, this is the entry point.

 help



if you're running qwen, check out what happens when you swap oMLX for kapitsa

its built just for qwen, healthy competition on tokens/sec


What is the experience of these local LLM models on this type of hardware? what are they capable of ? what do you use them for?

I do agentic coding and agentic search and document analysis just like what I do with Claude

they have 128k - 256k context windows now, subagents use the same weights loaded in memory so you have spawn many subagents at the same time just like Claude Code, and the tokens/sec has is like 150-200

I just do different projects based on its complexity or liability or its moat


A-OK, I would say. The pinnacle would be that I managed to develop a purpose made android app for my 2016 Honda dash unit with Hermes.

Hermes runs my personal to-do, manages 2 calendars, triages forwarded bills via it's own email address, orchestrates my docker container fleet. It also tends to a partly botched llm-wiki, that is built on Obsidian.


never heard of it, will check!



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: