← sources

Local models for personal agents. Is it worth it?

source: reddit (reddit:LocalLLaMA)published 2026-09-20 14:20ingested 2026-09-21 02:38

retained text

I have been working with agents for coding since Claude instant (not funny) and with coding being my main focus I kept working with the newest and fanciest models while improving the setup, the harness, the memory, etc.

Recently I have been experimenting with Hermes to take care of some personal matters (e.g., deep research stuff, managing calendars for me and my partner, small documents/letters for day to day stuff like maintenance of the building). I am still trying to figure out how many things I can do with it.

What is shocking for me is: I did all of that using GPT Luna. I am very very very surprised at how far that went. I could be happy with thai but that would be boring :P. I have been lurking here for a while but I was under the impression that local models would not have worked for my coding work. But now... Can I use a local model for Hermes? I can probably buy a Mac mini for 2k and host something around 25B. 

My questions:

\- who is running models for personal assistants, which models do work well? Any comparison with "frontier" models?

\- what's the minimum config I should invest into to have a decent setup?

\- anything I am overlooking? (I know I have to pay for electricity)

\- what did you build in this space? Satisfied?

document fetched 2026-09-21 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1wlic2r/local_models_for_personal_agents_is_it_worth_it