# What to run, on the 'idle' local LLM server? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/4339d4ecd6539569

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-09-26 04:44 ingested 2026-09-27 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1wqhgva/what_to_run_on_the_idle_local_llm_server/)

## retained text



```
It started as overnight model benchmarking, then I squeezed a last bit of performance, in critical functions, of the my astrometry app, by extended running autoresearch extension of the [pi.dev](http://pi.dev) coding agent.

My 128GB GMKtec X2, on the balanced performance settings, is quite efficient and capable to forge 80 million Qwen3.8 Flash Next tokens a month, for about $6 electricity consumed.

Soon I'm going to stay without code to optimize and I'm not eager to vibe code arcades and other unsolicited demo apps. My regular usage is about 30 million local tokens a month and remainder will be 'gone with the wind'.

Now, we come to the question from post title. I was thinking about refining Karpathy's wiki, or RAG my codebase, but here we probably have people smarter then I am, with better ideas.
```



document fetched 2026-09-27 00:39 · canonical https://reddit.com/r/LocalLLaMA/comments/1wqhgva/what_to_run_on_the_idle_local_llm_server
