Frontier vs local LLM behaviour

[giarmdok]

I've been working on some projects both for home and for work using OMO-Slim. I have tried very hard to keep my configs between home (using frontier models from ChatGPT and Google) and my work local LLMs which has various models available , GPT, Gemini, Nemotron. On the frontier models I notice a lot more activity handing off to the subagents than I do on the local models. The projects are dissimilar (.NET at home, Python at work) but I'm wondering if there are tweaks or things I should consider for it to be more effective at work on the local models. It rarely calls a subagent unless I direct it to.

[alvinunreal]

You will need to append to orchestrator prompt for less smart models, which in more details and tricks explains delegation rules.
Prompts also usually needs to be tuned per model,

[mhenke]

Answered by @alvinunreal: append to the orchestrator prompt for weaker models; prompts need per-model tuning.