Openclaw with small local model?

Openclaw Need The 80B+ models.
Big model are what need to for tool calls.

I dont have theese resource GB.

I want to be able to run openclaw "app" on multiple PC with 24gb RAM

tested the following models on Mac Studio M3 Ultra + 512GB unified storage

- qwen 2.5 coder 20b

- qwen 3 20b

But these models don't work properly with OpenClaw. The main problem is in the tool calls.

Im looking for any suggestion how t chunk the context or using a sliding window approach or one model for tool calling another support MCP and One write json instruction...

uppppp

no, dont do it. it will be useless.

use kimi 2.5 if its about cost