# Hermes profile leonard (local Windows): main model = gpt-oss-20b via freetoken e

category: tool
tags: hermes,profile,leonard,freetoken,models
stored: 20260915-151306

Hermes profile leonard (local Windows): main model = gpt-oss-20b via freetoken engine :1919 (Tony's choice, 2026-09-15); auxiliary compression = nous meituan/longcat-2.0:free. Engine :1919 effective context ~8.5K tokens (VRAM-capped). Hermes >= 0.21.2 hard-gates turns: compression model must be >=64K context — local freetoken models fail, compression must route remote.