diff --git a/composte-hermes.yml b/composte-hermes.yml
index 7957454..8f58b35 100644
--- a/composte-hermes.yml
+++ b/composte-hermes.yml
@@ -32,6 +32,8 @@ services:
        - --served-model-name=qwen3-14b-awq
        - --enable-auto-tool-choice
        - --tool-call-parser=hermes
+       - --enable-prefix-caching # 2x RTX 3060 (12 GB)
+       - --prefix-cache-threshold=0.8 # 24 GB VRAM
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8005/health"]
       interval: 30s