Everything below is a per-request sampling setting - llama-server accepts these directly in each chat request, so Apply just saves new defaults server-side and takes effect on your very next message, no restart. (Context size, GPU layers, and which model file loads are different: those are set at process launch in start-server.sh and genuinely do need a restart to change - they're not editable from here.)

Temperature Top P Top K Min P Repeat penalty Presence penalty Frequency penalty Max tokens DRY multiplier Autoconnect to last used model on page load

Apply

Identity and long-term memory are private to your own account by default. Individual facts can be marked "Public" in the Memory modal - that only means other users who separately opt in below can see that one fact; it doesn't change anything for you.

Aware of other users' shared memories

For the create_image tool (also needs the "Image generation" Power enabled - see the Powers tab). Your own ComfyUI instance, not shared with other accounts.

MCP Model (optional) -- none --