Local AI Local AI Foundations

Chat with Local Models

Use the built-in chat UI, compare models, and manage memory

50 of 66

The chat controls

After loading a model in LM Studio, the right panel shows:

  • System prompt: persistent instructions for every turn
  • Temperature: lower for code, higher for creativity
  • Max tokens: cap the response length
  • Context length: how much conversation history to keep 
System prompt: You are a terse, expert coding assistant.
Temperature: 0.2
Max tokens: 2048
Context length: 4096

Compare models side by side

Open two chat tabs, load different models, and ask them the same question. This is the fastest way to learn which model fits your workflow.

Manage GPU offload

In Model Configuration, adjust GPU offload layers. More layers = faster, but uses more VRAM. Start with the maximum your hardware allows. If generation is slow or crashes, reduce layers.

Save chats

LM Studio saves conversation history automatically. You can export a chat from the menu if you want to keep a record outside the app.

Working out which model to run this on? See The Codex. Packaging it as a reusable skill? See The Armory.