You can enable the lm studio server and use any openai compatible harness to use the models that are running inside it. OpenCode, pi, even Claude and Codex...