Set up LM Studio for PlotForge

Step 1: Install LM Studio

Download LM Studio from lmstudio.ai. It's available for Windows, macOS, and Linux. Install and launch like any other desktop app.

Hardware: 16 GB RAM is comfortable for the recommended 7–8B-parameter models. Apple Silicon and PCs with a recent NVIDIA GPU run dramatically faster than CPU-only.

Step 2: Download a writing-friendly model

Open LM Studio, click the search icon in the left sidebar, and download one of these prose-friendly models to start:

• llama-3.1-8b-instruct: solid baseline, ~5 GB, runs on most modern hardware
• mistral-nemo-12b-instruct: better prose quality, ~7 GB, needs ~12 GB RAM free
• qwen2.5-14b-instruct: strong creative writing, needs more memory

Stick with instruction-tuned variants. Base models tend to free-associate without following the prompt structure PlotForge sends.

Step 3: Start the local server

In LM Studio, switch to the Local Server tab (icon: two arrows). Load your downloaded model from the top dropdown, then click Start Server. By default it binds to http://localhost:1234.

Leave the CORS and Headless toggles at their defaults unless you have a specific reason to change them. PlotForge Desktop talks to LM Studio over the local network only.

Step 4: Connect PlotForge Desktop

In PlotForge Desktop, open Settings → Local AI. Set the provider to LM Studio and the URL to http://localhost:1234. Click Test Connection: PlotForge will list the loaded model and confirm it's reachable.

That's it. Drafting, refining, and Scene Surgery now route through LM Studio. Your manuscript stays on your machine.

Troubleshooting

• Test Connection fails: confirm LM Studio's Local Server tab shows “Server is running” and the port matches what PlotForge is pointing at.
• Slow generation: close other heavy apps, try a smaller model (7–8B), or set GPU offload to 100% in LM Studio if you have a GPU.
• Strange JSON errors on outline generation:some smaller models struggle with structured output. Switch to a 12B+ model or use the manual outline lane and only invoke AI on individual scenes.