Guides › LM Studio
AI review on your computer
LM Studio runs AI models on your own computer. Connect it to Novelar and the chapter you review and your novel's notes never leave your machine, the review works offline, and there are no usage fees.
Before you start
- A supported computer: a Mac with Apple silicon (M1 or later) on macOS 14 or later, Windows (x64 or ARM) or Linux (x64 or ARM64). LM Studio doesn't run on Intel Macs.
- 16 GB of memory or more, recommended. The more memory, the bigger the model you can use and the better the review.
- Disk space: each model takes between 3 and 20 GB.
- What to expect: a local model is slower than cloud services and usually less subtle. For proofreading and style, a mid-sized model does well.
1. Install LM Studio and download a model
- Download LM Studio from lmstudio.ai and install it.
- In LM Studio's model search, pick a chat model that writes well in your novel's language. Families such as Qwen, Gemma or Mistral work well in English and Spanish. LM Studio shows which ones fit your computer.
| Computer memory | Rough model size |
|---|---|
| 8 GB | Up to 4 billion parameters (4B). Simple reviews. |
| 16 GB | 7B to 12B. A good place to start. |
| 32 GB or more | 14B to 32B. Finer style and dialogue reviews. |
2. Load the model with enough context
Novelar sends the model the excerpt you review and your novel's notes (cards, outline and general notes). All of it has to fit in the model's context, which LM Studio lets you set when you load it.
- Load the model from LM Studio's top bar and open its load options.
- Raise Context Length to at least 16,384, or 32,768 if your chapters are long or you have many cards. More context uses more memory.
3. Start LM Studio's server
In the Developer tab, turn on the Start server switch. If you prefer a terminal:
lms server start
The server listens on http://localhost:1234 and only accepts connections from your own computer,
unless you turn on Serve on Local Network.
4. Connect it in Novelar
- Open Settings (⌘ , on a Mac, Ctrl , on Windows and Linux) and go to Services → AI review.
-
Choose LM Studio · local model. The Server address is already
filled in (
http://localhost:1234). Leave Model empty to use whichever one is loaded, or type its identifier exactly as LM Studio shows it. - Press Test connection. It should say Connected and the model's name.
- Press Use this one.
5. Ask for a review
Open View → AI Review (⌘ ⇧ E on a Mac, Ctrl Shift E on Windows and Linux), choose the kind of review (Proofreading, Style, Dialogue or Continuity) and press Review. The panel says the review runs in LM Studio, on a local model.
With a local model, a long chapter can take several minutes (Novelar waits up to 15). To speed things up, select an excerpt before pressing Review. The Continuity review includes the rest of the manuscript, so it needs the most context.
If something doesn't work
- “LM Studio cannot be reached”. Open LM Studio and turn on Start server in the Developer tab, or check the address in Settings.
- “LM Studio has no model loaded”. Load one. If you typed a model in Settings, check that the identifier matches LM Studio's exactly.
- “Don't fit in the model's context”, or the review is cut off. Load the model again with a longer Context Length, or review a shorter excerpt.
- The suggestions make no sense. Try a bigger model or another family; small models make more mistakes, especially in Spanish.
Optional: LM Studio on another computer on your network
If you have a more powerful computer at home, you can run LM Studio on it: turn on
Serve on Local Network on that computer and, in Novelar, enter its address (for example,
http://192.168.1.20:1234). The text leaves your computer, but not your network. Novelar doesn't send
an access key to the server, so only do this on a network you trust, never on a public one.
Want the checker's text to stay on your computer too? Follow the LanguageTool guide.