Once your model can hold a chat, start with a short edit and a question about a document you know. These tasks let you see whether it preserves your facts and finds the source text you need. If you're still getting started, install LM Studio and run your first chat or choose a local model for writing.
Save a reusable writing preset
A preset combines system instructions and generation settings for reuse across chats. System instructions tell the model how to handle each request. For a writing preset, try asking it to use plain English, keep names and dates, and flag missing information. Keep the audience and purpose of a particular piece in that chat's prompt so you can reuse the preset for other work.
- In Chat, open the presets dropdown in the right sidebar and choose + New Preset.
- Name it Fact-preserving edit. Enter the system instructions and keep your usual generation settings as a starting point. Save the named preset.
- Open a fresh chat, select that preset in the sidebar, and try the short edit below.
Use the named preset in a fresh chat and submit the Priya/Martin passage below. Verify who sent the draft, who owes comments, and the reply deadline before comparing tone. If you change temperature, keep this passage and the model fixed.
LM Studio's per-model defaults control how a model loads. They include context length, which limits how much text the model can work with at once, and GPU offload, which controls how much work the GPU takes on. Change these defaults in My Models if you want the model to use them the next time it loads. A longer context can use more memory. Keep the load settings fixed when comparing writing presets.
Try a short edit
Paste this into a fresh chat. Check the revision against the original for names, dates, times, and who is responsible for each action.
Rewrite the paragraph below as a short update for the project team.
Keep every name, date, time, and factual detail. Preserve who is
responsible for each action. Return only the revised paragraph.
On 12 May, Priya Shah sent the first draft to Martin Cole. Martin
will return comments by 3 p.m. on 15 May. Priya will send the revised
draft to the project team on 18 May.
Find a quote in an attached document
Attach a non-sensitive sample with a deadline you can locate yourself, such as a short public application notice. Run the prompt below. Then open the original and check the exact words, the cited location, and any nearby exception or qualification before you reuse the quote.
Using the attached document, find the sentence that states the
application deadline. Quote it word for word. Give the section
heading and page number when available. If either is unavailable,
say so. If you can't find the deadline, say "I couldn't find the
application deadline in the attached document." Do not guess.
LM Studio can process an attached document on your computer when you chat with a downloaded local model. It can include all the extracted text from a short attachment if it fits in the model's context. For longer files, it can search for relevant passages and give the model those passages to answer from. The search can miss a clause or exception. If an expected quote is missing, try a term or heading from the original, or give the model the relevant excerpt. For scanned pages, text extraction, and longer files, see Chat With PDFs Locally.
Use the model from Brave Leo if you prefer a browser
Brave Leo's Bring your own model option can send requests to a compatible endpoint. Use Brave's setup instructions with the OpenAI-compatible Chat Completions endpoint from your LM Studio server. For a connection on the same computer, keep the server bound to the device and use its loopback address, such as 127.0.0.1. Leo sends your request and any included page text to the selected endpoint. A remote endpoint receives that material off your device.