I wrote a guide on setting up Ollama with WindowSill a few days ago. Ollama is great, but it's mostly a command-line tool with a limited set of models. You pull models by typing commands in a terminal, you manage them by typing commands in a terminal, and if something goes wrong, you debug it in a terminal. The GUI they provide is quite limited as of 2026.
If you'd rather browse models visually, click a download button, and see everything in a proper GUI, LM Studio is the better fit.
This guide covers the same outcome: grammar checking, rewriting, tone adjustment, and translation running locally on your machine. Your prompts and responses stay on your computer; LM Studio doesn't send them to any cloud inference API. The only difference is the tool that runs the model.
What you'll need
- WindowSill installed from the Microsoft Store (free tier is fine for setup, but AI features require WindowSill+)
- LM Studio installed from lmstudio.ai
- 16 GB of RAM recommended (8 GB can work with smaller models and reduced context, but you'll hit limits quickly)
- A few gigabytes of disk space for the model you choose
Like Ollama, LM Studio runs on CPU if you don't have a dedicated GPU. An NVIDIA GPU with 6+ GB of VRAM will speed things up, but it isn't required.
Step 1: Install LM Studio and pick a model
Download LM Studio from lmstudio.ai and run the installer. When you open it, you'll see a model browser with a search bar. This is the main advantage over Ollama: you can search, filter, and compare models without leaving the app.
For writing tasks on a medium-range machine (16 GB RAM, no GPU or a mid-tier one), here are solid picks:
| Model | Size | Best for |
|---|---|---|
| Qwen 3.5 4B | ~4.5 GB | Good balance of speed and quality for grammar, tone, rewrites |
| Deepseek R1 8B | ~8 GB | Better output quality, needs more RAM |
| Phi-4 Mini 3.8B | ~3.5 GB | Fast responses, decent grammar on lighter hardware |
Search for one of these in the model browser, pick a quantization (Q4_K_M is a good default for balancing size and quality), and click download. LM Studio handles the rest. Sizes are approximate; exact GB depends on the quantization (Q4_K_M vs Q5_K_M vs Q8_0) and the specific build.
A note on quantization: You'll see options like Q4_K_M, Q5_K_M, Q8_0 next to model downloads. Lower numbers mean smaller files and faster inference, but slightly lower quality. Q4_K_M is the sweet spot for most writing tasks. Q8_0 is better quality but needs more RAM. Start with Q4_K_M and upgrade if you want to.
Step 2: Start the local server
LM Studio needs to run a local server for WindowSill to connect to it.
- Open LM Studio
- Go to the Developer tab (or Local Server tab, depending on your version)
- Select the model you downloaded
- Click Start Server
The server runs at http://localhost:1234 by default. You'll see a green status indicator when it's ready. Leave this running while you use WindowSill.
Step 3: Connect WindowSill to LM Studio
In WindowSill:
- Open the Settings from the command bar
- Go to the AI Writing & Analysis section
- Under AI Providers, select LM Studio
- The endpoint should already show
http://localhost:1234(the default) - Click the refresh button next to the model dropdown, and select the model you started in LM Studio
That's it. WindowSill now sends AI requests to LM Studio instead of a cloud service.
Step 4: Test it
Open any app where you write: Word, Outlook, Notion, Slack, a browser, anything. Type a sentence with a deliberate mistake:
Their going to the meeting tommorrow at 3pm, can you confirmed?
Select the text. WindowSill's Analyze / Rewrite sill should appear on the bar. Hit the Spell Check option.
If everything is connected, the corrected text comes back after a few seconds:
They're going to the meeting tomorrow at 3 PM. Can you confirm?
The first request might take a moment while LM Studio loads the model into memory. After that, responses are faster.
What you can do with local AI
Once connected, all of WindowSill's AI writing features work through your local model:
Grammar and spell check. Select text in any app, fix errors without opening a separate tool.
Rewriting. Highlight a paragraph and ask for a rewrite. Good for polishing drafts or simplifying dense writing.
Tone adjustment. Switch between professional, casual, confident, and six other tone presets. You can also create custom tones for recurring needs like "customer support reply" or "executive summary."
Translation. Select text and translate to any of 35+ supported languages. Quality depends on the model. Qwen 3.5 handles common language pairs (English/Spanish, English/French, English/German) well. Less common pairs benefit from a larger model.
Custom prompts. Build reusable prompts with variable injection. For example, a prompt that takes selected text and converts it into a formatted meeting recap.
Summarization. Select a long email or document section and get a summary.
All of this happens on your machine. Your prompts and responses stay local; LM Studio only talks to the internet to search and download models and check for updates.
LM Studio vs. Ollama: which should you pick?
Both tools do the same thing at the core: run a local model and expose an API that WindowSill connects to. The difference is in how you interact with them.
| LM Studio | Ollama | |
|---|---|---|
| Interface | GUI with model browser | Mostly command line |
| Model discovery | Search and browse in-app | Find model names online, then pull via CLI or GUI |
| Model management | Visual list, click to load/unload | Terminal commands (ollama list, ollama rm) |
| Server management | Click Start/Stop | Runs as background service automatically |
| Default port | localhost:1234 |
localhost:11434 |
Pick LM Studio if you prefer a visual interface and want to browse models without leaving the app. Pick Ollama if you're comfortable in the terminal and want something lighter running in the background.
You can also install both. WindowSill supports multiple AI providers at the same time, so you could have LM Studio and Ollama configured side by side.
Performance tips
Local models are slower than cloud APIs. Here's how to keep things comfortable:
- Unload models when you're done. LM Studio keeps the model in memory while the server is running. If you need the RAM for something else, stop the server or unload the model from the Developer tab.
- Use a smaller model for quick tasks. Grammar checks and spell fixes don't need a large model. Qwen 3.5 4B or Phi-4 Mini handle these well and respond faster than an 8B model.
- Use per-prompt model selection. WindowSill lets you assign different models to different tasks. Use a fast, small model for grammar and a larger one for complex rewrites. This also works across providers: keep local models for sensitive content, and use a cloud model for non-sensitive tasks.
- Upgrade your RAM. 16 GB gives you room to run LM Studio alongside a browser, editor, and other apps without swapping to disk. With 8 GB, LM Studio works but your system will feel slow when a model is loaded.
When local isn't enough
Local models are good, but they're not as capable as the latest cloud models for every task. Here's where a cloud provider still wins:
- Long, complex rewrites where nuance matters. Modern GPT-5-series and Claude Sonnet-series models still produce more polished output for multi-paragraph rewrites.
- Uncommon language translations. Local models handle major languages well, but for something like Finnish to Japanese, a cloud model with broader training data does better.
- Very fast response times. Cloud APIs respond in 1-2 seconds. Local models on CPU might take 5-15 seconds for a similar response.
You don't have to pick one. WindowSill supports both local and cloud providers at the same time. Route sensitive content through LM Studio and non-sensitive content through a cloud API. Per-prompt model selection makes this easy.
Frequently Asked Questions
Does LM Studio keep everything local?
Yes. LM Studio runs models on your hardware and exposes only a local API (localhost:1234). Your prompts and responses stay on your machine. The only data LM Studio sends over the network are model search and download requests and software update checks. You can verify this by disconnecting from the internet and confirming the AI features still work (with an already-downloaded model).
Which LM Studio model is best for grammar checking?
For grammar and spell checking on a medium-range machine, Qwen 3.5 4B (Q4_K_M quantization) offers a good balance of quality and speed. Deepseek R1 8B is better if you have 16+ GB of RAM and want higher quality output.
Can I use LM Studio and a cloud provider at the same time?
Yes. WindowSill supports multiple AI providers simultaneously. You can configure LM Studio for sensitive tasks and a cloud provider like OpenAI for non-sensitive tasks, then select which model to use on a per-prompt basis.
How much disk space do I need?
A 4B parameter model at Q4_K_M quantization takes about 3-5 GB of disk space. An 8B model takes about 5-8 GB. If you want multiple models available, plan for 15-20 GB. You can delete models from LM Studio's model manager at any time.
Does this work without a GPU?
Yes. LM Studio supports CPU-only inference. It's slower (5-15 seconds per response instead of 1-3 seconds with a GPU), but it works. Any modern CPU from the last few years can handle it.
I already use Ollama. Should I switch to LM Studio?
Not necessarily. Both work well with WindowSill. If you're happy with Ollama and comfortable in the terminal, there's no reason to switch. LM Studio is a better starting point for people who prefer a visual interface. You can also run both and let WindowSill connect to whichever you prefer for a given task.