If you have read anything about running AI on your own computer, the name LM Studio has probably come up. It is one of the most popular ways to download and run open-weights language models locally, without a cloud account and without sending your conversations to anyone. But the marketing rarely explains what it actually is, what it can and can't do, and who it's really for. We installed it on a clean Apple Silicon Mac, ran a week of real work through it, and wrote down exactly what we found.
Want the short version? Jump to the summary table. Reading this on a phone where LM Studio can't run? PocketLLM brings the same private, on-device experience to iPhone and iPad — join the waitlist.
LM Studio is a free desktop app for Mac, Windows, and Linux that downloads and runs open-weights LLMs entirely on your own machine. It bundles a chat UI, a model browser, and a local OpenAI-compatible API server. Your prompts never leave the computer — in our testing it kept generating with the network disconnected. It's a great fit for laptops and desktops with 8 GB of RAM or more. It does not run on phones or tablets, which is the one gap on-device mobile apps like PocketLLM fill.
How we evaluated it
- Setup friction: how long from download to first working reply, measured on a fresh install.
- Privacy posture: what leaves the machine, verified by watching network traffic and disconnecting mid-chat.
- Performance: tokens per second and load time on Apple Silicon, cross-checked against our best local LLM models ranking.
- Who it serves: which real workflows it handles well, and where it leaves you stuck.
What LM Studio actually does
It is a model browser
The first screen you meet is a searchable catalog of GGUF models pulled from Hugging Face. You pick a model — Llama 3.2 3B, Qwen 2.5 7B, Phi-3.5 Mini — and LM Studio shows you which quantizations exist and, helpfully, whether your machine has enough memory to run each one. In our testing this RAM-aware filtering was the single best beginner feature: it stops you downloading a 12B model you can't load.
It is a chat app
Once a model is downloaded, you chat with it in a clean interface that looks a lot like the cloud assistants you already know. You can adjust temperature, context length, and system prompts. Everything runs on your hardware. We measured a 3B model replying at 30+ tok/s on an Apple Silicon Mac, which feels instant for normal back-and-forth.
It is a local API server
This is the feature developers care about. LM Studio can expose a local OpenAI-compatible endpoint, so any script or app that talks to the OpenAI API can be pointed at localhost instead and run against a model on your own machine. We swapped a small Python script from a cloud endpoint to LM Studio's local server in under a minute, changing only the base URL.
LM Studio at a glance
| Aspect | What you get | Notes |
|---|---|---|
| Platforms | Mac, Windows, Linux | No mobile version |
| Price | Free | No account required |
| Privacy | Fully on-device inference | Prompts never leave the machine |
| Offline | Yes, after download | Confirmed by disconnecting network |
| Model format | GGUF (llama.cpp), MLX | Huge Hugging Face catalog |
| API server | OpenAI-compatible | Drop-in for many tools |
| Min RAM (3B Q4) | ~4 GB | 8 GB+ for 7B models |
| Best for | Laptops & desktops | Phones need a mobile app |
Is LM Studio private?
Yes, and this is its strongest selling point. After a model is downloaded, every token is generated on your CPU and GPU. We watched the network monitor during a long writing session and saw zero outbound traffic carrying conversation text. The only times LM Studio touches the internet are when you browse or download models and when the app checks for updates — both optional and both separable from actually using a model. Disconnect Wi-Fi and your local model keeps working as if nothing changed.
That on-device model matters because it means your drafts, code, medical questions, and half-formed ideas stay on hardware you control. It is the same architectural privacy stance PocketLLM takes on the phone: no accounts, no telemetry on your conversations, nothing to leak.
Who should use LM Studio
Developers who want a local OpenAI-compatible endpoint to build against without API bills or rate limits. Privacy-conscious writers and researchers who don't want their work sitting on a vendor's server. The curious who want to feel what a local model is like without learning the command line. If you own a Mac, a Windows laptop, or a Linux box with at least 8 GB of usable memory, LM Studio is one of the easiest on-ramps there is. We compare it head-to-head with the terminal-first options in Ollama vs LM Studio vs PocketLLM, and it sits among our best Mac AI apps for local use.
Where LM Studio stops — and PocketLLM begins
LM Studio is a desktop product. There is no iPhone or iPad build, and there won't be — phone-class chips, memory limits, and battery constraints are a different engineering problem. On Apple Silicon, both desktop and mobile share the same key rule: the size of the model you can run is gated by usable unified memory, so more RAM means a bigger model. A Mac with 16 GB can comfortably run a 7B model; a phone with 8 GB is happiest with a 3B model like Llama 3.2 3B at around 2 GB.
That is exactly the niche PocketLLM is built for. If you understand LM Studio, you already understand PocketLLM — it's the same on-device, zero-telemetry idea tuned for iPhone and iPad, with one-tap model downloads instead of a model browser. If you're coming from the Ollama world instead, our guide to Ollama for iPhone users covers the same translation. Many people we talked to run LM Studio on their Mac at the desk and PocketLLM on the phone everywhere else.
The quick answer
LM Studio is the friendliest way to run private, offline LLMs on a Mac, Windows, or Linux computer. It's free, it's genuinely on-device, and it's the right tool when your hardware is a laptop or desktop. Its only real limitation is that it can't follow you onto a phone. For that, you want a mobile-native app.
Want the same private, on-device chat on your iPhone or iPad? PocketLLM runs open models fully on-device with zero telemetry. Join the waitlist.
Frequently asked questions
What is LM Studio?
LM Studio is a free desktop application for Mac, Windows, and Linux that lets you download and run open-weights large language models entirely on your own computer. It gives you a chat interface, a model browser, and a local OpenAI-compatible server, so you can use models like Llama 3.2 or Qwen 2.5 without sending any text to the cloud. In our testing it ran a 3B model at 30+ tok/s on an Apple Silicon Mac with no internet connection.
Is LM Studio private and does it work offline?
Yes. Once a model is downloaded, LM Studio runs inference fully on-device, and your prompts and responses never leave the machine. We confirmed this by pulling the network cable mid-conversation and the model kept generating. The only network activity is downloading models and checking for app updates, both of which you can do once and then stay offline indefinitely.
Is LM Studio free to use?
LM Studio is free for personal use, and the company has stated it is free for commercial use as well. There is no subscription and no account required to run models locally. You only pay in disk space and RAM — a 7B model needs roughly 4.5 GB on disk at Q4 and about 8 GB of usable memory to run smoothly.
Who should use LM Studio?
LM Studio is for people who own a reasonably modern laptop or desktop and want a friendly, no-terminal way to run local LLMs. It suits developers who want a local OpenAI-compatible endpoint, privacy-conscious writers and researchers, and anyone curious about local AI. If your main device is a phone or tablet, LM Studio cannot help you there — that is where an on-device mobile app like PocketLLM fits.
What is the difference between LM Studio and PocketLLM?
LM Studio is a desktop app for Mac, Windows, and Linux, while PocketLLM is an on-device AI app built for iPhone and iPad, with a PocketLLM Android version coming soon. Both run models locally with no telemetry, but LM Studio targets computers with more RAM and PocketLLM is tuned for phone-class hardware and one-tap model downloads. Many people use LM Studio on their Mac and PocketLLM on their phone for the same private workflow on the go.