EMMA vs LM Studio
A workbench for running models, or a companion that knows you.
LM Studio is one of the best ways to download and run local models, and it's free. EMMA is a different kind of thing: not a place to try models, but one AI that lives on your computer, knows you, and does things for you. They also work together: EMMA can think with LM Studio's local server.
Side by side
| EMMA | LM Studio | |
|---|---|---|
| What it is | A private AI companion that lives on your computer: one ongoing relationship rather than a list of chats. | A desktop app to discover, download and run local models, chat with them, chat with documents, and serve them to other apps. |
| Price | £149 once, VAT included. A 60-day free trial with everything unlocked first; when it ends nothing is deleted. | Free, for home and for work. |
| Source code | Closed. The privacy is something you test (unplug her; read her audit log), not something you read the code for. | Closed (its terms describe the source as a trade secret). Its SDKs are published separately. |
| Runs on | Windows, Linux (.deb / AppImage) and macOS (Apple Silicon and Intel). The Mac build is behind the other two right now: it's on 1.11.6, and the current version follows soon. Thinks with Ollama, or llama.cpp, LM Studio, vLLM or anything that speaks the OpenAI API. | Apple Silicon Macs, Windows (x64 and ARM64) and x64 Linux, plus a headless mode. |
| Where your conversations live | On your machine, encrypted at rest (AES-256), key in your OS keyring. No account, no server. | On your machine when you run local models. |
| Remembers you between chats | Yes, by design: facts about you, your history, people in your life and your goals, carried across years and across model swaps. You can see every fact, and forget any of it. | Its docs describe chats, documents, MCP servers and an API server, not a memory of you that carries between chats. |
| Works on your computer for you | Yes: files, apps, commands, the browser, the screen. Gated by a safety dial (full / ask first / locked) and folders she may never touch. | Through MCP servers you connect; its newer Bionic agent advertises coding, automations and computer control. |
| Seeing and undoing what it did | Per-action undo, a panic stop, and a tamper-evident log of every action she has ever taken. | Not described in its docs. |
| Voice | She speaks, and she hears you, both on your machine. | Bionic advertises local, real-time voice transcription. |
| Also | Dream Mode (she goes over your day overnight), Her Studio (record video and podcasts), your phone on the home Wi-Fi, "Is this a scam?", offline Wikipedia, English / German / Dutch. | An OpenAI-compatible local API server; MLX and llama.cpp runtimes; a very wide model catalogue. |
Where LM Studio is the better choice
- It's free, for home and work. EMMA costs £149 after the trial.
- Model choice and speed tuning. If you want to try every new model the day it lands, compare quantisations and tune runtimes, LM Studio is built for exactly that.
- Serving models to other software is a first-class job for it.
Where EMMA is different
- She's one relationship, not a list of chats. EMMA builds a memory of you, the people in your life and what you're working on, and it outlives any model you swap in.
- She acts, and you can check her. A safety dial, undo, a panic stop and an audit log of every action.
- She keeps going when you stop. Dream Mode, a morning briefing, reminders and standing orders.
Or use both
Already happy with LM Studio? Keep it. Start LM Studio's local server and point EMMA at it: LM Studio does the model work, EMMA does the remembering and the acting.
Where the LM Studio column comes from
- LM Studio app docs
- LM Studio is free for use at work
- LM Studio app terms
- LM Studio home page (Bionic)
Try her yourself
Sixty days, everything unlocked, no card, no account. When the trial ends nothing is deleted.