LM Studio

Run open-weight models on your own machine

by Element Labs, Inc.

Overview

LM Studio is a desktop application from Element Labs that downloads open-weight language models and runs them on your own machine. It ships llama.cpp and MLX runtimes for local inference, a chat window, document chat that indexes dragged-in files without leaving the app, and a local server that answers on localhost with OpenAI-, Anthropic- and Jev-compatible endpoint shapes; the same stack is reachable from the CLI and a headless daemon. It is installed from the vendor's own site for macOS, Windows and Linux, and plans are bought and managed in the app or the dashboard (lmstudio.ai/docs/app/system-requirements, lmstudio.ai/docs/developer and lmstudio.ai/pricing, read 12 October 2026). Those prices are published in US dollars: Free is $0 and includes local models, Bionic Agent, offline voice transcription, LM Link for up to five devices and limited web search, Bionic+ is $20 USD a month and Pro is $100 USD a month. The pricing page names no region for those figures; the purchase channel it describes is the app and the dashboard. Two hard limits sit next to the free tier: on macOS the supported chips are Apple Silicon (M1/M2/M3/M4) with macOS 14.0 or newer and Intel-based Macs are not supported, and the licence covers personal and internal business use rather than resale.

What stays on the machine is the half that matters for confidential work. The offline page states that LM Studio can operate entirely offline and names three operations that need no connection at all - chatting with downloaded models, chatting with documents, and running a local server - adding that nothing you enter into LM Studio when chatting with LLMs leaves your device and that a dragged-in document stays on your machine because all document processing is done locally. The privacy policy says the same for local models: none of your messages, chat histories and documents are ever transmitted from your system. Requests that do go out to Cloud Services are processed transiently, are not retained after the request completes, and are not used for training or any other purpose, and every party involved works under Zero Data Retention or substantially equivalent terms. Two boundaries follow from those same pages rather than from any test of ours: the cloud models are US-hosted and are bought separately, so the on-device guarantee covers the local half only, and the local server can also be exposed to the local network, so a machine that must send nothing anywhere has to stay bound to localhost.

On the three boundaries this catalogue tracks, the answers are uneven. Commercial use has a source but is narrow: the desktop terms grant a non-exclusive, non-transferable licence to use the software solely for personal and internal business purposes, and separately prohibit sublicensing, service-bureau and software-as-a-service use. Training and data use has a source: local content is not transmitted, and cloud requests are not used for training or any other purpose under Zero Data Retention terms. The current pricing page says cloud credits cover usage beyond the plan’s included allowance. It does not establish endpoint-specific cloud API rates or a shared billing contract between the app’s local API and hosted inference. US hosting does not establish a sales region or final account checkout.

Editor's take

Reach for it when the deciding requirement is that prompts, documents and code never leave the machine: it is the one entry here whose own documentation states the model, prompt construction, document retrieval and local server all run on the device, and the local half costs nothing. Plan around what it is not - it is a model runtime with a chat window rather than a code editor, so editing a repository still needs another tool pointed at its localhost endpoint - and around the Mac requirement, which rules out Intel Macs. Read the cloud tier as a separate purchase that leaves the device, and keep the server bound to localhost if "nothing anywhere" is the actual policy.

Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.

Key Features

•Downloads and runs open-weight LLMs locally with llama.cpp and MLX runtimes
•Free tier at $0 including local models, Bionic Agent and limited web search
•Document chat and RAG that the vendor documents as fully local
•A local server on localhost with OpenAI-, Anthropic- and Jev-compatible endpoints, plus CLI and headless daemon
•Offline operation documented for models, documents and the local server alike
•Optional Bionic cloud models, US-hosted under Zero Data Retention

Use Cases

→Chatting with a private document or codebase without sending it anywhere
→Running a local OpenAI-compatible endpoint for another editor or script
→Working offline on a laptop with downloaded models
→Comparing open-weight models on your own hardware

Pros

✓The vendor documents local chat, document RAG and the local server as needing no connection
✓Free tier covers local models, so the on-device route costs $0
✓Data use is stated as no transmission locally and no training or retention for cloud requests
✓Local API, SDKs, CLI and a headless daemon are documented for scripting

Cons

✕It is a model runtime, not a code editor: the editing and repository side has to come from another tool
✕Intel-based Macs are not supported, and 16GB+ RAM is recommended
✕The licence covers personal and internal business use only, with sublicensing and SaaS use prohibited
✕The cloud half is US-hosted and bought separately, and the local server can be exposed to the local network if you configure it that way

Frequently Asked Questions

Is LM Studio free?

Yes. The Free plan is $0 and includes local model support, Bionic Agent, offline voice transcription, LM Link for up to five devices and limited web search, and the vendor confirms both that Bionic can be used for free and that local models keep working. Bionic+ at $20 USD a month and Pro at $100 USD a month add cloud usage; the pricing page names no region for those prices. The public pricing page was re-read on15October2026; US hosting is not a sales-region guarantee, and cloud credits cover use beyond the included allowance without an endpoint-specific API contract.

Do my prompts and documents leave my machine?

For the local half, the vendor says no: the offline page states that LM Studio can operate entirely offline, that nothing you enter when chatting with LLMs leaves your device, and that a document you drag in is processed locally. The cloud half is different - cloud models and web search send requests to US-hosted infrastructure, where the privacy policy says they are processed transiently, not retained, and not used for training or any other purpose under Zero Data Retention terms. The local server can also be published to the local network, so use localhost if nothing may leave the machine.

Can I use it for work and for commercial projects?

The terms grant a licence for personal and internal business purposes, so using it inside your own organisation is covered. They also prohibit sublicensing, service-bureau, application-service-provider and software-as-a-service use, so building a product that resells the software itself is not the documented route.

Is it an AI code editor?

No. It runs models and serves them locally; the chat window and the local API are the surfaces. Editing an existing repository is done by whatever editor or agent you point at http://localhost, which is the half the vendor does not supply. The API also expects a token in its example requests, so check the local server setup before wiring anything to it.

Task guides for this product

T7 · Check whether code and prompts stay on your Mac

Key facts and where they come from

PriceDocumented: LM Studio’s public Bionic pricing page lists Bionic+ at USD 20 per month and Pro at USD 100 per month, alongside a free local-model route. The page says cloud inference is US-hosted; this hosting statement is not a sales-region guarantee or a final tax-inclusive checkout.Source: LM Studio - pricing
AvailabilityDocumented: After model files are downloaded, local chats, document RAG and inference can run offline; document processing stays local and the server can serve localhost or the local network. On Mac, LM Studio requires Apple Silicon and macOS 14 or newer; Intel Macs are unsupported. At least 16 GB RAM is recommended, while 8 GB may work with small models and modest context.Source: LM Studio - offline operation · LM Studio - system requirements
Commercial useDocumented: Version August 23, 2026: 'Company grants to You a non-exclusive, non-transferable license to use the Software solely for Your personal and / or internal business purposes and solely in accordance with the Documentation.' The restrictions list prohibits sublicensing, service-bureau, application-service-provider and software-as-a-service use. The page contains no training clause.Source: LM Studio - desktop app terms of service
Training on your dataDocumented: Effective June 2026: 'LM Studio can run entirely on your device. If you download and run models locally, none of your messages, chat histories, and documents are ever transmitted from your system - everything is saved locally on your device.' Cloud requests 'are processed transiently and not stored after the request completes', are 'not used for training or any other purpose', and 'All parties involved in processing your data operate under Zero Data Retention or substantially equivalent terms.'Source: LM Studio - desktop app privacy policy
API billingNot established: The page says cloud credits cover usage beyond the allowance included with a plan. It does not establish endpoint-specific cloud API rates or that a local API endpoint and hosted inference share a billing contract.Source: LM Studio - pricing
Export formatsNot established: LM Studio's official chat guide establishes local JSON conversation storage accessible through Finder/File Explorer, and warns against relying on its structure. It does not establish a portable downloadable conversation/report export format contract; the attempted dedicated export URL was unreadable.Source: LM Studio - offline operation · LM Studio - developer docs, local APIs and SDKs · Official lm-studio reference
Usage limitsNot established: The page lists LM Link for up to five devices on Free and five-times usage limits on Pro. It does not specify an absolute cloud/search quota or maximum document capacity.Source: LM Studio - pricing

Official sources

The vendor pages we read for the claims on this page, what each one states, and the day we read it.

LM Studio - pricing ↗
LM Studio’s public Bionic pricing page lists Bionic+ at USD 20 per month and Pro at USD 100 per month, alongside a free local-model route. The page says cloud inference is US-hosted; this hosting statement is not a sales-region guarantee or a final tax-inclusive checkout. The page says cloud credits cover usage beyond the allowance included with a plan. It does not establish endpoint-specific cloud API rates or that a local API endpoint and hosted inference share a billing contract. The page lists LM Link for up to five devices on Free and five-times usage limits on Pro. It does not specify an absolute cloud/search quota or maximum document capacity.
Read on October 15, 2026
LM Studio - offline operation ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: 'LM Studio can operate entirely offline, just make sure to get some model files first.' The three operations listed as needing no connectivity are chatting with downloaded LLMs, chatting with documents, and running a local server: 'Nothing you enter into LM Studio when chatting with LLMs leaves your device.' / 'All document processing is done locally, and nothing you upload into LM Studio leaves the application.' / 'LM Studio can be used as a server to provide LLM inferencing on localhost or the local network.' Official body read on 2026-10-15, limited to this scope: After model files are downloaded, LM Studio can run chats, document RAG and inference offline. Document processing stays local; the server can serve localhost or the local network. This does not document every external repository-editing agent. After model files are downloaded, local chats, document RAG and inference can run offline; document processing stays local and the server can serve localhost or the local network.
Read on October 15, 2026
LM Studio - desktop app privacy policy ↗
Effective June 2026: 'LM Studio can run entirely on your device. If you download and run models locally, none of your messages, chat histories, and documents are ever transmitted from your system - everything is saved locally on your device.' Cloud requests 'are processed transiently and not stored after the request completes', are 'not used for training or any other purpose', and 'All parties involved in processing your data operate under Zero Data Retention or substantially equivalent terms.'
Read on October 12, 2026
LM Studio - desktop app terms of service ↗
Version August 23, 2026: 'Company grants to You a non-exclusive, non-transferable license to use the Software solely for Your personal and / or internal business purposes and solely in accordance with the Documentation.' The restrictions list prohibits sublicensing, service-bureau, application-service-provider and software-as-a-service use. The page contains no training clause.
Read on October 12, 2026
LM Studio - system requirements ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: 'macOS Chip: Apple Silicon (M1/M2/M3/M4). macOS 14.0 or newer is required. 16GB+ RAM recommended. … Intel-based Macs are currently not supported.' Windows (x64/ARM) and Linux (x64/ARM64) are also supported. Official body read on 2026-10-15, limited to this scope: macOS needs Apple Silicon and macOS14 or newer; 16GB+RAM is recommended, 8GB may work with small models/modest context, and Intel Macs are unsupported. On Mac, LM Studio requires Apple Silicon and macOS 14 or newer; Intel Macs are unsupported. At least 16 GB RAM is recommended, while 8 GB may work with small models and modest context.
Read on October 15, 2026
LM Studio - developer docs, local APIs and SDKs ↗
'Build with LM Studio's local APIs and SDKs' - OpenAI-, Anthropic- and Jev-compatible endpoints, TypeScript and Python SDKs, a CLI and a headless daemon. The example request runs 'lms server start --port 1234' and then calls http://localhost:1234/api/v1/chat with an Authorization bearer token. The page does not state what a subscription covers for cloud inference.
Read on October 12, 2026
Official lm-studio reference ↗
The official chat guide says conversations are locally stored as JSON and can be located with Reveal in Finder or Show in File Explorer. It warns against editing or depending on their structure. This is a storage-format statement, not a guaranteed portable report-export contract.
Read on October 15, 2026

Compare with other tools in Code Tools

Product	Maker	Pricing	Launched	Shared tags
Ollama	Ollama Inc.	Freemium	Launch date not verified	On-device, Open weights
Claude Code	Anthropic	Paid	2025-02-24	AI agent
Devin Desktop	Cognition	Freemium	2024-11-13	AI agent
Replit Agent	Replit	Freemium	2024-09	AI agent
MonkeyCode	Chaitin Tech	Freemium	2025-06	—
Lovable	Lovable	Freemium	2024-12	—