LM Studio: Run LLMs Locally

LM Studio is a platform for working with open language models. Its current homepage highlights Bionic, an agent app for work and code, while its documentation explains that Bionic is a separate application from LM Studio. That distinction matters if you are choosing between low-level model configuration and an agent that works on a project.

Use Bionic for project work

Bionic supports coding, research and document work with open models. Within a project, you can create separate sessions for separate tasks and run sessions across projects. The publisher also describes creating and editing documents, automations and computer control.

The Bionic documentation introduces these capabilities as agent work, while LM Studio remains available for advanced low-level configuration. An existing LM Studio user can therefore assess Bionic as an additional app rather than assuming the familiar application has been replaced.

Choose where the model runs

A Bionic session can use models on your computer, in LM Studio Secure Cloud, or on another device through LM Link. Local model downloads are available within the app, and the publisher identifies llama.cpp and MLX as the underlying local runtimes.

Bionic transcribes speech locally in real time. The homepage states that voice and audio data for this transcription stays on the device and that multiple languages are supported. Cloud inference is a separate option: the pricing page describes US-hosted open models with zero data retention. Local transcription and cloud inference should not be treated as the same processing route.

Compare access options

The Free plan includes running local models, the Bionic agent, offline voice transcription and LM Link for up to five devices, with limited web search. Paid Bionic+ and Pro plans add cloud-model access and different usage allowances.

For a first evaluation, create a project, choose where its model will run and ask for a bounded document or coding task. Inspect the saved changes before relying on them. This is a suggested evaluation approach, not a performance claim or a report that this directory has tested the app.