LM Studio has unveiled Bionic, a separate AI agent app that uses open models. Developers and document workers can choose on-device or cloud models for coding, research, and file tasks. Voice input is processed on the device, while cloud use requires a separate account and billing setup.
Local voice input and open models in one app
Bionic is separate from the existing LM Studio app. The company says users who need LM Studio's low-level settings, which adjust detailed system behavior, can use both apps together. In Bionic, users connect a project, select a model, and assign tasks to an agent.
The company has promised Bionic users zero data retention and no training on their data. Models can run directly on the device, connect through LM Link, or use LM Studio Secure Cloud. The product aims to give users more control over costs by letting them choose models and runtimes suited to each task.
On-device voice transcription wherever the cursor is
The voice keyboard can be activated in any app and starts dictation at the cursor's location. Transcription converts speech into text using an on-device audio model, without sending content to the cloud for voice processing. At launch, it offers Mistral AI's Voxtral, which the company describes as a multilingual real-time transcription model.
Review changes to code and documents
Coding starts in a project linked to a local folder. Users can request research, editing, and debugging, monitor progress, and review changes through line-by-line diffs comparing lines before and after edits. The agent can search code to find relevant files, trace behavior, or explain unfamiliar code. The company's examples of open models include GLM 5.2 and Kimi K2.7 Code.
Document work runs in separate projects with isolated environments, keeping it apart from other tasks. Bionic supports organizing local directories, editing files, summarizing material, and searching the web. Automatic checkpoints, records of the state at the time of changes, let users review or undo changes. The company also presents it as a tool for handling documents, PDFs, presentations, and spreadsheets, as well as creating new materials. It says support for additional file formats in the in-app preview is expanding.
Control costs and task routing through model selection
Running models on the device uses the LM Studio runtime, the components needed to run models, for everything from simple chats to agent tasks. For heavier tasks, users can choose cloud-hosted open models. The company says requests are used only during processing and are not retained afterward. Cloud models require an LM Studio account and billing setup.
The benefit of this setup is the ability to control data processing routes and spending by choosing local or cloud execution for each task. However, users must decide which model and runtime are suitable, and cloud use requires account and billing setup. The original announcement provides no pricing or hardware requirements, so users need to check their own setup and cost conditions before choosing.
Bionic goes beyond simply running open models: it is an agent app that lets users choose where models run for coding and document work.