LM Studio Bionic is a new AI agent for coding and document work using local models or open models in the cloud.
It turns LM Studio from a model runner into a complete agent workflow.
LM Studio made running local AI models accessible. You could download a model, chat with it, and expose a local API without building the entire setup yourself.
LM Studio Bionic takes the next step.
You give the agent a project. It can inspect files, make changes, search the web, and produce documents.
Bionic is a separate app
First, notice that Bionic is not a new screen inside LM Studio.
It is a separate app.
LM Studio remains available when you want low-level control over models and inference. Bionic provides the agent interface around that model execution.
This separation makes sense.
Running a model and trusting an agent to edit your files are different jobs. They need different interfaces and different safety controls.
You choose where the model runs
Bionic gives you three ways to run a model:
- locally on your computer
- through LM Link
- through LM Studio Secure Cloud
Local models keep the work on your device. Cloud models give you more compute for harder tasks.
You can choose a smaller local model for private or routine work, then switch to a larger open model when the task needs it.
This also gives you more control over cost. You do not need the largest model for every rename, summary, or simple edit.
LM Studio says its cloud inference uses zero data retention. Requests are processed without being stored after completion, and your data is not used for training.
Bionic for coding
Create a Code project and point Bionic at a local folder.
You can then ask it to:
- investigate a codebase
- explain unfamiliar code
- find relevant files
- edit an implementation
- debug a problem
Bionic shows inline diffs, so you can review the changes as it works.
This is the part I care about.
The model matters, but the workflow around the model matters too. File search, clear diffs, and a good review loop often determine whether an agent feels useful.
An excellent model inside a poor agent interface still creates friction.
Bionic for documents, slides, and spreadsheets
Bionic also has Work projects for general document tasks.
You can give it PDFs, documents, slides, and spreadsheets. It can summarize files, organize a directory, edit existing material, or create something new.
The work runs in a sandboxed environment. This keeps the rest of your computer separate from the files available to the agent.
Automatic checkpoints let you review or roll back changes. In-app previews keep the result beside the conversation.
These controls are not decorative features.
An agent that can edit files needs a clear boundary and an undo button.
Local voice input
Bionic includes a voice keyboard with local transcription.
You can start it from any app and dictate where your cursor is. At launch, it uses Mistral AI's Voxtral model for multilingual real-time transcription.
The audio stays on your device.
I can see this being useful for long prompts, notes, and edits. Speaking an idea is often faster than turning it into a polished request first.
Why Bionic is interesting
Open models are improving quickly, but using them for real work still takes effort.
You need a runtime, model downloads, tool access, file permissions, a review interface, and some way to recover when the agent makes a bad change.
Bionic packages those parts into one app.
This makes open models easier to use as agents, not just chatbots.
My advice is still to review every diff and keep the agent inside a focused project folder. Local execution improves privacy, but it does not make every agent action correct.
The interesting choice is no longer only which model should I use?
Now you can also choose where it runs, what it can access, and how much control you want around its work.