What Bionic is, and what it isn't
LM Studio launched Bionic as a new app for Mac and Windows that uses open models for coding, research, and complex work with documents and files. LM Studio's own launch post is dated July 16, 2026. It describes an agentic app that works across code, documents, and other files. Bionic does not replace the LM Studio you may already have. LM Studio stays as the model runner and chat window, while Bionic is the agent that edits files, runs commands and builds documents.
The two apps now run side by side. Since Bionic 1.1.8 they coexist, and Bionic uses LM Studio for local inference while LM Studio is running (LM Studio 0.4.26 or newer is required). The XDA reviewer found that models he had already downloaded in LM Studio showed up in Bionic's model picker with no setup. That fits the shared runtime, but it is one person's experience, not a guarantee for every install.
How the workspace is organised
Bionic works in projects and sessions. LM Studio's documentation says a project keeps related sessions and files together. Separate sessions inside it handle separate tasks. Sessions open as tabs. Switching tabs or projects does not stop a running session, because Bionic keeps working in the background.
There are two main project types:
- Work projects cover documents, PDFs, decks and spreadsheets. LM Studio says Bionic handles these in a sandboxed environment and uses automatic checkpoints so you can review or roll back changes.
- Code projects are for repositories. You enable Allow coding and pick a working folder. LM Studio says Bionic can then search the repo, edit files, use Git and run shell commands in that directory.
The official quick-start tells you to review Bionic's diffs and command output before keeping changes. That advice matters more for local models, which are usually smaller and more error-prone than hosted frontier models.
A simple workflow for a document task
LM Studio's documentation describes this routine:
- Create a new project.
- In a new session, attach files or describe what Bionic should research.
- Ask for a specific result and file format. A typical request is a decision memo saved as a Markdown file.
- Review the response and the generated files.
- Keep revising in the same session, or open a new session for a separate task.
For slide work, LM Studio suggests naming the audience, length, tone and output filename. Vague prompts produce vague files, and this is even more true of 9B models.
The caveat: external files are edited in place
The XDA reviewer flags one detail that deserves attention. LM Studio's documentation confirms it. Files managed by a project live in its workspace. External files stay where they are but remain directly accessible to Bionic. Checkpoints let you review or roll back changes, but that is not the same as a backup. LM Studio's changelog notes that version 1.1.4 added approval prompts for file operations that exceed checkpoint limits, so there are limits to what checkpoints cover.
Sensible habits:
- Start in a managed project, or on copies of anything you care about.
- Read every diff before accepting it.
- Keep real backups of irreplaceable files.
- Treat MCP servers as real permissions. LM Studio's general MCP guidance warns that some servers can run arbitrary code and reach files and the network, so avoid servers from untrusted sources.
Local, remote or cloud
Each session can use any of three execution modes, according to LM Studio's model documentation:
| Mode | Where inference runs | Requirements |
|---|---|---|
| Local | On your PC | Model downloaded and small enough for your hardware |
| Remote (LM Link) | Another device you own | That device must be available and sharing its models |
| Cloud | LM Studio Secure Cloud | Internet, signed-in account, billing, and credits |
For remote models, the conversation stays on the device where you use Bionic. You can change the model on an existing session from its picker, so a local model can handle routine tidying in one tab while a cloud model takes a harder report in another.
Privacy
LM Studio says cloud requests are processed transiently under Zero Data Retention and are not stored afterwards. That is the company's stated policy. Cloud inference still means your request leaves your machine. LM Studio's own pricing FAQ also warns that web search carries privacy and security risks of its own. It advises treating web content as untrusted.
Pricing, as listed now
From LM Studio's current pricing page:
- Free ($0): the Bionic agent, local models, offline voice transcription, LM Link for up to 5 devices, and limited web search.
- Bionic+ ($20 USD/month): US-hosted open models, discounted bulk tokens, LM Link for up to 20 devices, and web search with page extraction.
- Pro ($100 USD/month): five times the usage limits, LM Link for up to 100 devices, included usage for specified hosted models, and early access to new features.
Cloud credits cover usage beyond what your plan includes. Plans and prices can change, so check the pricing page before relying on these numbers.
What has changed since launch
The XDA piece says external API models are only experimental. The changelog shows how quickly Bionic is moving. Version 1.2.1 is dated November 16, 2026. It lists:
- Experimental support for external API models.
- LM Link remote chats and projects on other Bionic devices.
- An "Allow remote agent control" switch.
- Local, Cloud and All filters in the model picker.
- A fix for model loading failures after switching to the ARM64 build on Windows.
- Clearer guidance to install the ARM64 build on Windows ARM PCs.
The in-app terminal reached public builds in version 1.1.10. Version 1.1.8 added in-app previews for Word documents, spreadsheets and presentations. Other coverage says Linux support arrived after launch. One outlet notes that Bionic is closed source, as some Hacker News commenters have pointed out. That matters if you want to audit what an app with file and shell access actually does.
How far can an 8GB card go?
The reviewer reports that Qwen 3.5 9B handles admin tasks fine on his 8GB card, with a ceiling above that. He keeps reasoning mode off to save context. These are individual test results, not benchmarks. LM Studio's guidance is to pick a model that supports the tools and image input your task needs, and to weigh the memory and speed of the machine running it.
Bionic can't be treated as a drop-in replacement for a frontier hosted model on heavy work. The reviewer says as much. He still uses Claude for the harder jobs, and he says MCP and web search were where the free tier got limiting. He also says Notion's MCP requires a Notion sign-in and that Bionic's built-in web search required an LM Studio account. LM Studio's pricing confirms that search is limited on Free and fuller on paid plans. The exact sign-in behaviour is his report.
Bottom line
Bionic's appeal is a project-based agent with diffs, checkpoints and previews, running on models you can keep on your own PC. The trade-offs are real. Small models have limits. External files are edited directly. Cloud use costs credits. MCP connections widen what the agent can touch. Test it on copies first. If it handles your routine document and code chores, you may find you need a hosted agent less often.
References
- LM Studio's new agent app handles the work I used to pay Claude for XDA · 2026-11-19T19:00:22+00:00
- LM Studio Bionic: A Local-First AI Agent for Open Models developersdigest.tech
- LM Studio launches Bionic, a new AI agent app for open models - 9to5Mac 9to5mac.com