LM Studio's new Bionic agent runs open models locally or in its secure cloud

LM Studio has launched Bionic, a standalone agent app aimed at people who want to do real work—coding, research, and document tasks—using open-weight models rather than proprietary frontier APIs. The pitch centers on control: users can run models entirely on-device via the LM Studio runtime, connect through LM Link, or offload heavier jobs to LM Studio Secure Cloud, choosing the compute environment per task to manage both privacy and spend. The company commits to zero data retention and says it won’t train on user data; cloud requests are processed transiently and discarded after completion.

For coding, Bionic can point at a local folder, search a codebase agentically, trace behavior, explain unfamiliar code, and propose edits shown as inline diffs, working with open models like GLM 5.2 and Kimi K2.7 Code. A separate ‘Work’ project mode handles documents, PDFs, decks, and spreadsheets inside a sandboxed environment with automatic checkpoints for rollback and native web search for outside context. The app also ships an offline voice keyboard powered by Mistral’s Voxtral transcription model, usable to dictate into any app locally.

The significance is positioning: LM Studio is betting that open-source models will keep closing the gap on coding, reasoning, and long-context tasks, and Bionic is its wager that privacy-conscious users and developers will pay for an agent that keeps that work local or in a no-retention cloud. Bionic is separate from the existing LM Studio app, which remains for low-level configuration; cloud use requires an account for billing.