Bionic runs open models locally, from code to spreadsheets

Bionic runs open models locally to code and analyze documents. A solution by LM Studio with zero data retention.

Separate from the classic LM Studio application, Bionic is an agent built to make open models work on concrete tasks: code, research, and document and file manipulation. Execution remains flexible. Models run locally via the LM Studio runtime, connect through LM Link, or switch to the largest open models hosted in the LM Studio Secure Cloud for heavy workloads. The publisher boasts zero data retention (ZDR) by default and commits to not training its models on user data.

On the coding side, the agent inspects a local folder, explains unfamiliar code, proposes modifications presented in inline diffs, and finds relevant files through agentic code search. LM Studio cites GLM 5.2 and Kimi K2.7 Code among the compatible models. For office work, Bionic processes documents, PDFs, presentations, and spreadsheets in a sandboxed environment that isolates the rest of the machine, featuring native web search, automatic checkpoints to roll back changes, and built-in previews. It can review and annotate contracts or fill out forms.

Voice input is handled entirely on-device. A voice keyboard can be triggered from any application and transcribes wherever the cursor is located, without sending any audio to the cloud. At launch, LM Studio embeds Voxtral, Mistral AI's multilingual transcription model, running via the ExecuTorch runtime. Local models are downloaded directly within the application; cloud usage requires an LM Studio account for billing. As a detail reported by the publisher, part of Bionic's development was conducted using Bionic itself.