LM Studio shipped an AI agent built only for open models, not closed APIs
Bionic inspects codebases, edits files with inline diffs, handles documents, and transcribes voice offline using Mistral's Voxtral model, running entirely on local or open weight models with zero data retention promised.
LM Studio shipped an agent that only runs open models, not closed APIs, with zero data retention promised.
LM Studio released Bionic this week, an AI agent built specifically to run on open weight models rather than closed APIs. It inspects codebases, edits files with inline diffs, works across PDFs and spreadsheets, and transcribes voice offline using Mistral's Voxtral model.
What it actually does
Bionic runs local models downloaded inside the app, connects out over LM Link, or reaches frontier open weight models like GLM 5.2 and Kimi K2.7 Code through LM Studio's own Secure Cloud. Coding support includes agentic code search and inline diff edits, not just chat. Document handling covers PDFs, presentations, and spreadsheets, including creating new ones from scratch.
The data commitment
LM Studio states a zero data retention policy for all Bionic users and says it never trains on user data, with cloud requests processed transiently. Bionic itself is not open source, it sits as a proprietary layer on top of the open models it runs, which is a distinction worth keeping straight: open model, closed agent shell.
Why a build studio cares
AI workflows and custom agents are one of the three things we build. Most agent tooling so far assumes a closed API underneath it. Bionic is a sign that agent orchestration for open weight models specifically, not just chat interfaces for them, is maturing into something usable, which matters for any client build where data residency or model cost rules out a closed API entirely.
Next step: read LM Studio's announcement for the full feature list and supported models. If you are weighing an open model agent stack for a build, write to us at hello@gattyworks.com.