An AI agent that can read, draft, and organize your files, runs entirely on your own computer, and is free to try, launched last week.
What's happening
LM Studio shipped Bionic on July 16: a standalone app that turns free open AI models into an agent that does work on your files. Not chatting. Doing. Point it at a folder of client notes, contracts, or proposals and it can summarize them, draft a one-pager from your scattered notes, or tell you what changed across a pile of documents. It handles coding too, but the document-and-file angle is what makes this interesting for a small shop.
The whole thing runs on open models on your own machine. You pull in a free local model (Kimi K2 and Qwen3 both work well for document tasks) and Bionic runs it locally. No per-token API costs, and nothing leaves your computer. If you hit a heavy task, you can optionally route to a cloud model, but the default use case is completely offline. LM Studio promises zero data retention on cloud calls too.
There's also a local voice keyboard powered by Mistral's Voxtral model. You dictate, the transcription happens on-device, and not a word touches any external speech-to-text API. For anyone who's been nervous about client documents touching cloud servers, that matters.
Try this this week
- Download Bionic at lmstudio.ai (free, Mac and Windows, currently in preview)
- Inside the app, pull a local model: Qwen3 8B or Kimi K2 are solid starting points for document work
- Create a "Work project" and point it at a folder of your notes, proposals, or past reports
- Ask it to summarize everything in the folder, draft a short client brief, or surface every file that mentions a specific topic
- Try one voice-dictated task: describe what you need, let Bionic turn it into a document
The bigger picture
Six months ago, "local AI" meant a chat window on your laptop. Now it means an agent that can actually touch your files, act on your documents, and grind through the slow work you keep putting off, without an internet connection or a subscription. The question is shifting from "can AI do this?" to "how do I set it up so it just runs?" That is exactly the kind of smallest-working-thing setup build9 helps small teams get running.