LM Studio Bionic setup and local agent experiments

The presenter installs LM Studio Bionic on Windows 11 with a Ryzen 5600G, 16 GB of RAM and no dedicated GPU. Bionic is described as available for Windows, Mac and Linux, with local or cloud models. The tests use Gemma 4 E4B to run models locally. A simple greeting takes 2 minutes and 8 seconds on this PC; the presenter recommends ordinary LM Studio for basic chat because it is lighter.

The setup shows how to choose a model and associate a project with a folder. Bionic successfully adds a line to an existing text file. Web search requires signing into an LM Studio account in this demonstration. Voice support at the time of testing requires CUDA and an Nvidia GPU, so this machine cannot test it.

The web results need checking. An initial BBC headline request returns stories from late June rather than the September test date. The presenter suspects caching but does not establish the cause. Later requests find current stories, though page ranking remains imperfect, and a stock-price task misses Intel and AMD.

Document creation is less reliable in these experiments. Bionic produces an HTML image gallery, but Word and PDF attempts fail to deliver files, even after an update and claims of success. It does summarize an existing Word document. These results show what this local AI agent managed on the tested hardware and models, rather than establishing a general limit on Bionic.