LM Studio Bionic: Windows setup and local coding demo

LM Studio Bionic adds an AI agent workflow to local model chat, with projects tied to folders on the user's PC. This tutorial covers installation, model selection and a coding assistant that creates files and runs programs. Although the source title mentions macOS, the speaker says Bionic is available only for Windows at the time of recording.

The walkthrough starts with a model identified as Gemma 4 with 12 billion parameters and a 7 GB download. The speaker explains that GPU memory affects which models fit and how quickly they run. CPU execution is possible, but he describes it as very slow. He also points out the interface's indicator for full GPU offloading.

For the first project, Bionic writes a C program to find the first million prime numbers using a sieve. The Windows machine already has GCC and make installed. The demonstration produces source code and an executable, which the speaker runs in a terminal. A second project creates a flat type-length-value encoding specification, then implements it in C with a header and a test program. The generated tests report success.

The speaker says these tasks run locally without uploading project data. He also reports limits on more complex work: a language interpreter project still had bugs and memory leaks with the largest model he could run on an RTX 5090 with 32 GB of VRAM. He attributes those failures to model capability rather than Bionic's agent interface.