wandermyz/vhotseat is a Python tool that mediates so a local LLM and ComfyUI take turns using GPU resources without interrupting work in progress. The latest release recorded on the GitHub repository is v0.1.0. The provided FACTS list no newly added or changed features for this version, so the change log cannot be confirmed. The core idea the repository presents is handing GPU resources to the next job without cutting off an in-flight answer or render.
The tool schedules the timing of jobs that share GPU memory. If one side's job is in progress, it waits until that job finishes before handing over the resources; if a ComfyUI job is queued, it rejects the LLM load and checks available memory. It also provides a resource headroom check, per-job connection panels, a batch job tool and a status dashboard. Settings can link other graphics jobs, and the README documents commands and hand-off tests using a mock server. The license is MIT.
The repository records one star. The provided material contains no throughput or switching-latency numbers from actual GPUs. The description of the tests is also based on a mock server, so it cannot be treated as evidence of behavior on real devices across various job combinations. What can currently be confirmed is therefore limited to a resource hand-off structure that does not interrupt jobs, plus the mock tests; performance and stability in real environments are hard to judge.