Wednesday was a shipping day. I took Amanuensis from "runs somewhere" to "runs fast on your own laptop" and pushed the whole thing live.
The headline was Intel iGPU acceleration. I added an iGPU-optimized WhisperX build with speaker diarization, which touched forty-five files. That's the GPU path, the install scripts, the Docker compose files, the benchmark script, the model guide, the bare-metal installer, the GPU selector, the start script. The big diff was eleven thousand lines added against seven hundred fifty-four removed. The server routing got restructured to match the Unicorn Orator pattern so everything wires together cleanly. Whisper Large v3 is now the default model and the GUI explicitly says all models are iGPU-optimized.
I didn't want this to disappear into a pull request and evaporate. I rewrote the README into a real pitch for hardware optimization and added a dedicated setup guide. Two hundred fifty lines of docs telling people exactly how to get acceleration working. I also dropped the Unicorn logo into the web interface, replacing the stock header with the real brand mark across the templates and styles.
I added an iGPU-optimized WhisperX build with speaker diarization, which touched forty-five files.
The workflow for someone who wants speed now is straightforward. They run the iGPU install script, pick their GPU if they have more than one, and start the server. The Docker path is there too if they prefer containers. The benchmark script lets them verify the numbers. The old AMD NPU and OpenVINO guides stayed in place for the people who went that route.
I uploaded a screenshot of the updated interface so the visual change is visible without cloning the repo.
Amanuensis is a transcription tool and the bottleneck has always been the model. Putting Whisper Large v3 on an Intel iGPU changes the calculus for people who don't have a data center sitting next to their desk. Eighteen thousand lines of changes across docs, scripts, templates, and the core server. The feature ships.