Hermes Agent (+ faster-whisper) is the official Nous Research Hermes Agent extended with faster-whisper for local, GPU-accelerated speech-to-text. This image is built from nousresearch/hermes-agent:latest with faster-whisper and its CUDA runtime wheels (nvidia-cudnn-cu12, nvidia-cublas-cu12) added on top. No separate CUDA toolkit installation is required on the host — GPU acceleration works out of the box with the NVIDIA Unraid plugin. It exposes the Dashboard WebUI on port 9119 and API access on port 8642. A custom Docker network and a shared volume are used so the other Hermes templates can see this container. Please see https://github.com/lbouriez/Unraid-Templates for deployment notes. Refer to https://hermes-agent.nousresearch.com/docs/user-guide/docker for full Hermes configuration options. WHISPER_MODEL controls the model size (tiny / base / small / medium / large-v3). Larger models are slower to load but more accurate. Start with "small" unless you have ample VRAM. If you encounter issues with this template (not the underlying application), please open an issue at https://github.com/lbouriez/Unraid-Templates/issues

This template's WebUI is the Hermes Dashboard. To access the gateway, connect to port 8642 (running in API server mode). Make sure to use the other "Hermes" components from the same maintainer.