#3243: Feature Request: Add GPU hardware acceleration libraries (for video editing scenario) to the Docker image

Feature Request: Add GPU hardware acceleration libraries to the Docker image

Many users run Hermes WebUI in containers on machines with dedicated or integrated GPU hardware (Intel, AMD, NVIDIA). Hardware acceleration enables significant performance gains for video transcoding, AI inference, and other GPU-bound workloads commonly performed inside the agent context (e.g. PyAV decoding, Whisper inference, LLM serving).

Request

Add the following GPU driver packages to the Dockerfile. These add minimal footprint but unlock hardware acceleration for all three major GPU vendors:

# Intel GPU (Arc, Iris Xe, UHD Graphics) — VA-API + Quick Sync Video
RUN apt-get update && apt-get install -y \
    intel-media-va-driver-non-free \
    libmfx1 \
    libva2 \
    vainfo \
    && rm -rf /var/lib/apt/lists/*

# AMD GPU (Radeon, Instinct) — VA-API
RUN apt-get update && apt-get install -y \
    mesa-va-drivers \
    libva2 \
    vainfo \
    && rm -rf /var/lib/apt/lists/*

# NVIDIA GPU (GeForce, Quadro, A/H/L-series) — CUDA + NVENC/NVDEC
RUN apt-get update && apt-get install -y \
    nvidia-container-toolkit \
    && rm -rf /var/lib/apt/lists/*

This allows users passing GPU devices via docker-compose/podman-compose to use hardware-accelerated video processing and AI inference without building custom images.

Context

Users passing GPU devices via compose need only the user-space libraries inside the container — the host kernel drivers are shared via the passed device nodes. Example compose addition:

# Intel
hermes-webui:
  devices:
    - /dev/dri/renderD128:/dev/dri/renderD128
    - /dev/dri/card1:/dev/dri/card1
  group_add:
    - video
    - render

# AMD
hermes-webui:
  devices:
    - /dev/dri/renderD128:/dev/dri/renderD128
  group_add:
    - video

# NVIDIA
hermes-webui:
  devices:
    - /dev/nvidia0:/dev/nvidia0
    - /dev/nvidiactl:/dev/nvidiactl
    - /dev/nvidia-uvm:/dev/nvidia-uvm
  runtime: nvidia

With these packages pre-installed, common Python libraries (PyAV, faster-whisper, torch, opencv) can transparently use hardware decode/inference without users needing to build custom images.

Severity

Low — quality-of-life improvement for GPU users. No impact on CPU-only setups. The packages are small (~10–50 MB total) and add zero runtime overhead when no GPU is present.