[
{ type: install
  message: <<EOM
The machine learning service listens on 127.0.0.1:3003 by default.
Point the Immich server at it in the admin web interface under
Administration > Settings > Machine Learning.

What to expect:
 * Models are downloaded on first use into %%CACHEDIR%%.  Plan for
   about 1 GB with the default models for smart search, faces and OCR.
 * Inference runs on the CPU.  There is no GPU acceleration on FreeBSD,
   onnxruntime does not support it here.
 * Each worker holds its models in memory, roughly 200 MB for the
   smallest set.  Tune it in %%PREFIX%%/etc/immich-ml.env, for instance
   MACHINE_LEARNING_WORKERS or MACHINE_LEARNING_MODEL_TTL.
EOM
}
]