Register & sync AI providers¶

The admin wizard's step 5 does this once on a fresh deployment. The two underlying commands:

# On the AI Manager — registers PredictionModel + AIRuntime rows
#   from the trapper-schemas manifest.
python manage.py sync_models_from_schemas

# On the Expert — pulls the AI Manager's catalog + bundled
#   trapper-schemas category YAMLs, creates / updates AIProvider
#   rows, and resolves gbifSpeciesKey -> Species.taxon_id for each
#   species AttributeLabel. Idempotent.
python manage.py sync_ai_models

The full catalog of bundled models lives in trapper-schemas/src/trapper_schemas/data/model_manifest.yaml and the per-model species mappings under trapper-schemas/src/trapper_schemas/data/categories/. To add a new model: edit those files, cut a trapper-schemas release, bump the pinned rev in trapper/, trapper-ai/, and trapper-ai-worker/ pyproject.toml, then re-run the two sync commands.

Most operators won't need this — sync_ai_models covers the canonical case. Reach for the manual flow when wiring up a brand-new model not yet in trapper-schemas, or testing a one-off configuration.

1. Navigate to Media classification → Ai providers, click the + icon, and pick trapper ai provider on the type-selection screen that appears first.

   

2. Set the basic identification fields: Name, Version, Description, Provider type (Detection for MegaDetector-style models, Classification for species classifiers, Depth Estimation or Pose for the other job types), Connection (an AI Provider Connection pointing at your AI Manager), Remote model ID (the PredictionModel.id UUID from the AI Manager), Model file hash (SHA-256 of the weights file).

3. Configure capability flags: Video support, Object based, Crop based (for classifiers operating on pre-cropped detections).

1. Provide the categories JSON (shape documented in the CategoryMapping Pydantic model in trapper-schemas). Minimal detection mapping:

   {
     "attributes": [
       {
         "type": "observationType",
         "labels": [
           { "aiLabel": "1", "value": "animal" },
           { "aiLabel": "2", "value": "human" },
           { "aiLabel": "3", "value": "vehicle" }
         ]
       }
     ]
   }
   

   A classification mapping carries the species join keys:

   {
     "attributes": [
       {
         "type": "species",
         "labels": [
           {
             "aiLabel": "0",
             "scientificName": "Meles meles",
             "commonName": "European badger",
             "gbifSpeciesKey": 5219243
           }
         ]
       }
     ]
   }
   

   On save, sync_ai_models resolves each gbifSpeciesKey to a local Species row and writes the PK as speciesId on each label.

2. Save. To keep manual edits to categories from being overwritten on the next sync, untick Keep synced.