--model

Why we added this flag: vectorization must use the same model registry as prepare-datasets to ensure metadata alignment and schema validation.

What it does

  • Accepts a Python import path (package.module:ATTRIBUTE) resolving to a Mapping[str, ModelSpec].
  • Drives per-model token policies, metadata enrichment, and schema signature validation during indexing.
  • Required for every idxr vectorize command.

Typical usage

idxr vectorize index \
  --model kb.std.ecc_6_0_ehp_7.registry:MODEL_REGISTRY \
  --partition-manifest workdir/partitions/manifest.json \
  --partition-out-dir workdir/chroma_partitions

Tips

  • Keep a shared environment variable (IDXR_MODEL) so all automation references the same registry string.
  • If you maintain multiple registries, note them in your manifest to avoid mixing incompatible partitions.