2.0 KiB
2.0 KiB
Vectorize configuration routes
| Task | Current documentation |
|---|---|
| Create an index, choose dimensions and metric | Create indexes |
| Bind an index to a Worker, develop, deploy, and verify queries | Introduction to Vectorize |
| Configure bindings and generate types | Binding and TypeScript guidance |
| Create, list, or delete metadata indexes | Metadata filtering and Wrangler commands |
| Manage indexes and vectors through the CLI | Wrangler commands |
| Upload NDJSON and batch ingestion | Insert vectors |
| Check capacity, payload, namespace, or batch constraints | Limits |
Configuration decisions
Confirm the embedding model, output dimensions, and distance metric before provisioning: dimensions and metric cannot be changed in place. Plan a new index and re-embedding where needed when changing models.
Create metadata indexes before ingesting vectors that must be filterable. If adding one to an existing dataset, plan to re-upsert the affected vectors after index creation.
Choose metadata granularity around actual queries. For range filters over high-cardinality fields, consider buckets that preserve the application's required precision; do not bucket identifiers used for exact matches. Fetch the cardinality guidance before designing the schema.