Inference
Embeddings & rerank
Vector and ranking workloads without product data ownership
Octoryn routes transient embeddings and ranking requests. It does not persist vectors, index documents or attach results to product records.
Embeddings
POST /v1/embeddings accepts supported scalar or batch inputs. Select a model whose dimensions, context and region are approved for the workspace.
Rerank
POST /v1/rerank scores supplied documents against a query. The documents protocol field is not persisted by Router; the product controls source authorization and retention.
Moderation
POST /v1/moderations exposes approved moderation routes. A model score informs product policy but does not replace the product's authorization or regulated review process.
Batch deliberately
Bound batch size, request deadline and retry behaviour. A partial provider failure should not silently reorder records or create duplicate writes in the product.
