Embeddings

Manage vector embedding resources for semantic search. See Embeddings for the full end-user REST API reference.

Read Queries

Operation	GraphQL
List resources	`app_embedding_resources(where: {application_id: {_eq: $appId}}, order_by: {created_at: desc})`
List items	`app_embedding_resource_items(where: {resource_id: {_eq: $resourceId}}, order_by: {created_at: desc}, limit: $limit)`
Available models	`app_available_model_catalog_v1(where: {_or: [{application_id: {_eq: $appId}}, {application_id: {_is_null: true}}]}, order_by: [{provider: asc}, {model: asc}])` (view)

Mutations (Hasura Actions)

Operation	GraphQL
Create resource	`BuilderEmbeddingsCreateResource(input: {key, name, description, model, accessMode, items})` (Hasura Action)
Update resource	`BuilderEmbeddingsUpdateResource(input: {resourceKey, name, description})` (Hasura Action)
Delete resource	`BuilderEmbeddingsDeleteResource(input: {resourceKey})` (Hasura Action)
Upsert items	`BuilderEmbeddingsUpsertItems(input: {resourceKey, items: [{itemId, text, vector, metadata}]})` (Hasura Action)
Delete item	`BuilderEmbeddingsDeleteItem(input: {resourceKey, itemId})` (Hasura Action)
Query	`BuilderEmbeddingsQuery(input: {resourceKey, query, queryVector, topK, minScore, filter})` (Hasura Action)
Import files	`BuilderEmbeddingsImportFiles(input: {resourceKey, fileIds, truncateAtChars})` (Hasura Action)

Key parameters

model (create): Provider-prefixed embedding model. Required for the GraphQL/Builder path. For model-free (direct-vector-only) resources, use the REST API instead (POST /v1/embeddings/resources with dimensions and no model).
accessMode (create): 'application_shared' (default) or 'user_private'. Immutable after creation. Private resources are invisible to non-owners.
vector (upsert items): Pre-computed embedding vector (array of floats). When provided, text is optional and server-side embedding is skipped. Dimensions must match the resource.
queryVector (query): Pre-computed query vector. When provided, query text is optional. At least one of query or queryVector is required.

Note: Model-free resources (created via REST with dimensions instead of model) only accept pre-computed vectors. Text upserts and text queries on model-free resources return a 400 error.