supabase.storage.vector — io.github.supabase-community/storage 0.4.0

Vector buckets, indexes, and vector data against Supabase Storage.

Provides vector bucket CRUD, per-bucket index operations, and per-index vector data operations (put, get, list, query, delete). Index ops take an instance returned by from; vector data ops take an instance further scoped by index.

Example

(require '[supabase.core.client :as client]
         '[supabase.storage.vector :as vector])

(def c (client/make-client "https://abc.supabase.co" "anon-key"))

(vector/create-bucket (vector/from c) "embeddings")

(def b (vector/from c "embeddings"))
(vector/create-index b {:index-name "docs"
                        :data-type :float32
                        :dimension 384
                        :distance-metric :cosine})

(def i (vector/index b "docs"))
(vector/put-vectors i [{:key "doc-1" :data {:float32 [0.1 0.2]}}])
(vector/query-vectors i {:query-vector {:float32 [0.1 0.2]} :top-k 5})

Each function returns {:status :body :headers} on success or an anomaly map on failure. All endpoints are POSTs under {storage-url}/vector with camelCase JSON bodies. See https://supabase.com/docs/reference/javascript/storage-vector-api

create-bucket clj

(create-bucket v name)

Creates a vector bucket named name.

create-index clj

(create-index v params)

Creates a vector index within the instance's bucket.

Params

• :index-name — unique name within the bucket (required)
• :data-type — :float32 (required, the only supported type)
• :dimension — vector dimensionality, e.g. 384 (required)
• :distance-metric — :cosine, :euclidean, or :dotproduct (required)
• :metadata-configuration — optional map with :non-filterable-metadata-keys

delete-bucket clj

(delete-bucket v name)

Deletes the vector bucket name. The bucket must be empty first.

delete-index clj

(delete-index v index-name)

Deletes the index index-name and all its data from the instance's bucket.

delete-vectors clj

(delete-vectors v keys)

Deletes vectors by their keys in batch (1-500 per request).

from clj

(from client)

(from client bucket-name)

Returns a vector storage instance, optionally bound to bucket-name. Bucket CRUD works on either form; index and vector data operations require the bound form.

(def v (from client "embeddings"))
(def i (index v "docs"))

get-bucket clj

(get-bucket v name)

Retrieves metadata for the vector bucket name.

get-index clj

(get-index v index-name)

Retrieves metadata for the index index-name in the instance's bucket.

get-vectors clj

(get-vectors v opts)

Retrieves vectors by their keys in batch.

Options

• :keys — vector keys to fetch (required)
• :return-data — include vector data in the response
• :return-metadata — include metadata in the response

index clj

(index v index-name)

Returns v scoped to index-name. Pass the result as the first argument to vector data operations.

list-buckets clj

(list-buckets v)

(list-buckets v opts)

Lists vector buckets, optionally filtered and paginated.

Options

• :prefix — name prefix filter
• :max-results — page size (default 100)
• :next-token — pagination token from a previous response

list-indexes clj

(list-indexes v)

(list-indexes v opts)

Lists vector indexes in the instance's bucket. Options match list-buckets.

list-vectors clj

(list-vectors v)

(list-vectors v opts)

Lists vectors in the instance's index with pagination.

Options

• :max-results — page size
• :next-token — pagination token from a previous response
• :return-data — include vector data in the response
• :return-metadata — include metadata in the response
• :segment-count — parallel scan segment count (1-16)
• :segment-index — segment to return; with :segment-count given, must be within [0, segment-count)

put-vectors clj

(put-vectors v vectors)

Inserts or updates vectors in batch (1-500 per request).

Each vector is API-shaped and passed through as-is:

{:key "doc-1" :data {:float32 [0.1 0.2 ...]} :metadata {...}}

query-vectors clj

(query-vectors v query)

Queries for similar vectors using approximate nearest neighbor search.

Query

• :query-vector — {:float32 [...]} (required)
• :top-k — number of nearest neighbors to return
• :filter — metadata filter map
• :return-distance — include distances in the response
• :return-metadata — include metadata in the response