Supabase, Upstash Vector, Cloudflare Vectorize, and Redis adapters

VecStore SDK 0.1.1 adds four adapters. The same filter you wrote for Qdrant now compiles for Supabase, Upstash Vector, Cloudflare Vectorize, and Redis.

Supabase

createSupabaseStore from vecstore-sdk/supabase reaches Supabase Vector through supabase-js, so vector search runs in Edge Functions and the browser. PostgREST has no syntax for order by embedding <=> $1, so every verb calls a SQL function from the new sql/supabase.sql, which you install once per project. The functions are security invoker, so row level security still applies. The table layout matches the pgvector adapter, so both adapters can read the same table.

Upstash Vector

createUpstashStore from vecstore-sdk/upstash wraps an Index from @upstash/vector. Upstash cannot create an index from the client, so an index name maps to an Upstash namespace inside the index your URL and token point at. namespaceMode picks where the namespace option lives: "metadata" emulates it with a _namespace key, and "native" gives each index and namespace pair its own Upstash namespace. Upstash has no NOT, so compileUpstashFilter pushes negation to the leaves with De Morgan’s laws.

Cloudflare Vectorize

createVectorizeStore from vecstore-sdk/vectorize takes a Cloudflare client plus your accountId and drives the Vectorize v2 HTTP API, so createIndex, deleteIndex, and listIndexes all work. A Vectorize filter only matches on a property that carries a metadata index, so metadataIndexes names the properties you filter on and createIndex opens them. Namespaces are native. A Vectorize filter joins every field with AND, so or and exists come back as an unsupported error before the request is sent.

Redis

createRedisStore from vecstore-sdk/redis takes a node-redis client and drives the Redis Query Engine over JSON documents, which ship with Redis 8, Redis Stack 7.4 or later, and Redis Cloud. Records are JSON rather than hashes, so metadata keeps its types. Redis searches a metadata field only when the index schema declares it, so metadataFields names the fields you filter on as tag or numeric, and a filter on an undeclared field returns invalid_argument instead of an empty page.

Each provider page in the docs lists what its adapter does differently. Start with the provider differences overview.