vecstore-sdk@0.1.1

Patch Changes

• b19b2b8: Add a Cloudflare Vectorize adapter at vecstore-sdk/vectorize.

  createVectorizeStore takes a Cloudflare client from the cloudflare package, which is now an optional peer dependency, plus your accountId. It drives the Vectorize v2 HTTP API, so an index is a real Vectorize index and createIndex, deleteIndex, and listIndexes all work. The Workers binding is not the client, because a binding reaches one index and has no call that creates, lists, or drops one. A Vectorize filter only matches on a property that carries a metadata index, and a metadata index has to exist before you insert, so metadataIndexes names the properties you filter on and createIndex opens them.

  Namespaces are native, and the adapter scopes every write and every query to them. A Vectorize id is unique across the whole index rather than within a namespace and is capped at 64 bytes, so a namespaced record, or one with an id that is too long, is stored under a deterministic UUID with the original id in the _id metadata key, the way the Qdrant adapter does it.

  compileVectorizeFilter compiles the filter AST to a Vectorize filter object. It is the one compiler that returns a Result, because a Vectorize filter joins every field with AND and has no operator for a missing property. not is pushed to the leaves with De Morgan's laws, and what is left over comes back as an unsupported error with feature: "orFilter" or "existsFilter" before the request is sent. delete({ filter }) and delete({ all: true }) are unsupported too, because Vectorize deletes by id only.

• 12f7509: Add the Supabase adapter. 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, and filters cross as JSON rather than SQL text. The table layout matches the pgvector adapter, so both adapters can read the same table.

• bb85fb9: Add a Redis adapter at vecstore-sdk/redis.

  createRedisStore takes a client with the ft, json, and unlink calls that node-redis gives you, which is now an optional peer dependency, and drives the Redis Query Engine over JSON documents. An index is a real Redis index: createIndex runs FT.CREATE ... ON JSON PREFIX 1 vecstore:{index}:, a record is stored at vecstore:{index}:{namespace}:{id} as { namespace, vector, metadata }, and deleteIndex drops the index and its documents. The server needs the query engine and JSON, which ship with Redis 8, Redis Stack 7.4 or later, and Redis Cloud; without them every verb returns unsupported with feature: "queryEngine".

  Records are JSON rather than hashes, so metadata keeps its types, a string list indexes one tag per element, and a vector reads back at full precision while the index holds a FLOAT32 copy. Ids live in the key rather than in a reserved metadata field, so nothing is hidden from what you read back. Namespaces are a tag on the document and a segment in the key, and the default namespace stores the empty tag, which is why every tag field carries INDEXEMPTY.

  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 createIndex adds one schema entry each. compileRedisFilter takes that same list and returns a Result, so a filter on an undeclared field, or one that compares a number against a tag field, comes back as invalid_argument instead of the empty page Redis would answer with. upsert runs the same check, because Redis leaves a whole document out of the index when one value contradicts the schema. Every metadata field carries INDEXMISSING, which gives exists an operator to compile to.

  A Redis score is the vector distance, where lower is closer for every metric, unlike the higher-is-better score the other adapters return for cosine and dot.

• 12f7509: Add an Upstash Vector adapter at vecstore-sdk/upstash.

  createUpstashStore wraps an Index from @upstash/vector, which is now an optional peer dependency. @upstash/vector cannot create a vector index, so an index name is an Upstash namespace inside the index your URL and token point at: createIndex checks the dimension and metric against info(), deleteIndex deletes the namespace, and listIndexes returns them. The index name uses the one namespace level Upstash gives you, so namespaceMode picks where the namespace option lives. The default, "metadata", emulates it with a _namespace metadata key and a namespace-prefixed stored id, keeping the original id in _id. "native" gives each index and namespace pair its own Upstash namespace named {index}~{namespace}, which leaves ids and metadata untouched and queries unfiltered, at the cost of Upstash's namespace cap.

  compileUpstashFilter compiles the filter AST to the Upstash filter string. Upstash has no NOT, so the compiler pushes negation to the leaves with De Morgan's laws. Because the filter is a string, the compiler rejects a field name that breaks Upstash's identifier rule, a string value holding a backslash or both quote characters, and a non-finite number. The verb returns an invalid_argument error instead.

vecstore-sdk@0.1.0

Minor Changes

• 0ae2949: Initial release: filter AST with Qdrant, pgvector, and Pinecone compilers, the core vector store API, normalized errors, and adapters for all three providers.