Create an Upstash Vector index and upsert plus query vectors with metadata filtering

Documented steps

1. Create the index in the Upstash console (Vector > Create Index) or via the Management API: POST https://api.upstash.com/v2/vector/index with HTTP Basic auth (email:management_api_key) and body {"name":...,"region":...,"similarity_function":"COSINE|EUCLIDEAN|DOT_PRODUCT","dimension_count":<int>,"type":"free|payg|fixed"}. The response includes endpoint, token and read_only_token.
2. Choose dimension_count to match your embedding model and similarity_function by use case: COSINE for most NLP/recommendation work, EUCLIDEAN when magnitude matters, DOT_PRODUCT only for unit-normalized vectors.
3. Copy UPSTASH_VECTOR_REST_URL and UPSTASH_VECTOR_REST_TOKEN from the console Connect tab (or the create response's endpoint/token fields).
4. Upsert: POST {UPSTASH_VECTOR_REST_URL}/upsert with header 'Authorization: Bearer $UPSTASH_VECTOR_REST_TOKEN'. Body is one object or an array of objects: {"id":"<string>","vector":[<numbers matching dimension_count>],"metadata":{...optional...}}.
5. Target a non-default namespace with /upsert/{namespace}. Namespaces are created implicitly on first use; omitting the segment targets the default namespace (internally named "").
6. A successful upsert returns {"result":"Success"}. Batch by sending a JSON array in the body.
7. Query: POST {UPSTASH_VECTOR_REST_URL}/query (or /query/{namespace}) with the same Bearer header and body {"vector":[...],"topK":10,"includeMetadata":true,"includeVectors":false,"filter":"<filter string>"}. topK defaults to 10; includeMetadata and includeVectors default to false.
8. Metadata filters are a SQL-like string, e.g. "genre = 'sci-fi' AND year > 2000". Supported operators: =, !=, <, <=, >, >=, GLOB / NOT GLOB, IN / NOT IN, CONTAINS / NOT CONTAINS, HAS FIELD / HAS NOT FIELD. Nested fields use dot notation; arrays index with [0] or [#-1]; AND binds tighter than OR unless parenthesized; booleans are written as 1/0.
9. Query response: {"result":[{"id":...,"score":<number>,"metadata":{...},"vector":[...]}]}. Dense-index scores are normalized 0-1 with 1 most similar.
10. Check ingestion state with GET {UPSTASH_VECTOR_REST_URL}/info — it returns vectorCount (queryable), pendingVectorCount (still indexing), dimension and similarityFunction. Upstash documents that info is updated eventually.
11. Free-tier limits per the Vector FAQ: maximum dimension 1536, and up to 10 indexes per account. Use an array size of up to 1000 for efficient batch inserts; metadata sizing guidance is up to ~48KB per vector.
12. Official docs: https://upstash.com/docs/vector/overall/getstarted | https://upstash.com/docs/vector/api/endpoints/upsert | https://upstash.com/docs/vector/api/endpoints/query | https://upstash.com/docs/vector/features/filtering | https://upstash.com/docs/vector/help/faq