+
+ LiteLLM searches a k-NN index you already have in an OpenSearch Serverless collection or an
+ OpenSearch Service domain. It does not create the index or upload documents. Before creating this
+ vector store, make sure:
+
+
+ -
+ The index has a knn_vector field holding each document's embedding. Enter the index name as
+ the Vector Store ID, or under Index Name when another endpoint already uses that ID
+
+ -
+ You know which embedding model created those embeddings. That model must be added to this proxy
+ under Models so you can pick it below. Using a different model returns wrong results
+
+ -
+ The proxy has AWS credentials that can read the index, either its own (environment variables or
+ an IAM role) or a saved credential picked under Existing Credentials. A Serverless collection
+ needs a data access policy for them, and a domain needs its access policy to allow them
+
+
+
+