Collection

Source: src/AWS/OpenSearchServerless/Collection.ts

An Amazon OpenSearch Serverless collection — a group of OpenSearch indexes that scales OpenSearch Compute Units (OCUs) automatically. A collection of type VECTORSEARCH is the vector store required to back an Amazon Bedrock Knowledge Base.

A collection requires a matching encryption SecurityPolicy to exist before creation, plus a network security policy and a data AccessPolicy to be reachable and usable. Creation is asynchronous — the provider polls (bounded, ~5 minutes) until the collection reaches ACTIVE.

Creating Collections

import * as AWS from "alchemy/AWS";

const encryption = yield* AWS.OpenSearchServerless.SecurityPolicy("Enc", {
  policyName: "kb-enc",
  type: "encryption",
  policy: {
    Rules: [{ ResourceType: "collection", Resource: ["collection/kb"] }],
    AWSOwnedKey: true,
  },
});
const network = yield* AWS.OpenSearchServerless.SecurityPolicy("Net", {
  policyName: "kb-net",
  type: "network",
  policy: [
    {
      Rules: [
        { ResourceType: "collection", Resource: ["collection/kb"] },
        { ResourceType: "dashboard", Resource: ["collection/kb"] },
      ],
      AllowFromPublic: true,
    },
  ],
});
const collection = yield* AWS.OpenSearchServerless.Collection("KB", {
  collectionName: "kb",
  type: "VECTORSEARCH",
});
// collection.collectionEndpoint is the aoss data-plane endpoint

Search Collections

const collection = yield* AWS.OpenSearchServerless.Collection("Search", {
  collectionName: "logs",
  type: "SEARCH",
  description: "application logs",
});