@langchain/core 1.2.9 → 1.2.11

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # @langchain/core
 
+## 1.2.11
+
+### Patch Changes
+
+- [#11603](https://github.com/langchain-ai/langchainjs/pull/11603) [`fec9cd8`](https://github.com/langchain-ai/langchainjs/commit/fec9cd87b01976014dd549bd2cf7849aee89a566) Thanks [@thushanth-bengre-langchain](https://github.com/thushanth-bengre-langchain)! - fix(core): build streaming `llmOutput.tokenUsage` from the fully-accumulated chunk instead of whichever individual chunk's `usage_metadata` arrived last
+
+  Affects both core streaming paths — `.stream()`/`.streamEvents()` (`_streamIterator`) and `.invoke()`/`.generate()` when a streaming-preferring callback is attached (`_generateWithCache`'s `hasStreamingHandler` branch). Previously, `llmOutput.tokenUsage` was overwritten by each chunk in turn, so only the last chunk carrying `usage_metadata` won — correct for providers that emit one cumulative total on a final chunk, but wrong for providers (e.g. `@langchain/google`, `@langchain/anthropic`) that emit `usage_metadata` as a per-chunk delta across multiple chunks, where the values must be summed.
+
+  Note for provider authors: this assumes each streamed chunk's `usage_metadata` is either a per-chunk delta or appears only on a single final chunk. A provider that instead repeats a cumulative total on every chunk will now see it summed (and inflated) in `llmOutput.tokenUsage`, matching the existing behavior of the correctly-working `message.usage_metadata` field.
+
+  Also fixes `@langchain/google`'s `invoke({streaming: true})` path (no streaming-preferring callback attached), where `llmOutput` was never populated at all.
+
+- [#11590](https://github.com/langchain-ai/langchainjs/pull/11590) [`ffebdc2`](https://github.com/langchain-ai/langchainjs/commit/ffebdc2f00f3290d19f85e5afd6a297920ae584c) Thanks [@thushanth-bengre-langchain](https://github.com/thushanth-bengre-langchain)! - Fix OpenAI Responses API replay under Zero Data Retention when a response contains more than one reasoning item, for both v0 and v1. In v0, the default replay path now reuses `response_metadata.output` directly, preserving every reasoning item's `id`/`encrypted_content` in original order. In v1, `AIMessage.contentBlocks` (`outputVersion: "v1"`) is fixed the same way. `additional_kwargs.reasoning` is unchanged.
+
+## 1.2.10
+
+### Patch Changes
+
+- [#10047](https://github.com/langchain-ai/langchainjs/pull/10047) [`ff1248f`](https://github.com/langchain-ai/langchainjs/commit/ff1248fd49dacdb35f5da128278cd777068481d7) Thanks [@afirstenberg](https://github.com/afirstenberg)! - fix(core): BaseMessage.text use contentBlocks
+
 ## 1.2.9
 
 ### Patch Changes

package/dist/_virtual/_rolldown/runtime.cjs

@@ -24,7 +24,7 @@ var __copyProps = (to, from, except, desc) => {
 	}
 	return to;
 };
-var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", {
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(isNodeMode || !mod || !mod.__esModule || !__hasOwnProp.call(mod, "default") ? __defProp(target, "default", {
 	value: mod,
 	enumerable: true
 }) : target, mod));

package/dist/callbacks/dispatch/index.cjs

@@ -40,7 +40,8 @@ require_index.AsyncLocalStorageProviderSingleton.initializeGlobalInstance(new no
 * ```
 */
 async function dispatchCustomEvent(eventName, payload, config) {
-	await require_callbacks_dispatch_web.dispatchCustomEvent(eventName, payload, require_config.ensureConfig(config));
+	const ensuredConfig = require_config.ensureConfig(config);
+	await require_callbacks_dispatch_web.dispatchCustomEvent(eventName, payload, ensuredConfig);
 }
 //#endregion
 exports.dispatchCustomEvent = dispatchCustomEvent;

package/dist/callbacks/dispatch/index.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.cjs","names":["AsyncLocalStorageProviderSingleton","AsyncLocalStorage","dispatchCustomEventWeb","ensureConfig"],"sources":["../../../src/callbacks/dispatch/index.ts"],"sourcesContent":["/* __LC_ALLOW_ENTRYPOINT_SIDE_EFFECTS__ */\n\nimport { AsyncLocalStorage } from \"node:async_hooks\";\nimport { dispatchCustomEvent as dispatchCustomEventWeb } from \"./web.js\";\nimport { type RunnableConfig, ensureConfig } from \"../../runnables/config.js\";\nimport { AsyncLocalStorageProviderSingleton } from \"../../singletons/index.js\";\n\nAsyncLocalStorageProviderSingleton.initializeGlobalInstance(\n  new AsyncLocalStorage()\n);\n\n/**\n * Dispatch a custom event.\n *\n * Note: this method is only supported in non-web environments\n * due to usage of async_hooks to infer config.\n *\n * If you are using this method in the browser, please import and use\n * from \"@langchain/core/callbacks/dispatch/web\".\n *\n * @param name The name of the custom event.\n * @param payload The data for the custom event.\n *   Ideally should be JSON serializable to avoid serialization issues downstream, but not enforced.\n * @param config Optional config object.\n *\n * @example\n * ```typescript\n * import { dispatchCustomEvent } from \"@langchain/core/callbacks/dispatch\";\n *\n * const foo = RunnableLambda.from(async (input: string) => {\n *   await dispatchCustomEvent(\"my_custom_event\", { arbitraryField: \"someval\" });\n *   return input;\n * });\n *\n * const callbacks = [{\n *   handleCustomEvent: (eventName: string, payload: any) => {\n *     // Logs \"my_custom_event\" and { arbitraryField: \"someval\" }\n *     console.log(eventName, payload);\n *   }\n * }];\n *\n * await foo.invoke(\"hi\", { callbacks })\n * ```\n */\nexport async function dispatchCustomEvent(\n  eventName: string,\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  payload: any,\n  config?: RunnableConfig\n) {\n  const ensuredConfig = ensureConfig(config);\n  await dispatchCustomEventWeb(eventName, payload, ensuredConfig);\n}\n"],"mappings":";;;;;;;AAOAA,cAAAA,mCAAmC,yBACjC,IAAIC,iBAAAA,kBAAkB,CACxB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmCA,eAAsB,oBACpB,WAEA,SACA,QACA;CAEA,MAAMC,+BAAAA,oBAAuB,WAAW,SADlBC,eAAAA,aAAa,MAC0B,CAAC;AAChE"}
+{"version":3,"file":"index.cjs","names":["AsyncLocalStorageProviderSingleton","AsyncLocalStorage","ensureConfig","dispatchCustomEventWeb"],"sources":["../../../src/callbacks/dispatch/index.ts"],"sourcesContent":["/* __LC_ALLOW_ENTRYPOINT_SIDE_EFFECTS__ */\n\nimport { AsyncLocalStorage } from \"node:async_hooks\";\nimport { dispatchCustomEvent as dispatchCustomEventWeb } from \"./web.js\";\nimport { type RunnableConfig, ensureConfig } from \"../../runnables/config.js\";\nimport { AsyncLocalStorageProviderSingleton } from \"../../singletons/index.js\";\n\nAsyncLocalStorageProviderSingleton.initializeGlobalInstance(\n  new AsyncLocalStorage()\n);\n\n/**\n * Dispatch a custom event.\n *\n * Note: this method is only supported in non-web environments\n * due to usage of async_hooks to infer config.\n *\n * If you are using this method in the browser, please import and use\n * from \"@langchain/core/callbacks/dispatch/web\".\n *\n * @param name The name of the custom event.\n * @param payload The data for the custom event.\n *   Ideally should be JSON serializable to avoid serialization issues downstream, but not enforced.\n * @param config Optional config object.\n *\n * @example\n * ```typescript\n * import { dispatchCustomEvent } from \"@langchain/core/callbacks/dispatch\";\n *\n * const foo = RunnableLambda.from(async (input: string) => {\n *   await dispatchCustomEvent(\"my_custom_event\", { arbitraryField: \"someval\" });\n *   return input;\n * });\n *\n * const callbacks = [{\n *   handleCustomEvent: (eventName: string, payload: any) => {\n *     // Logs \"my_custom_event\" and { arbitraryField: \"someval\" }\n *     console.log(eventName, payload);\n *   }\n * }];\n *\n * await foo.invoke(\"hi\", { callbacks })\n * ```\n */\nexport async function dispatchCustomEvent(\n  eventName: string,\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  payload: any,\n  config?: RunnableConfig\n) {\n  const ensuredConfig = ensureConfig(config);\n  await dispatchCustomEventWeb(eventName, payload, ensuredConfig);\n}\n"],"mappings":";;;;;;;AAOAA,cAAAA,mCAAmC,yBACjC,IAAIC,iBAAAA,kBAAkB,CACxB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmCA,eAAsB,oBACpB,WAEA,SACA,QACA;CACA,MAAM,gBAAgBC,eAAAA,aAAa,MAAM;CACzC,MAAMC,+BAAAA,oBAAuB,WAAW,SAAS,aAAa;AAChE"}

package/dist/callbacks/dispatch/index.js

@@ -39,7 +39,8 @@ AsyncLocalStorageProviderSingleton.initializeGlobalInstance(new AsyncLocalStorag
 * ```
 */
 async function dispatchCustomEvent(eventName, payload, config) {
-	await dispatchCustomEvent$1(eventName, payload, ensureConfig(config));
+	const ensuredConfig = ensureConfig(config);
+	await dispatchCustomEvent$1(eventName, payload, ensuredConfig);
 }
 //#endregion
 export { dispatchCustomEvent };

package/dist/callbacks/dispatch/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":["dispatchCustomEventWeb"],"sources":["../../../src/callbacks/dispatch/index.ts"],"sourcesContent":["/* __LC_ALLOW_ENTRYPOINT_SIDE_EFFECTS__ */\n\nimport { AsyncLocalStorage } from \"node:async_hooks\";\nimport { dispatchCustomEvent as dispatchCustomEventWeb } from \"./web.js\";\nimport { type RunnableConfig, ensureConfig } from \"../../runnables/config.js\";\nimport { AsyncLocalStorageProviderSingleton } from \"../../singletons/index.js\";\n\nAsyncLocalStorageProviderSingleton.initializeGlobalInstance(\n  new AsyncLocalStorage()\n);\n\n/**\n * Dispatch a custom event.\n *\n * Note: this method is only supported in non-web environments\n * due to usage of async_hooks to infer config.\n *\n * If you are using this method in the browser, please import and use\n * from \"@langchain/core/callbacks/dispatch/web\".\n *\n * @param name The name of the custom event.\n * @param payload The data for the custom event.\n *   Ideally should be JSON serializable to avoid serialization issues downstream, but not enforced.\n * @param config Optional config object.\n *\n * @example\n * ```typescript\n * import { dispatchCustomEvent } from \"@langchain/core/callbacks/dispatch\";\n *\n * const foo = RunnableLambda.from(async (input: string) => {\n *   await dispatchCustomEvent(\"my_custom_event\", { arbitraryField: \"someval\" });\n *   return input;\n * });\n *\n * const callbacks = [{\n *   handleCustomEvent: (eventName: string, payload: any) => {\n *     // Logs \"my_custom_event\" and { arbitraryField: \"someval\" }\n *     console.log(eventName, payload);\n *   }\n * }];\n *\n * await foo.invoke(\"hi\", { callbacks })\n * ```\n */\nexport async function dispatchCustomEvent(\n  eventName: string,\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  payload: any,\n  config?: RunnableConfig\n) {\n  const ensuredConfig = ensureConfig(config);\n  await dispatchCustomEventWeb(eventName, payload, ensuredConfig);\n}\n"],"mappings":";;;;;;AAOA,mCAAmC,yBACjC,IAAI,kBAAkB,CACxB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmCA,eAAsB,oBACpB,WAEA,SACA,QACA;CAEA,MAAMA,sBAAuB,WAAW,SADlB,aAAa,MAC0B,CAAC;AAChE"}
+{"version":3,"file":"index.js","names":["dispatchCustomEventWeb"],"sources":["../../../src/callbacks/dispatch/index.ts"],"sourcesContent":["/* __LC_ALLOW_ENTRYPOINT_SIDE_EFFECTS__ */\n\nimport { AsyncLocalStorage } from \"node:async_hooks\";\nimport { dispatchCustomEvent as dispatchCustomEventWeb } from \"./web.js\";\nimport { type RunnableConfig, ensureConfig } from \"../../runnables/config.js\";\nimport { AsyncLocalStorageProviderSingleton } from \"../../singletons/index.js\";\n\nAsyncLocalStorageProviderSingleton.initializeGlobalInstance(\n  new AsyncLocalStorage()\n);\n\n/**\n * Dispatch a custom event.\n *\n * Note: this method is only supported in non-web environments\n * due to usage of async_hooks to infer config.\n *\n * If you are using this method in the browser, please import and use\n * from \"@langchain/core/callbacks/dispatch/web\".\n *\n * @param name The name of the custom event.\n * @param payload The data for the custom event.\n *   Ideally should be JSON serializable to avoid serialization issues downstream, but not enforced.\n * @param config Optional config object.\n *\n * @example\n * ```typescript\n * import { dispatchCustomEvent } from \"@langchain/core/callbacks/dispatch\";\n *\n * const foo = RunnableLambda.from(async (input: string) => {\n *   await dispatchCustomEvent(\"my_custom_event\", { arbitraryField: \"someval\" });\n *   return input;\n * });\n *\n * const callbacks = [{\n *   handleCustomEvent: (eventName: string, payload: any) => {\n *     // Logs \"my_custom_event\" and { arbitraryField: \"someval\" }\n *     console.log(eventName, payload);\n *   }\n * }];\n *\n * await foo.invoke(\"hi\", { callbacks })\n * ```\n */\nexport async function dispatchCustomEvent(\n  eventName: string,\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  payload: any,\n  config?: RunnableConfig\n) {\n  const ensuredConfig = ensureConfig(config);\n  await dispatchCustomEventWeb(eventName, payload, ensuredConfig);\n}\n"],"mappings":";;;;;;AAOA,mCAAmC,yBACjC,IAAI,kBAAkB,CACxB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmCA,eAAsB,oBACpB,WAEA,SACA,QACA;CACA,MAAM,gBAAgB,aAAa,MAAM;CACzC,MAAMA,sBAAuB,WAAW,SAAS,aAAa;AAChE"}

package/dist/indexing/base.cjs

@@ -59,11 +59,13 @@ var _HashedDocument = class {
 		return doc;
 	}
 	_hashStringToUUID(inputString) {
-		return require_utils_uuid_index.v5(this.keyEncoder(inputString), require_record_manager.UUIDV5_NAMESPACE);
+		const hash_value = this.keyEncoder(inputString);
+		return require_utils_uuid_index.v5(hash_value, require_record_manager.UUIDV5_NAMESPACE);
 	}
 	_hashNestedDictToUUID(data) {
 		const serialized_data = JSON.stringify(data, Object.keys(data).sort());
-		return require_utils_uuid_index.v5(this.keyEncoder(serialized_data), require_record_manager.UUIDV5_NAMESPACE);
+		const hash_value = this.keyEncoder(serialized_data);
+		return require_utils_uuid_index.v5(hash_value, require_record_manager.UUIDV5_NAMESPACE);
 	}
 };
 function _batch(size, iterable) {
@@ -144,10 +146,12 @@ async function index(args) {
 		const docsToUpdate = [];
 		const seenDocs = /* @__PURE__ */ new Set();
 		hashedDocs.forEach((hashedDoc, i) => {
-			if (batchExists[i]) if (forceUpdate) seenDocs.add(hashedDoc.uid);
-			else {
-				docsToUpdate.push(hashedDoc.uid);
-				return;
+			if (batchExists[i]) {
+				if (forceUpdate) seenDocs.add(hashedDoc.uid);
+				else {
+					docsToUpdate.push(hashedDoc.uid);
+					return;
+				}
 			}
 			uids.push(hashedDoc.uid);
 			docsToIndex.push(hashedDoc.toDocument());

package/dist/indexing/base.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"base.cjs","names":["sha256","Document","uuidv5","UUIDV5_NAMESPACE"],"sources":["../../src/indexing/base.ts"],"sourcesContent":["import { v5 as uuidv5 } from \"../utils/uuid/index.js\";\nimport { VectorStore } from \"../vectorstores.js\";\nimport { RecordManagerInterface, UUIDV5_NAMESPACE } from \"./record_manager.js\";\nimport { sha256, type HashKeyEncoder } from \"../utils/hash.js\";\nimport { DocumentInterface, Document } from \"../documents/document.js\";\nimport { BaseDocumentLoader } from \"../document_loaders/base.js\";\n\ntype Metadata = Record<string, unknown>;\n\ntype IndexingResult = {\n  numAdded: number;\n  numDeleted: number;\n  numUpdated: number;\n  numSkipped: number;\n};\n\ntype StringOrDocFunc = string | ((doc: DocumentInterface) => string);\n\nexport interface HashedDocumentInterface extends DocumentInterface {\n  uid: string;\n  hash_?: string;\n  contentHash?: string;\n  metadataHash?: string;\n  pageContent: string;\n  metadata: Metadata;\n  calculateHashes(): void;\n  toDocument(): DocumentInterface;\n}\n\ninterface HashedDocumentArgs {\n  pageContent: string;\n  metadata: Metadata;\n  uid: string;\n}\n\n/**\n * HashedDocument is a Document with hashes calculated.\n * Hashes are calculated based on page content and metadata.\n * It is used for indexing.\n */\nexport class _HashedDocument implements HashedDocumentInterface {\n  uid: string;\n\n  hash_?: string;\n\n  contentHash?: string;\n\n  metadataHash?: string;\n\n  pageContent: string;\n\n  metadata: Metadata;\n\n  private keyEncoder: HashKeyEncoder = sha256;\n\n  constructor(fields: HashedDocumentArgs) {\n    this.uid = fields.uid;\n    this.pageContent = fields.pageContent;\n    this.metadata = fields.metadata;\n  }\n\n  makeDefaultKeyEncoder(keyEncoderFn: HashKeyEncoder): void {\n    this.keyEncoder = keyEncoderFn;\n  }\n\n  calculateHashes(): void {\n    const forbiddenKeys = [\"hash_\", \"content_hash\", \"metadata_hash\"];\n\n    for (const key of forbiddenKeys) {\n      if (key in this.metadata) {\n        throw new Error(\n          `Metadata cannot contain key ${key} as it is reserved for internal use. Restricted keys: [${forbiddenKeys.join(\n            \", \"\n          )}]`\n        );\n      }\n    }\n\n    const contentHash = this._hashStringToUUID(this.pageContent);\n\n    try {\n      const metadataHash = this._hashNestedDictToUUID(this.metadata);\n      this.contentHash = contentHash;\n      this.metadataHash = metadataHash;\n    } catch (e) {\n      throw new Error(\n        `Failed to hash metadata: ${e}. Please use a dict that can be serialized using json.`\n      );\n    }\n\n    this.hash_ = this._hashStringToUUID(this.contentHash + this.metadataHash);\n\n    if (!this.uid) {\n      this.uid = this.hash_;\n    }\n  }\n\n  toDocument(): DocumentInterface {\n    return new Document({\n      pageContent: this.pageContent,\n      metadata: this.metadata,\n    });\n  }\n\n  static fromDocument(\n    document: DocumentInterface,\n    uid?: string\n  ): _HashedDocument {\n    const doc = new this({\n      pageContent: document.pageContent,\n      metadata: document.metadata,\n      uid: uid || (document as DocumentInterface & { uid: string }).uid,\n    });\n    doc.calculateHashes();\n    return doc;\n  }\n\n  private _hashStringToUUID(inputString: string): string {\n    const hash_value = this.keyEncoder(inputString);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n\n  private _hashNestedDictToUUID(data: Record<string, unknown>): string {\n    const serialized_data = JSON.stringify(data, Object.keys(data).sort());\n    const hash_value = this.keyEncoder(serialized_data);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n}\n\nexport type CleanupMode = \"full\" | \"incremental\";\n\nexport type IndexOptions = {\n  /**\n   * The number of documents to index in one batch.\n   */\n  batchSize?: number;\n  /**\n   * The cleanup mode to use. Can be \"full\", \"incremental\" or undefined.\n   * - **Incremental**: Cleans up all documents that haven't been updated AND\n   *   that are associated with source ids that were seen\n   *   during indexing.\n   *   Clean up is done continuously during indexing helping\n   *   to minimize the probability of users seeing duplicated\n   *   content.\n   * - **Full**: Delete all documents that haven to been returned by the loader.\n   *   Clean up runs after all documents have been indexed.\n   *   This means that users may see duplicated content during indexing.\n   * - **undefined**: Do not delete any documents.\n   */\n  cleanup?: CleanupMode;\n  /**\n   * Optional key that helps identify the original source of the document.\n   * Must either be a string representing the key of the source in the metadata\n   * or a function that takes a document and returns a string representing the source.\n   * **Required when cleanup is incremental**.\n   */\n  sourceIdKey?: StringOrDocFunc;\n  /**\n   * Batch size to use when cleaning up documents.\n   */\n  cleanupBatchSize?: number;\n  /**\n   * Force update documents even if they are present in the\n   * record manager. Useful if you are re-indexing with updated embeddings.\n   */\n  forceUpdate?: boolean;\n};\n\nexport function _batch<T>(size: number, iterable: T[]): T[][] {\n  const batches: T[][] = [];\n  let currentBatch: T[] = [];\n\n  iterable.forEach((item) => {\n    currentBatch.push(item);\n\n    if (currentBatch.length >= size) {\n      batches.push(currentBatch);\n      currentBatch = [];\n    }\n  });\n\n  if (currentBatch.length > 0) {\n    batches.push(currentBatch);\n  }\n\n  return batches;\n}\n\nexport function _deduplicateInOrder(\n  hashedDocuments: HashedDocumentInterface[]\n): HashedDocumentInterface[] {\n  const seen = new Set<string>();\n  const deduplicated: HashedDocumentInterface[] = [];\n\n  for (const hashedDoc of hashedDocuments) {\n    if (!hashedDoc.hash_) {\n      throw new Error(\"Hashed document does not have a hash\");\n    }\n\n    if (!seen.has(hashedDoc.hash_)) {\n      seen.add(hashedDoc.hash_);\n      deduplicated.push(hashedDoc);\n    }\n  }\n  return deduplicated;\n}\n\nexport function _getSourceIdAssigner(\n  sourceIdKey: StringOrDocFunc | null\n): (doc: DocumentInterface) => string | null {\n  if (sourceIdKey === null) {\n    return (_doc: DocumentInterface) => null;\n  } else if (typeof sourceIdKey === \"string\") {\n    return (doc: DocumentInterface) => doc.metadata[sourceIdKey];\n  } else if (typeof sourceIdKey === \"function\") {\n    return sourceIdKey;\n  } else {\n    throw new Error(\n      `sourceIdKey should be null, a string or a function, got ${typeof sourceIdKey}`\n    );\n  }\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport const _isBaseDocumentLoader = (arg: any): arg is BaseDocumentLoader => {\n  if (\n    \"load\" in arg &&\n    typeof arg.load === \"function\" &&\n    \"loadAndSplit\" in arg &&\n    typeof arg.loadAndSplit === \"function\"\n  ) {\n    return true;\n  }\n  return false;\n};\n\ninterface IndexArgs {\n  docsSource: BaseDocumentLoader | DocumentInterface[];\n  recordManager: RecordManagerInterface;\n  vectorStore: VectorStore;\n  options?: IndexOptions;\n}\n\n/**\n * Index data from the doc source into the vector store.\n *\n * Indexing functionality uses a manager to keep track of which documents\n * are in the vector store.\n *\n * This allows us to keep track of which documents were updated, and which\n * documents were deleted, which documents should be skipped.\n *\n * For the time being, documents are indexed using their hashes, and users\n *  are not able to specify the uid of the document.\n *\n * @param {IndexArgs} args\n * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.\n * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.\n * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.\n * @param {IndexOptions | undefined} args.options Options for indexing.\n * @returns {Promise<IndexingResult>}\n */\nexport async function index(args: IndexArgs): Promise<IndexingResult> {\n  const { docsSource, recordManager, vectorStore, options } = args;\n  const {\n    batchSize = 100,\n    cleanup,\n    sourceIdKey,\n    cleanupBatchSize = 1000,\n    forceUpdate = false,\n  } = options ?? {};\n\n  if (cleanup === \"incremental\" && !sourceIdKey) {\n    throw new Error(\n      \"sourceIdKey is required when cleanup mode is incremental. Please provide through 'options.sourceIdKey'.\"\n    );\n  }\n\n  const docs = _isBaseDocumentLoader(docsSource)\n    ? await docsSource.load()\n    : docsSource;\n\n  const sourceIdAssigner = _getSourceIdAssigner(sourceIdKey ?? null);\n\n  const indexStartDt = await recordManager.getTime();\n  let numAdded = 0;\n  let numDeleted = 0;\n  let numUpdated = 0;\n  let numSkipped = 0;\n\n  const batches = _batch<DocumentInterface>(batchSize ?? 100, docs);\n\n  for (const batch of batches) {\n    const hashedDocs = _deduplicateInOrder(\n      batch.map((doc) => _HashedDocument.fromDocument(doc))\n    );\n\n    const sourceIds = hashedDocs.map((doc) => sourceIdAssigner(doc));\n\n    if (cleanup === \"incremental\") {\n      hashedDocs.forEach((_hashedDoc, index) => {\n        const source = sourceIds[index];\n        if (source === null) {\n          throw new Error(\n            \"sourceIdKey must be provided when cleanup is incremental\"\n          );\n        }\n      });\n    }\n\n    const batchExists = await recordManager.exists(\n      hashedDocs.map((doc) => doc.uid)\n    );\n\n    const uids: string[] = [];\n    const docsToIndex: DocumentInterface[] = [];\n    const docsToUpdate: string[] = [];\n    const seenDocs = new Set<string>();\n    hashedDocs.forEach((hashedDoc, i) => {\n      const docExists = batchExists[i];\n      if (docExists) {\n        if (forceUpdate) {\n          seenDocs.add(hashedDoc.uid);\n        } else {\n          docsToUpdate.push(hashedDoc.uid);\n          return;\n        }\n      }\n      uids.push(hashedDoc.uid);\n      docsToIndex.push(hashedDoc.toDocument());\n    });\n\n    if (docsToUpdate.length > 0) {\n      await recordManager.update(docsToUpdate, { timeAtLeast: indexStartDt });\n      numSkipped += docsToUpdate.length;\n    }\n\n    if (docsToIndex.length > 0) {\n      await vectorStore.addDocuments(docsToIndex, { ids: uids });\n      numAdded += docsToIndex.length - seenDocs.size;\n      numUpdated += seenDocs.size;\n    }\n\n    await recordManager.update(\n      hashedDocs.map((doc) => doc.uid),\n      { timeAtLeast: indexStartDt, groupIds: sourceIds }\n    );\n\n    if (cleanup === \"incremental\") {\n      sourceIds.forEach((sourceId) => {\n        if (!sourceId) throw new Error(\"Source id cannot be null\");\n      });\n      const uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        groupIds: sourceIds,\n      });\n\n      if (uidsToDelete.length > 0) {\n        await vectorStore.delete({ ids: uidsToDelete });\n        await recordManager.deleteKeys(uidsToDelete);\n        numDeleted += uidsToDelete.length;\n      }\n    }\n  }\n\n  if (cleanup === \"full\") {\n    let uidsToDelete = await recordManager.listKeys({\n      before: indexStartDt,\n      limit: cleanupBatchSize,\n    });\n    while (uidsToDelete.length > 0) {\n      await vectorStore.delete({ ids: uidsToDelete });\n      await recordManager.deleteKeys(uidsToDelete);\n      numDeleted += uidsToDelete.length;\n      uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        limit: cleanupBatchSize,\n      });\n    }\n  }\n\n  return {\n    numAdded,\n    numDeleted,\n    numUpdated,\n    numSkipped,\n  };\n}\n"],"mappings":";;;;;;;;;;;AAwCA,IAAa,kBAAb,MAAgE;CAC9D;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA,aAAqCA,aAAAA;CAErC,YAAY,QAA4B;EACtC,KAAK,MAAM,OAAO;EAClB,KAAK,cAAc,OAAO;EAC1B,KAAK,WAAW,OAAO;CACzB;CAEA,sBAAsB,cAAoC;EACxD,KAAK,aAAa;CACpB;CAEA,kBAAwB;EACtB,MAAM,gBAAgB;GAAC;GAAS;GAAgB;EAAe;EAE/D,KAAK,MAAM,OAAO,eAChB,IAAI,OAAO,KAAK,UACd,MAAM,IAAI,MACR,+BAA+B,IAAI,yDAAyD,cAAc,KACxG,IACF,EAAE,EACJ;EAIJ,MAAM,cAAc,KAAK,kBAAkB,KAAK,WAAW;EAE3D,IAAI;GACF,MAAM,eAAe,KAAK,sBAAsB,KAAK,QAAQ;GAC7D,KAAK,cAAc;GACnB,KAAK,eAAe;EACtB,SAAS,GAAG;GACV,MAAM,IAAI,MACR,4BAA4B,EAAE,uDAChC;EACF;EAEA,KAAK,QAAQ,KAAK,kBAAkB,KAAK,cAAc,KAAK,YAAY;EAExE,IAAI,CAAC,KAAK,KACR,KAAK,MAAM,KAAK;CAEpB;CAEA,aAAgC;EAC9B,OAAO,IAAIC,iBAAAA,SAAS;GAClB,aAAa,KAAK;GAClB,UAAU,KAAK;EACjB,CAAC;CACH;CAEA,OAAO,aACL,UACA,KACiB;EACjB,MAAM,MAAM,IAAI,KAAK;GACnB,aAAa,SAAS;GACtB,UAAU,SAAS;GACnB,KAAK,OAAQ,SAAiD;EAChE,CAAC;EACD,IAAI,gBAAgB;EACpB,OAAO;CACT;CAEA,kBAA0B,aAA6B;EAErD,OAAOC,yBAAAA,GADY,KAAK,WAAW,WACZ,GAAGC,uBAAAA,gBAAgB;CAC5C;CAEA,sBAA8B,MAAuC;EACnE,MAAM,kBAAkB,KAAK,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC;EAErE,OAAOD,yBAAAA,GADY,KAAK,WAAW,eACZ,GAAGC,uBAAAA,gBAAgB;CAC5C;AACF;AAyCA,SAAgB,OAAU,MAAc,UAAsB;CAC5D,MAAM,UAAiB,CAAC;CACxB,IAAI,eAAoB,CAAC;CAEzB,SAAS,SAAS,SAAS;EACzB,aAAa,KAAK,IAAI;EAEtB,IAAI,aAAa,UAAU,MAAM;GAC/B,QAAQ,KAAK,YAAY;GACzB,eAAe,CAAC;EAClB;CACF,CAAC;CAED,IAAI,aAAa,SAAS,GACxB,QAAQ,KAAK,YAAY;CAG3B,OAAO;AACT;AAEA,SAAgB,oBACd,iBAC2B;CAC3B,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,eAA0C,CAAC;CAEjD,KAAK,MAAM,aAAa,iBAAiB;EACvC,IAAI,CAAC,UAAU,OACb,MAAM,IAAI,MAAM,sCAAsC;EAGxD,IAAI,CAAC,KAAK,IAAI,UAAU,KAAK,GAAG;GAC9B,KAAK,IAAI,UAAU,KAAK;GACxB,aAAa,KAAK,SAAS;EAC7B;CACF;CACA,OAAO;AACT;AAEA,SAAgB,qBACd,aAC2C;CAC3C,IAAI,gBAAgB,MAClB,QAAQ,SAA4B;MAC/B,IAAI,OAAO,gBAAgB,UAChC,QAAQ,QAA2B,IAAI,SAAS;MAC3C,IAAI,OAAO,gBAAgB,YAChC,OAAO;MAEP,MAAM,IAAI,MACR,2DAA2D,OAAO,aACpE;AAEJ;AAGA,MAAa,yBAAyB,QAAwC;CAC5E,IACE,UAAU,OACV,OAAO,IAAI,SAAS,cACpB,kBAAkB,OAClB,OAAO,IAAI,iBAAiB,YAE5B,OAAO;CAET,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AA4BA,eAAsB,MAAM,MAA0C;CACpE,MAAM,EAAE,YAAY,eAAe,aAAa,YAAY;CAC5D,MAAM,EACJ,YAAY,KACZ,SACA,aACA,mBAAmB,KACnB,cAAc,UACZ,WAAW,CAAC;CAEhB,IAAI,YAAY,iBAAiB,CAAC,aAChC,MAAM,IAAI,MACR,yGACF;CAGF,MAAM,OAAO,sBAAsB,UAAU,IACzC,MAAM,WAAW,KAAK,IACtB;CAEJ,MAAM,mBAAmB,qBAAqB,eAAe,IAAI;CAEjE,MAAM,eAAe,MAAM,cAAc,QAAQ;CACjD,IAAI,WAAW;CACf,IAAI,aAAa;CACjB,IAAI,aAAa;CACjB,IAAI,aAAa;CAEjB,MAAM,UAAU,OAA0B,aAAa,KAAK,IAAI;CAEhE,KAAK,MAAM,SAAS,SAAS;EAC3B,MAAM,aAAa,oBACjB,MAAM,KAAK,QAAQ,gBAAgB,aAAa,GAAG,CAAC,CACtD;EAEA,MAAM,YAAY,WAAW,KAAK,QAAQ,iBAAiB,GAAG,CAAC;EAE/D,IAAI,YAAY,eACd,WAAW,SAAS,YAAY,UAAU;GAExC,IADe,UAAU,WACV,MACb,MAAM,IAAI,MACR,0DACF;EAEJ,CAAC;EAGH,MAAM,cAAc,MAAM,cAAc,OACtC,WAAW,KAAK,QAAQ,IAAI,GAAG,CACjC;EAEA,MAAM,OAAiB,CAAC;EACxB,MAAM,cAAmC,CAAC;EAC1C,MAAM,eAAyB,CAAC;EAChC,MAAM,2BAAW,IAAI,IAAY;EACjC,WAAW,SAAS,WAAW,MAAM;GAEnC,IADkB,YAAY,IAE5B,IAAI,aACF,SAAS,IAAI,UAAU,GAAG;QACrB;IACL,aAAa,KAAK,UAAU,GAAG;IAC/B;GACF;GAEF,KAAK,KAAK,UAAU,GAAG;GACvB,YAAY,KAAK,UAAU,WAAW,CAAC;EACzC,CAAC;EAED,IAAI,aAAa,SAAS,GAAG;GAC3B,MAAM,cAAc,OAAO,cAAc,EAAE,aAAa,aAAa,CAAC;GACtE,cAAc,aAAa;EAC7B;EAEA,IAAI,YAAY,SAAS,GAAG;GAC1B,MAAM,YAAY,aAAa,aAAa,EAAE,KAAK,KAAK,CAAC;GACzD,YAAY,YAAY,SAAS,SAAS;GAC1C,cAAc,SAAS;EACzB;EAEA,MAAM,cAAc,OAClB,WAAW,KAAK,QAAQ,IAAI,GAAG,GAC/B;GAAE,aAAa;GAAc,UAAU;EAAU,CACnD;EAEA,IAAI,YAAY,eAAe;GAC7B,UAAU,SAAS,aAAa;IAC9B,IAAI,CAAC,UAAU,MAAM,IAAI,MAAM,0BAA0B;GAC3D,CAAC;GACD,MAAM,eAAe,MAAM,cAAc,SAAS;IAChD,QAAQ;IACR,UAAU;GACZ,CAAC;GAED,IAAI,aAAa,SAAS,GAAG;IAC3B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;IAC9C,MAAM,cAAc,WAAW,YAAY;IAC3C,cAAc,aAAa;GAC7B;EACF;CACF;CAEA,IAAI,YAAY,QAAQ;EACtB,IAAI,eAAe,MAAM,cAAc,SAAS;GAC9C,QAAQ;GACR,OAAO;EACT,CAAC;EACD,OAAO,aAAa,SAAS,GAAG;GAC9B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;GAC9C,MAAM,cAAc,WAAW,YAAY;GAC3C,cAAc,aAAa;GAC3B,eAAe,MAAM,cAAc,SAAS;IAC1C,QAAQ;IACR,OAAO;GACT,CAAC;EACH;CACF;CAEA,OAAO;EACL;EACA;EACA;EACA;CACF;AACF"}
+{"version":3,"file":"base.cjs","names":["sha256","Document","uuidv5","UUIDV5_NAMESPACE"],"sources":["../../src/indexing/base.ts"],"sourcesContent":["import { v5 as uuidv5 } from \"../utils/uuid/index.js\";\nimport { VectorStore } from \"../vectorstores.js\";\nimport { RecordManagerInterface, UUIDV5_NAMESPACE } from \"./record_manager.js\";\nimport { sha256, type HashKeyEncoder } from \"../utils/hash.js\";\nimport { DocumentInterface, Document } from \"../documents/document.js\";\nimport { BaseDocumentLoader } from \"../document_loaders/base.js\";\n\ntype Metadata = Record<string, unknown>;\n\ntype IndexingResult = {\n  numAdded: number;\n  numDeleted: number;\n  numUpdated: number;\n  numSkipped: number;\n};\n\ntype StringOrDocFunc = string | ((doc: DocumentInterface) => string);\n\nexport interface HashedDocumentInterface extends DocumentInterface {\n  uid: string;\n  hash_?: string;\n  contentHash?: string;\n  metadataHash?: string;\n  pageContent: string;\n  metadata: Metadata;\n  calculateHashes(): void;\n  toDocument(): DocumentInterface;\n}\n\ninterface HashedDocumentArgs {\n  pageContent: string;\n  metadata: Metadata;\n  uid: string;\n}\n\n/**\n * HashedDocument is a Document with hashes calculated.\n * Hashes are calculated based on page content and metadata.\n * It is used for indexing.\n */\nexport class _HashedDocument implements HashedDocumentInterface {\n  uid: string;\n\n  hash_?: string;\n\n  contentHash?: string;\n\n  metadataHash?: string;\n\n  pageContent: string;\n\n  metadata: Metadata;\n\n  private keyEncoder: HashKeyEncoder = sha256;\n\n  constructor(fields: HashedDocumentArgs) {\n    this.uid = fields.uid;\n    this.pageContent = fields.pageContent;\n    this.metadata = fields.metadata;\n  }\n\n  makeDefaultKeyEncoder(keyEncoderFn: HashKeyEncoder): void {\n    this.keyEncoder = keyEncoderFn;\n  }\n\n  calculateHashes(): void {\n    const forbiddenKeys = [\"hash_\", \"content_hash\", \"metadata_hash\"];\n\n    for (const key of forbiddenKeys) {\n      if (key in this.metadata) {\n        throw new Error(\n          `Metadata cannot contain key ${key} as it is reserved for internal use. Restricted keys: [${forbiddenKeys.join(\n            \", \"\n          )}]`\n        );\n      }\n    }\n\n    const contentHash = this._hashStringToUUID(this.pageContent);\n\n    try {\n      const metadataHash = this._hashNestedDictToUUID(this.metadata);\n      this.contentHash = contentHash;\n      this.metadataHash = metadataHash;\n    } catch (e) {\n      throw new Error(\n        `Failed to hash metadata: ${e}. Please use a dict that can be serialized using json.`\n      );\n    }\n\n    this.hash_ = this._hashStringToUUID(this.contentHash + this.metadataHash);\n\n    if (!this.uid) {\n      this.uid = this.hash_;\n    }\n  }\n\n  toDocument(): DocumentInterface {\n    return new Document({\n      pageContent: this.pageContent,\n      metadata: this.metadata,\n    });\n  }\n\n  static fromDocument(\n    document: DocumentInterface,\n    uid?: string\n  ): _HashedDocument {\n    const doc = new this({\n      pageContent: document.pageContent,\n      metadata: document.metadata,\n      uid: uid || (document as DocumentInterface & { uid: string }).uid,\n    });\n    doc.calculateHashes();\n    return doc;\n  }\n\n  private _hashStringToUUID(inputString: string): string {\n    const hash_value = this.keyEncoder(inputString);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n\n  private _hashNestedDictToUUID(data: Record<string, unknown>): string {\n    const serialized_data = JSON.stringify(data, Object.keys(data).sort());\n    const hash_value = this.keyEncoder(serialized_data);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n}\n\nexport type CleanupMode = \"full\" | \"incremental\";\n\nexport type IndexOptions = {\n  /**\n   * The number of documents to index in one batch.\n   */\n  batchSize?: number;\n  /**\n   * The cleanup mode to use. Can be \"full\", \"incremental\" or undefined.\n   * - **Incremental**: Cleans up all documents that haven't been updated AND\n   *   that are associated with source ids that were seen\n   *   during indexing.\n   *   Clean up is done continuously during indexing helping\n   *   to minimize the probability of users seeing duplicated\n   *   content.\n   * - **Full**: Delete all documents that haven to been returned by the loader.\n   *   Clean up runs after all documents have been indexed.\n   *   This means that users may see duplicated content during indexing.\n   * - **undefined**: Do not delete any documents.\n   */\n  cleanup?: CleanupMode;\n  /**\n   * Optional key that helps identify the original source of the document.\n   * Must either be a string representing the key of the source in the metadata\n   * or a function that takes a document and returns a string representing the source.\n   * **Required when cleanup is incremental**.\n   */\n  sourceIdKey?: StringOrDocFunc;\n  /**\n   * Batch size to use when cleaning up documents.\n   */\n  cleanupBatchSize?: number;\n  /**\n   * Force update documents even if they are present in the\n   * record manager. Useful if you are re-indexing with updated embeddings.\n   */\n  forceUpdate?: boolean;\n};\n\nexport function _batch<T>(size: number, iterable: T[]): T[][] {\n  const batches: T[][] = [];\n  let currentBatch: T[] = [];\n\n  iterable.forEach((item) => {\n    currentBatch.push(item);\n\n    if (currentBatch.length >= size) {\n      batches.push(currentBatch);\n      currentBatch = [];\n    }\n  });\n\n  if (currentBatch.length > 0) {\n    batches.push(currentBatch);\n  }\n\n  return batches;\n}\n\nexport function _deduplicateInOrder(\n  hashedDocuments: HashedDocumentInterface[]\n): HashedDocumentInterface[] {\n  const seen = new Set<string>();\n  const deduplicated: HashedDocumentInterface[] = [];\n\n  for (const hashedDoc of hashedDocuments) {\n    if (!hashedDoc.hash_) {\n      throw new Error(\"Hashed document does not have a hash\");\n    }\n\n    if (!seen.has(hashedDoc.hash_)) {\n      seen.add(hashedDoc.hash_);\n      deduplicated.push(hashedDoc);\n    }\n  }\n  return deduplicated;\n}\n\nexport function _getSourceIdAssigner(\n  sourceIdKey: StringOrDocFunc | null\n): (doc: DocumentInterface) => string | null {\n  if (sourceIdKey === null) {\n    return (_doc: DocumentInterface) => null;\n  } else if (typeof sourceIdKey === \"string\") {\n    return (doc: DocumentInterface) => doc.metadata[sourceIdKey];\n  } else if (typeof sourceIdKey === \"function\") {\n    return sourceIdKey;\n  } else {\n    throw new Error(\n      `sourceIdKey should be null, a string or a function, got ${typeof sourceIdKey}`\n    );\n  }\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport const _isBaseDocumentLoader = (arg: any): arg is BaseDocumentLoader => {\n  if (\n    \"load\" in arg &&\n    typeof arg.load === \"function\" &&\n    \"loadAndSplit\" in arg &&\n    typeof arg.loadAndSplit === \"function\"\n  ) {\n    return true;\n  }\n  return false;\n};\n\ninterface IndexArgs {\n  docsSource: BaseDocumentLoader | DocumentInterface[];\n  recordManager: RecordManagerInterface;\n  vectorStore: VectorStore;\n  options?: IndexOptions;\n}\n\n/**\n * Index data from the doc source into the vector store.\n *\n * Indexing functionality uses a manager to keep track of which documents\n * are in the vector store.\n *\n * This allows us to keep track of which documents were updated, and which\n * documents were deleted, which documents should be skipped.\n *\n * For the time being, documents are indexed using their hashes, and users\n *  are not able to specify the uid of the document.\n *\n * @param {IndexArgs} args\n * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.\n * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.\n * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.\n * @param {IndexOptions | undefined} args.options Options for indexing.\n * @returns {Promise<IndexingResult>}\n */\nexport async function index(args: IndexArgs): Promise<IndexingResult> {\n  const { docsSource, recordManager, vectorStore, options } = args;\n  const {\n    batchSize = 100,\n    cleanup,\n    sourceIdKey,\n    cleanupBatchSize = 1000,\n    forceUpdate = false,\n  } = options ?? {};\n\n  if (cleanup === \"incremental\" && !sourceIdKey) {\n    throw new Error(\n      \"sourceIdKey is required when cleanup mode is incremental. Please provide through 'options.sourceIdKey'.\"\n    );\n  }\n\n  const docs = _isBaseDocumentLoader(docsSource)\n    ? await docsSource.load()\n    : docsSource;\n\n  const sourceIdAssigner = _getSourceIdAssigner(sourceIdKey ?? null);\n\n  const indexStartDt = await recordManager.getTime();\n  let numAdded = 0;\n  let numDeleted = 0;\n  let numUpdated = 0;\n  let numSkipped = 0;\n\n  const batches = _batch<DocumentInterface>(batchSize ?? 100, docs);\n\n  for (const batch of batches) {\n    const hashedDocs = _deduplicateInOrder(\n      batch.map((doc) => _HashedDocument.fromDocument(doc))\n    );\n\n    const sourceIds = hashedDocs.map((doc) => sourceIdAssigner(doc));\n\n    if (cleanup === \"incremental\") {\n      hashedDocs.forEach((_hashedDoc, index) => {\n        const source = sourceIds[index];\n        if (source === null) {\n          throw new Error(\n            \"sourceIdKey must be provided when cleanup is incremental\"\n          );\n        }\n      });\n    }\n\n    const batchExists = await recordManager.exists(\n      hashedDocs.map((doc) => doc.uid)\n    );\n\n    const uids: string[] = [];\n    const docsToIndex: DocumentInterface[] = [];\n    const docsToUpdate: string[] = [];\n    const seenDocs = new Set<string>();\n    hashedDocs.forEach((hashedDoc, i) => {\n      const docExists = batchExists[i];\n      if (docExists) {\n        if (forceUpdate) {\n          seenDocs.add(hashedDoc.uid);\n        } else {\n          docsToUpdate.push(hashedDoc.uid);\n          return;\n        }\n      }\n      uids.push(hashedDoc.uid);\n      docsToIndex.push(hashedDoc.toDocument());\n    });\n\n    if (docsToUpdate.length > 0) {\n      await recordManager.update(docsToUpdate, { timeAtLeast: indexStartDt });\n      numSkipped += docsToUpdate.length;\n    }\n\n    if (docsToIndex.length > 0) {\n      await vectorStore.addDocuments(docsToIndex, { ids: uids });\n      numAdded += docsToIndex.length - seenDocs.size;\n      numUpdated += seenDocs.size;\n    }\n\n    await recordManager.update(\n      hashedDocs.map((doc) => doc.uid),\n      { timeAtLeast: indexStartDt, groupIds: sourceIds }\n    );\n\n    if (cleanup === \"incremental\") {\n      sourceIds.forEach((sourceId) => {\n        if (!sourceId) throw new Error(\"Source id cannot be null\");\n      });\n      const uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        groupIds: sourceIds,\n      });\n\n      if (uidsToDelete.length > 0) {\n        await vectorStore.delete({ ids: uidsToDelete });\n        await recordManager.deleteKeys(uidsToDelete);\n        numDeleted += uidsToDelete.length;\n      }\n    }\n  }\n\n  if (cleanup === \"full\") {\n    let uidsToDelete = await recordManager.listKeys({\n      before: indexStartDt,\n      limit: cleanupBatchSize,\n    });\n    while (uidsToDelete.length > 0) {\n      await vectorStore.delete({ ids: uidsToDelete });\n      await recordManager.deleteKeys(uidsToDelete);\n      numDeleted += uidsToDelete.length;\n      uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        limit: cleanupBatchSize,\n      });\n    }\n  }\n\n  return {\n    numAdded,\n    numDeleted,\n    numUpdated,\n    numSkipped,\n  };\n}\n"],"mappings":";;;;;;;;;;;AAwCA,IAAa,kBAAb,MAAgE;CAC9D;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA,aAAqCA,aAAAA;CAErC,YAAY,QAA4B;EACtC,KAAK,MAAM,OAAO;EAClB,KAAK,cAAc,OAAO;EAC1B,KAAK,WAAW,OAAO;CACzB;CAEA,sBAAsB,cAAoC;EACxD,KAAK,aAAa;CACpB;CAEA,kBAAwB;EACtB,MAAM,gBAAgB;GAAC;GAAS;GAAgB;EAAe;EAE/D,KAAK,MAAM,OAAO,eAChB,IAAI,OAAO,KAAK,UACd,MAAM,IAAI,MACR,+BAA+B,IAAI,yDAAyD,cAAc,KACxG,IACF,EAAE,EACJ;EAIJ,MAAM,cAAc,KAAK,kBAAkB,KAAK,WAAW;EAE3D,IAAI;GACF,MAAM,eAAe,KAAK,sBAAsB,KAAK,QAAQ;GAC7D,KAAK,cAAc;GACnB,KAAK,eAAe;EACtB,SAAS,GAAG;GACV,MAAM,IAAI,MACR,4BAA4B,EAAE,uDAChC;EACF;EAEA,KAAK,QAAQ,KAAK,kBAAkB,KAAK,cAAc,KAAK,YAAY;EAExE,IAAI,CAAC,KAAK,KACR,KAAK,MAAM,KAAK;CAEpB;CAEA,aAAgC;EAC9B,OAAO,IAAIC,iBAAAA,SAAS;GAClB,aAAa,KAAK;GAClB,UAAU,KAAK;EACjB,CAAC;CACH;CAEA,OAAO,aACL,UACA,KACiB;EACjB,MAAM,MAAM,IAAI,KAAK;GACnB,aAAa,SAAS;GACtB,UAAU,SAAS;GACnB,KAAK,OAAQ,SAAiD;EAChE,CAAC;EACD,IAAI,gBAAgB;EACpB,OAAO;CACT;CAEA,kBAA0B,aAA6B;EACrD,MAAM,aAAa,KAAK,WAAW,WAAW;EAC9C,OAAOC,yBAAAA,GAAO,YAAYC,uBAAAA,gBAAgB;CAC5C;CAEA,sBAA8B,MAAuC;EACnE,MAAM,kBAAkB,KAAK,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC;EACrE,MAAM,aAAa,KAAK,WAAW,eAAe;EAClD,OAAOD,yBAAAA,GAAO,YAAYC,uBAAAA,gBAAgB;CAC5C;AACF;AAyCA,SAAgB,OAAU,MAAc,UAAsB;CAC5D,MAAM,UAAiB,CAAC;CACxB,IAAI,eAAoB,CAAC;CAEzB,SAAS,SAAS,SAAS;EACzB,aAAa,KAAK,IAAI;EAEtB,IAAI,aAAa,UAAU,MAAM;GAC/B,QAAQ,KAAK,YAAY;GACzB,eAAe,CAAC;EAClB;CACF,CAAC;CAED,IAAI,aAAa,SAAS,GACxB,QAAQ,KAAK,YAAY;CAG3B,OAAO;AACT;AAEA,SAAgB,oBACd,iBAC2B;CAC3B,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,eAA0C,CAAC;CAEjD,KAAK,MAAM,aAAa,iBAAiB;EACvC,IAAI,CAAC,UAAU,OACb,MAAM,IAAI,MAAM,sCAAsC;EAGxD,IAAI,CAAC,KAAK,IAAI,UAAU,KAAK,GAAG;GAC9B,KAAK,IAAI,UAAU,KAAK;GACxB,aAAa,KAAK,SAAS;EAC7B;CACF;CACA,OAAO;AACT;AAEA,SAAgB,qBACd,aAC2C;CAC3C,IAAI,gBAAgB,MAClB,QAAQ,SAA4B;MAC/B,IAAI,OAAO,gBAAgB,UAChC,QAAQ,QAA2B,IAAI,SAAS;MAC3C,IAAI,OAAO,gBAAgB,YAChC,OAAO;MAEP,MAAM,IAAI,MACR,2DAA2D,OAAO,aACpE;AAEJ;AAGA,MAAa,yBAAyB,QAAwC;CAC5E,IACE,UAAU,OACV,OAAO,IAAI,SAAS,cACpB,kBAAkB,OAClB,OAAO,IAAI,iBAAiB,YAE5B,OAAO;CAET,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AA4BA,eAAsB,MAAM,MAA0C;CACpE,MAAM,EAAE,YAAY,eAAe,aAAa,YAAY;CAC5D,MAAM,EACJ,YAAY,KACZ,SACA,aACA,mBAAmB,KACnB,cAAc,UACZ,WAAW,CAAC;CAEhB,IAAI,YAAY,iBAAiB,CAAC,aAChC,MAAM,IAAI,MACR,yGACF;CAGF,MAAM,OAAO,sBAAsB,UAAU,IACzC,MAAM,WAAW,KAAK,IACtB;CAEJ,MAAM,mBAAmB,qBAAqB,eAAe,IAAI;CAEjE,MAAM,eAAe,MAAM,cAAc,QAAQ;CACjD,IAAI,WAAW;CACf,IAAI,aAAa;CACjB,IAAI,aAAa;CACjB,IAAI,aAAa;CAEjB,MAAM,UAAU,OAA0B,aAAa,KAAK,IAAI;CAEhE,KAAK,MAAM,SAAS,SAAS;EAC3B,MAAM,aAAa,oBACjB,MAAM,KAAK,QAAQ,gBAAgB,aAAa,GAAG,CAAC,CACtD;EAEA,MAAM,YAAY,WAAW,KAAK,QAAQ,iBAAiB,GAAG,CAAC;EAE/D,IAAI,YAAY,eACd,WAAW,SAAS,YAAY,UAAU;GAExC,IADe,UAAU,WACV,MACb,MAAM,IAAI,MACR,0DACF;EAEJ,CAAC;EAGH,MAAM,cAAc,MAAM,cAAc,OACtC,WAAW,KAAK,QAAQ,IAAI,GAAG,CACjC;EAEA,MAAM,OAAiB,CAAC;EACxB,MAAM,cAAmC,CAAC;EAC1C,MAAM,eAAyB,CAAC;EAChC,MAAM,2BAAW,IAAI,IAAY;EACjC,WAAW,SAAS,WAAW,MAAM;GAEnC,IADkB,YAAY,IACf;IACb,IAAI,aACF,SAAS,IAAI,UAAU,GAAG;SACrB;KACL,aAAa,KAAK,UAAU,GAAG;KAC/B;IACF;GACF;GACA,KAAK,KAAK,UAAU,GAAG;GACvB,YAAY,KAAK,UAAU,WAAW,CAAC;EACzC,CAAC;EAED,IAAI,aAAa,SAAS,GAAG;GAC3B,MAAM,cAAc,OAAO,cAAc,EAAE,aAAa,aAAa,CAAC;GACtE,cAAc,aAAa;EAC7B;EAEA,IAAI,YAAY,SAAS,GAAG;GAC1B,MAAM,YAAY,aAAa,aAAa,EAAE,KAAK,KAAK,CAAC;GACzD,YAAY,YAAY,SAAS,SAAS;GAC1C,cAAc,SAAS;EACzB;EAEA,MAAM,cAAc,OAClB,WAAW,KAAK,QAAQ,IAAI,GAAG,GAC/B;GAAE,aAAa;GAAc,UAAU;EAAU,CACnD;EAEA,IAAI,YAAY,eAAe;GAC7B,UAAU,SAAS,aAAa;IAC9B,IAAI,CAAC,UAAU,MAAM,IAAI,MAAM,0BAA0B;GAC3D,CAAC;GACD,MAAM,eAAe,MAAM,cAAc,SAAS;IAChD,QAAQ;IACR,UAAU;GACZ,CAAC;GAED,IAAI,aAAa,SAAS,GAAG;IAC3B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;IAC9C,MAAM,cAAc,WAAW,YAAY;IAC3C,cAAc,aAAa;GAC7B;EACF;CACF;CAEA,IAAI,YAAY,QAAQ;EACtB,IAAI,eAAe,MAAM,cAAc,SAAS;GAC9C,QAAQ;GACR,OAAO;EACT,CAAC;EACD,OAAO,aAAa,SAAS,GAAG;GAC9B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;GAC9C,MAAM,cAAc,WAAW,YAAY;GAC3C,cAAc,aAAa;GAC3B,eAAe,MAAM,cAAc,SAAS;IAC1C,QAAQ;IACR,OAAO;GACT,CAAC;EACH;CACF;CAEA,OAAO;EACL;EACA;EACA;EACA;CACF;AACF"}

package/dist/indexing/base.js

@@ -59,11 +59,13 @@ var _HashedDocument = class {
 		return doc;
 	}
 	_hashStringToUUID(inputString) {
-		return v5(this.keyEncoder(inputString), UUIDV5_NAMESPACE);
+		const hash_value = this.keyEncoder(inputString);
+		return v5(hash_value, UUIDV5_NAMESPACE);
 	}
 	_hashNestedDictToUUID(data) {
 		const serialized_data = JSON.stringify(data, Object.keys(data).sort());
-		return v5(this.keyEncoder(serialized_data), UUIDV5_NAMESPACE);
+		const hash_value = this.keyEncoder(serialized_data);
+		return v5(hash_value, UUIDV5_NAMESPACE);
 	}
 };
 function _batch(size, iterable) {
@@ -144,10 +146,12 @@ async function index(args) {
 		const docsToUpdate = [];
 		const seenDocs = /* @__PURE__ */ new Set();
 		hashedDocs.forEach((hashedDoc, i) => {
-			if (batchExists[i]) if (forceUpdate) seenDocs.add(hashedDoc.uid);
-			else {
-				docsToUpdate.push(hashedDoc.uid);
-				return;
+			if (batchExists[i]) {
+				if (forceUpdate) seenDocs.add(hashedDoc.uid);
+				else {
+					docsToUpdate.push(hashedDoc.uid);
+					return;
+				}
 			}
 			uids.push(hashedDoc.uid);
 			docsToIndex.push(hashedDoc.toDocument());

package/dist/indexing/base.js.map

@@ -1 +1 @@
-{"version":3,"file":"base.js","names":["uuidv5"],"sources":["../../src/indexing/base.ts"],"sourcesContent":["import { v5 as uuidv5 } from \"../utils/uuid/index.js\";\nimport { VectorStore } from \"../vectorstores.js\";\nimport { RecordManagerInterface, UUIDV5_NAMESPACE } from \"./record_manager.js\";\nimport { sha256, type HashKeyEncoder } from \"../utils/hash.js\";\nimport { DocumentInterface, Document } from \"../documents/document.js\";\nimport { BaseDocumentLoader } from \"../document_loaders/base.js\";\n\ntype Metadata = Record<string, unknown>;\n\ntype IndexingResult = {\n  numAdded: number;\n  numDeleted: number;\n  numUpdated: number;\n  numSkipped: number;\n};\n\ntype StringOrDocFunc = string | ((doc: DocumentInterface) => string);\n\nexport interface HashedDocumentInterface extends DocumentInterface {\n  uid: string;\n  hash_?: string;\n  contentHash?: string;\n  metadataHash?: string;\n  pageContent: string;\n  metadata: Metadata;\n  calculateHashes(): void;\n  toDocument(): DocumentInterface;\n}\n\ninterface HashedDocumentArgs {\n  pageContent: string;\n  metadata: Metadata;\n  uid: string;\n}\n\n/**\n * HashedDocument is a Document with hashes calculated.\n * Hashes are calculated based on page content and metadata.\n * It is used for indexing.\n */\nexport class _HashedDocument implements HashedDocumentInterface {\n  uid: string;\n\n  hash_?: string;\n\n  contentHash?: string;\n\n  metadataHash?: string;\n\n  pageContent: string;\n\n  metadata: Metadata;\n\n  private keyEncoder: HashKeyEncoder = sha256;\n\n  constructor(fields: HashedDocumentArgs) {\n    this.uid = fields.uid;\n    this.pageContent = fields.pageContent;\n    this.metadata = fields.metadata;\n  }\n\n  makeDefaultKeyEncoder(keyEncoderFn: HashKeyEncoder): void {\n    this.keyEncoder = keyEncoderFn;\n  }\n\n  calculateHashes(): void {\n    const forbiddenKeys = [\"hash_\", \"content_hash\", \"metadata_hash\"];\n\n    for (const key of forbiddenKeys) {\n      if (key in this.metadata) {\n        throw new Error(\n          `Metadata cannot contain key ${key} as it is reserved for internal use. Restricted keys: [${forbiddenKeys.join(\n            \", \"\n          )}]`\n        );\n      }\n    }\n\n    const contentHash = this._hashStringToUUID(this.pageContent);\n\n    try {\n      const metadataHash = this._hashNestedDictToUUID(this.metadata);\n      this.contentHash = contentHash;\n      this.metadataHash = metadataHash;\n    } catch (e) {\n      throw new Error(\n        `Failed to hash metadata: ${e}. Please use a dict that can be serialized using json.`\n      );\n    }\n\n    this.hash_ = this._hashStringToUUID(this.contentHash + this.metadataHash);\n\n    if (!this.uid) {\n      this.uid = this.hash_;\n    }\n  }\n\n  toDocument(): DocumentInterface {\n    return new Document({\n      pageContent: this.pageContent,\n      metadata: this.metadata,\n    });\n  }\n\n  static fromDocument(\n    document: DocumentInterface,\n    uid?: string\n  ): _HashedDocument {\n    const doc = new this({\n      pageContent: document.pageContent,\n      metadata: document.metadata,\n      uid: uid || (document as DocumentInterface & { uid: string }).uid,\n    });\n    doc.calculateHashes();\n    return doc;\n  }\n\n  private _hashStringToUUID(inputString: string): string {\n    const hash_value = this.keyEncoder(inputString);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n\n  private _hashNestedDictToUUID(data: Record<string, unknown>): string {\n    const serialized_data = JSON.stringify(data, Object.keys(data).sort());\n    const hash_value = this.keyEncoder(serialized_data);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n}\n\nexport type CleanupMode = \"full\" | \"incremental\";\n\nexport type IndexOptions = {\n  /**\n   * The number of documents to index in one batch.\n   */\n  batchSize?: number;\n  /**\n   * The cleanup mode to use. Can be \"full\", \"incremental\" or undefined.\n   * - **Incremental**: Cleans up all documents that haven't been updated AND\n   *   that are associated with source ids that were seen\n   *   during indexing.\n   *   Clean up is done continuously during indexing helping\n   *   to minimize the probability of users seeing duplicated\n   *   content.\n   * - **Full**: Delete all documents that haven to been returned by the loader.\n   *   Clean up runs after all documents have been indexed.\n   *   This means that users may see duplicated content during indexing.\n   * - **undefined**: Do not delete any documents.\n   */\n  cleanup?: CleanupMode;\n  /**\n   * Optional key that helps identify the original source of the document.\n   * Must either be a string representing the key of the source in the metadata\n   * or a function that takes a document and returns a string representing the source.\n   * **Required when cleanup is incremental**.\n   */\n  sourceIdKey?: StringOrDocFunc;\n  /**\n   * Batch size to use when cleaning up documents.\n   */\n  cleanupBatchSize?: number;\n  /**\n   * Force update documents even if they are present in the\n   * record manager. Useful if you are re-indexing with updated embeddings.\n   */\n  forceUpdate?: boolean;\n};\n\nexport function _batch<T>(size: number, iterable: T[]): T[][] {\n  const batches: T[][] = [];\n  let currentBatch: T[] = [];\n\n  iterable.forEach((item) => {\n    currentBatch.push(item);\n\n    if (currentBatch.length >= size) {\n      batches.push(currentBatch);\n      currentBatch = [];\n    }\n  });\n\n  if (currentBatch.length > 0) {\n    batches.push(currentBatch);\n  }\n\n  return batches;\n}\n\nexport function _deduplicateInOrder(\n  hashedDocuments: HashedDocumentInterface[]\n): HashedDocumentInterface[] {\n  const seen = new Set<string>();\n  const deduplicated: HashedDocumentInterface[] = [];\n\n  for (const hashedDoc of hashedDocuments) {\n    if (!hashedDoc.hash_) {\n      throw new Error(\"Hashed document does not have a hash\");\n    }\n\n    if (!seen.has(hashedDoc.hash_)) {\n      seen.add(hashedDoc.hash_);\n      deduplicated.push(hashedDoc);\n    }\n  }\n  return deduplicated;\n}\n\nexport function _getSourceIdAssigner(\n  sourceIdKey: StringOrDocFunc | null\n): (doc: DocumentInterface) => string | null {\n  if (sourceIdKey === null) {\n    return (_doc: DocumentInterface) => null;\n  } else if (typeof sourceIdKey === \"string\") {\n    return (doc: DocumentInterface) => doc.metadata[sourceIdKey];\n  } else if (typeof sourceIdKey === \"function\") {\n    return sourceIdKey;\n  } else {\n    throw new Error(\n      `sourceIdKey should be null, a string or a function, got ${typeof sourceIdKey}`\n    );\n  }\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport const _isBaseDocumentLoader = (arg: any): arg is BaseDocumentLoader => {\n  if (\n    \"load\" in arg &&\n    typeof arg.load === \"function\" &&\n    \"loadAndSplit\" in arg &&\n    typeof arg.loadAndSplit === \"function\"\n  ) {\n    return true;\n  }\n  return false;\n};\n\ninterface IndexArgs {\n  docsSource: BaseDocumentLoader | DocumentInterface[];\n  recordManager: RecordManagerInterface;\n  vectorStore: VectorStore;\n  options?: IndexOptions;\n}\n\n/**\n * Index data from the doc source into the vector store.\n *\n * Indexing functionality uses a manager to keep track of which documents\n * are in the vector store.\n *\n * This allows us to keep track of which documents were updated, and which\n * documents were deleted, which documents should be skipped.\n *\n * For the time being, documents are indexed using their hashes, and users\n *  are not able to specify the uid of the document.\n *\n * @param {IndexArgs} args\n * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.\n * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.\n * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.\n * @param {IndexOptions | undefined} args.options Options for indexing.\n * @returns {Promise<IndexingResult>}\n */\nexport async function index(args: IndexArgs): Promise<IndexingResult> {\n  const { docsSource, recordManager, vectorStore, options } = args;\n  const {\n    batchSize = 100,\n    cleanup,\n    sourceIdKey,\n    cleanupBatchSize = 1000,\n    forceUpdate = false,\n  } = options ?? {};\n\n  if (cleanup === \"incremental\" && !sourceIdKey) {\n    throw new Error(\n      \"sourceIdKey is required when cleanup mode is incremental. Please provide through 'options.sourceIdKey'.\"\n    );\n  }\n\n  const docs = _isBaseDocumentLoader(docsSource)\n    ? await docsSource.load()\n    : docsSource;\n\n  const sourceIdAssigner = _getSourceIdAssigner(sourceIdKey ?? null);\n\n  const indexStartDt = await recordManager.getTime();\n  let numAdded = 0;\n  let numDeleted = 0;\n  let numUpdated = 0;\n  let numSkipped = 0;\n\n  const batches = _batch<DocumentInterface>(batchSize ?? 100, docs);\n\n  for (const batch of batches) {\n    const hashedDocs = _deduplicateInOrder(\n      batch.map((doc) => _HashedDocument.fromDocument(doc))\n    );\n\n    const sourceIds = hashedDocs.map((doc) => sourceIdAssigner(doc));\n\n    if (cleanup === \"incremental\") {\n      hashedDocs.forEach((_hashedDoc, index) => {\n        const source = sourceIds[index];\n        if (source === null) {\n          throw new Error(\n            \"sourceIdKey must be provided when cleanup is incremental\"\n          );\n        }\n      });\n    }\n\n    const batchExists = await recordManager.exists(\n      hashedDocs.map((doc) => doc.uid)\n    );\n\n    const uids: string[] = [];\n    const docsToIndex: DocumentInterface[] = [];\n    const docsToUpdate: string[] = [];\n    const seenDocs = new Set<string>();\n    hashedDocs.forEach((hashedDoc, i) => {\n      const docExists = batchExists[i];\n      if (docExists) {\n        if (forceUpdate) {\n          seenDocs.add(hashedDoc.uid);\n        } else {\n          docsToUpdate.push(hashedDoc.uid);\n          return;\n        }\n      }\n      uids.push(hashedDoc.uid);\n      docsToIndex.push(hashedDoc.toDocument());\n    });\n\n    if (docsToUpdate.length > 0) {\n      await recordManager.update(docsToUpdate, { timeAtLeast: indexStartDt });\n      numSkipped += docsToUpdate.length;\n    }\n\n    if (docsToIndex.length > 0) {\n      await vectorStore.addDocuments(docsToIndex, { ids: uids });\n      numAdded += docsToIndex.length - seenDocs.size;\n      numUpdated += seenDocs.size;\n    }\n\n    await recordManager.update(\n      hashedDocs.map((doc) => doc.uid),\n      { timeAtLeast: indexStartDt, groupIds: sourceIds }\n    );\n\n    if (cleanup === \"incremental\") {\n      sourceIds.forEach((sourceId) => {\n        if (!sourceId) throw new Error(\"Source id cannot be null\");\n      });\n      const uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        groupIds: sourceIds,\n      });\n\n      if (uidsToDelete.length > 0) {\n        await vectorStore.delete({ ids: uidsToDelete });\n        await recordManager.deleteKeys(uidsToDelete);\n        numDeleted += uidsToDelete.length;\n      }\n    }\n  }\n\n  if (cleanup === \"full\") {\n    let uidsToDelete = await recordManager.listKeys({\n      before: indexStartDt,\n      limit: cleanupBatchSize,\n    });\n    while (uidsToDelete.length > 0) {\n      await vectorStore.delete({ ids: uidsToDelete });\n      await recordManager.deleteKeys(uidsToDelete);\n      numDeleted += uidsToDelete.length;\n      uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        limit: cleanupBatchSize,\n      });\n    }\n  }\n\n  return {\n    numAdded,\n    numDeleted,\n    numUpdated,\n    numSkipped,\n  };\n}\n"],"mappings":";;;;;;;;;;;AAwCA,IAAa,kBAAb,MAAgE;CAC9D;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA,aAAqC;CAErC,YAAY,QAA4B;EACtC,KAAK,MAAM,OAAO;EAClB,KAAK,cAAc,OAAO;EAC1B,KAAK,WAAW,OAAO;CACzB;CAEA,sBAAsB,cAAoC;EACxD,KAAK,aAAa;CACpB;CAEA,kBAAwB;EACtB,MAAM,gBAAgB;GAAC;GAAS;GAAgB;EAAe;EAE/D,KAAK,MAAM,OAAO,eAChB,IAAI,OAAO,KAAK,UACd,MAAM,IAAI,MACR,+BAA+B,IAAI,yDAAyD,cAAc,KACxG,IACF,EAAE,EACJ;EAIJ,MAAM,cAAc,KAAK,kBAAkB,KAAK,WAAW;EAE3D,IAAI;GACF,MAAM,eAAe,KAAK,sBAAsB,KAAK,QAAQ;GAC7D,KAAK,cAAc;GACnB,KAAK,eAAe;EACtB,SAAS,GAAG;GACV,MAAM,IAAI,MACR,4BAA4B,EAAE,uDAChC;EACF;EAEA,KAAK,QAAQ,KAAK,kBAAkB,KAAK,cAAc,KAAK,YAAY;EAExE,IAAI,CAAC,KAAK,KACR,KAAK,MAAM,KAAK;CAEpB;CAEA,aAAgC;EAC9B,OAAO,IAAI,SAAS;GAClB,aAAa,KAAK;GAClB,UAAU,KAAK;EACjB,CAAC;CACH;CAEA,OAAO,aACL,UACA,KACiB;EACjB,MAAM,MAAM,IAAI,KAAK;GACnB,aAAa,SAAS;GACtB,UAAU,SAAS;GACnB,KAAK,OAAQ,SAAiD;EAChE,CAAC;EACD,IAAI,gBAAgB;EACpB,OAAO;CACT;CAEA,kBAA0B,aAA6B;EAErD,OAAOA,GADY,KAAK,WAAW,WACZ,GAAG,gBAAgB;CAC5C;CAEA,sBAA8B,MAAuC;EACnE,MAAM,kBAAkB,KAAK,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC;EAErE,OAAOA,GADY,KAAK,WAAW,eACZ,GAAG,gBAAgB;CAC5C;AACF;AAyCA,SAAgB,OAAU,MAAc,UAAsB;CAC5D,MAAM,UAAiB,CAAC;CACxB,IAAI,eAAoB,CAAC;CAEzB,SAAS,SAAS,SAAS;EACzB,aAAa,KAAK,IAAI;EAEtB,IAAI,aAAa,UAAU,MAAM;GAC/B,QAAQ,KAAK,YAAY;GACzB,eAAe,CAAC;EAClB;CACF,CAAC;CAED,IAAI,aAAa,SAAS,GACxB,QAAQ,KAAK,YAAY;CAG3B,OAAO;AACT;AAEA,SAAgB,oBACd,iBAC2B;CAC3B,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,eAA0C,CAAC;CAEjD,KAAK,MAAM,aAAa,iBAAiB;EACvC,IAAI,CAAC,UAAU,OACb,MAAM,IAAI,MAAM,sCAAsC;EAGxD,IAAI,CAAC,KAAK,IAAI,UAAU,KAAK,GAAG;GAC9B,KAAK,IAAI,UAAU,KAAK;GACxB,aAAa,KAAK,SAAS;EAC7B;CACF;CACA,OAAO;AACT;AAEA,SAAgB,qBACd,aAC2C;CAC3C,IAAI,gBAAgB,MAClB,QAAQ,SAA4B;MAC/B,IAAI,OAAO,gBAAgB,UAChC,QAAQ,QAA2B,IAAI,SAAS;MAC3C,IAAI,OAAO,gBAAgB,YAChC,OAAO;MAEP,MAAM,IAAI,MACR,2DAA2D,OAAO,aACpE;AAEJ;AAGA,MAAa,yBAAyB,QAAwC;CAC5E,IACE,UAAU,OACV,OAAO,IAAI,SAAS,cACpB,kBAAkB,OAClB,OAAO,IAAI,iBAAiB,YAE5B,OAAO;CAET,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AA4BA,eAAsB,MAAM,MAA0C;CACpE,MAAM,EAAE,YAAY,eAAe,aAAa,YAAY;CAC5D,MAAM,EACJ,YAAY,KACZ,SACA,aACA,mBAAmB,KACnB,cAAc,UACZ,WAAW,CAAC;CAEhB,IAAI,YAAY,iBAAiB,CAAC,aAChC,MAAM,IAAI,MACR,yGACF;CAGF,MAAM,OAAO,sBAAsB,UAAU,IACzC,MAAM,WAAW,KAAK,IACtB;CAEJ,MAAM,mBAAmB,qBAAqB,eAAe,IAAI;CAEjE,MAAM,eAAe,MAAM,cAAc,QAAQ;CACjD,IAAI,WAAW;CACf,IAAI,aAAa;CACjB,IAAI,aAAa;CACjB,IAAI,aAAa;CAEjB,MAAM,UAAU,OAA0B,aAAa,KAAK,IAAI;CAEhE,KAAK,MAAM,SAAS,SAAS;EAC3B,MAAM,aAAa,oBACjB,MAAM,KAAK,QAAQ,gBAAgB,aAAa,GAAG,CAAC,CACtD;EAEA,MAAM,YAAY,WAAW,KAAK,QAAQ,iBAAiB,GAAG,CAAC;EAE/D,IAAI,YAAY,eACd,WAAW,SAAS,YAAY,UAAU;GAExC,IADe,UAAU,WACV,MACb,MAAM,IAAI,MACR,0DACF;EAEJ,CAAC;EAGH,MAAM,cAAc,MAAM,cAAc,OACtC,WAAW,KAAK,QAAQ,IAAI,GAAG,CACjC;EAEA,MAAM,OAAiB,CAAC;EACxB,MAAM,cAAmC,CAAC;EAC1C,MAAM,eAAyB,CAAC;EAChC,MAAM,2BAAW,IAAI,IAAY;EACjC,WAAW,SAAS,WAAW,MAAM;GAEnC,IADkB,YAAY,IAE5B,IAAI,aACF,SAAS,IAAI,UAAU,GAAG;QACrB;IACL,aAAa,KAAK,UAAU,GAAG;IAC/B;GACF;GAEF,KAAK,KAAK,UAAU,GAAG;GACvB,YAAY,KAAK,UAAU,WAAW,CAAC;EACzC,CAAC;EAED,IAAI,aAAa,SAAS,GAAG;GAC3B,MAAM,cAAc,OAAO,cAAc,EAAE,aAAa,aAAa,CAAC;GACtE,cAAc,aAAa;EAC7B;EAEA,IAAI,YAAY,SAAS,GAAG;GAC1B,MAAM,YAAY,aAAa,aAAa,EAAE,KAAK,KAAK,CAAC;GACzD,YAAY,YAAY,SAAS,SAAS;GAC1C,cAAc,SAAS;EACzB;EAEA,MAAM,cAAc,OAClB,WAAW,KAAK,QAAQ,IAAI,GAAG,GAC/B;GAAE,aAAa;GAAc,UAAU;EAAU,CACnD;EAEA,IAAI,YAAY,eAAe;GAC7B,UAAU,SAAS,aAAa;IAC9B,IAAI,CAAC,UAAU,MAAM,IAAI,MAAM,0BAA0B;GAC3D,CAAC;GACD,MAAM,eAAe,MAAM,cAAc,SAAS;IAChD,QAAQ;IACR,UAAU;GACZ,CAAC;GAED,IAAI,aAAa,SAAS,GAAG;IAC3B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;IAC9C,MAAM,cAAc,WAAW,YAAY;IAC3C,cAAc,aAAa;GAC7B;EACF;CACF;CAEA,IAAI,YAAY,QAAQ;EACtB,IAAI,eAAe,MAAM,cAAc,SAAS;GAC9C,QAAQ;GACR,OAAO;EACT,CAAC;EACD,OAAO,aAAa,SAAS,GAAG;GAC9B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;GAC9C,MAAM,cAAc,WAAW,YAAY;GAC3C,cAAc,aAAa;GAC3B,eAAe,MAAM,cAAc,SAAS;IAC1C,QAAQ;IACR,OAAO;GACT,CAAC;EACH;CACF;CAEA,OAAO;EACL;EACA;EACA;EACA;CACF;AACF"}
+{"version":3,"file":"base.js","names":["uuidv5"],"sources":["../../src/indexing/base.ts"],"sourcesContent":["import { v5 as uuidv5 } from \"../utils/uuid/index.js\";\nimport { VectorStore } from \"../vectorstores.js\";\nimport { RecordManagerInterface, UUIDV5_NAMESPACE } from \"./record_manager.js\";\nimport { sha256, type HashKeyEncoder } from \"../utils/hash.js\";\nimport { DocumentInterface, Document } from \"../documents/document.js\";\nimport { BaseDocumentLoader } from \"../document_loaders/base.js\";\n\ntype Metadata = Record<string, unknown>;\n\ntype IndexingResult = {\n  numAdded: number;\n  numDeleted: number;\n  numUpdated: number;\n  numSkipped: number;\n};\n\ntype StringOrDocFunc = string | ((doc: DocumentInterface) => string);\n\nexport interface HashedDocumentInterface extends DocumentInterface {\n  uid: string;\n  hash_?: string;\n  contentHash?: string;\n  metadataHash?: string;\n  pageContent: string;\n  metadata: Metadata;\n  calculateHashes(): void;\n  toDocument(): DocumentInterface;\n}\n\ninterface HashedDocumentArgs {\n  pageContent: string;\n  metadata: Metadata;\n  uid: string;\n}\n\n/**\n * HashedDocument is a Document with hashes calculated.\n * Hashes are calculated based on page content and metadata.\n * It is used for indexing.\n */\nexport class _HashedDocument implements HashedDocumentInterface {\n  uid: string;\n\n  hash_?: string;\n\n  contentHash?: string;\n\n  metadataHash?: string;\n\n  pageContent: string;\n\n  metadata: Metadata;\n\n  private keyEncoder: HashKeyEncoder = sha256;\n\n  constructor(fields: HashedDocumentArgs) {\n    this.uid = fields.uid;\n    this.pageContent = fields.pageContent;\n    this.metadata = fields.metadata;\n  }\n\n  makeDefaultKeyEncoder(keyEncoderFn: HashKeyEncoder): void {\n    this.keyEncoder = keyEncoderFn;\n  }\n\n  calculateHashes(): void {\n    const forbiddenKeys = [\"hash_\", \"content_hash\", \"metadata_hash\"];\n\n    for (const key of forbiddenKeys) {\n      if (key in this.metadata) {\n        throw new Error(\n          `Metadata cannot contain key ${key} as it is reserved for internal use. Restricted keys: [${forbiddenKeys.join(\n            \", \"\n          )}]`\n        );\n      }\n    }\n\n    const contentHash = this._hashStringToUUID(this.pageContent);\n\n    try {\n      const metadataHash = this._hashNestedDictToUUID(this.metadata);\n      this.contentHash = contentHash;\n      this.metadataHash = metadataHash;\n    } catch (e) {\n      throw new Error(\n        `Failed to hash metadata: ${e}. Please use a dict that can be serialized using json.`\n      );\n    }\n\n    this.hash_ = this._hashStringToUUID(this.contentHash + this.metadataHash);\n\n    if (!this.uid) {\n      this.uid = this.hash_;\n    }\n  }\n\n  toDocument(): DocumentInterface {\n    return new Document({\n      pageContent: this.pageContent,\n      metadata: this.metadata,\n    });\n  }\n\n  static fromDocument(\n    document: DocumentInterface,\n    uid?: string\n  ): _HashedDocument {\n    const doc = new this({\n      pageContent: document.pageContent,\n      metadata: document.metadata,\n      uid: uid || (document as DocumentInterface & { uid: string }).uid,\n    });\n    doc.calculateHashes();\n    return doc;\n  }\n\n  private _hashStringToUUID(inputString: string): string {\n    const hash_value = this.keyEncoder(inputString);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n\n  private _hashNestedDictToUUID(data: Record<string, unknown>): string {\n    const serialized_data = JSON.stringify(data, Object.keys(data).sort());\n    const hash_value = this.keyEncoder(serialized_data);\n    return uuidv5(hash_value, UUIDV5_NAMESPACE);\n  }\n}\n\nexport type CleanupMode = \"full\" | \"incremental\";\n\nexport type IndexOptions = {\n  /**\n   * The number of documents to index in one batch.\n   */\n  batchSize?: number;\n  /**\n   * The cleanup mode to use. Can be \"full\", \"incremental\" or undefined.\n   * - **Incremental**: Cleans up all documents that haven't been updated AND\n   *   that are associated with source ids that were seen\n   *   during indexing.\n   *   Clean up is done continuously during indexing helping\n   *   to minimize the probability of users seeing duplicated\n   *   content.\n   * - **Full**: Delete all documents that haven to been returned by the loader.\n   *   Clean up runs after all documents have been indexed.\n   *   This means that users may see duplicated content during indexing.\n   * - **undefined**: Do not delete any documents.\n   */\n  cleanup?: CleanupMode;\n  /**\n   * Optional key that helps identify the original source of the document.\n   * Must either be a string representing the key of the source in the metadata\n   * or a function that takes a document and returns a string representing the source.\n   * **Required when cleanup is incremental**.\n   */\n  sourceIdKey?: StringOrDocFunc;\n  /**\n   * Batch size to use when cleaning up documents.\n   */\n  cleanupBatchSize?: number;\n  /**\n   * Force update documents even if they are present in the\n   * record manager. Useful if you are re-indexing with updated embeddings.\n   */\n  forceUpdate?: boolean;\n};\n\nexport function _batch<T>(size: number, iterable: T[]): T[][] {\n  const batches: T[][] = [];\n  let currentBatch: T[] = [];\n\n  iterable.forEach((item) => {\n    currentBatch.push(item);\n\n    if (currentBatch.length >= size) {\n      batches.push(currentBatch);\n      currentBatch = [];\n    }\n  });\n\n  if (currentBatch.length > 0) {\n    batches.push(currentBatch);\n  }\n\n  return batches;\n}\n\nexport function _deduplicateInOrder(\n  hashedDocuments: HashedDocumentInterface[]\n): HashedDocumentInterface[] {\n  const seen = new Set<string>();\n  const deduplicated: HashedDocumentInterface[] = [];\n\n  for (const hashedDoc of hashedDocuments) {\n    if (!hashedDoc.hash_) {\n      throw new Error(\"Hashed document does not have a hash\");\n    }\n\n    if (!seen.has(hashedDoc.hash_)) {\n      seen.add(hashedDoc.hash_);\n      deduplicated.push(hashedDoc);\n    }\n  }\n  return deduplicated;\n}\n\nexport function _getSourceIdAssigner(\n  sourceIdKey: StringOrDocFunc | null\n): (doc: DocumentInterface) => string | null {\n  if (sourceIdKey === null) {\n    return (_doc: DocumentInterface) => null;\n  } else if (typeof sourceIdKey === \"string\") {\n    return (doc: DocumentInterface) => doc.metadata[sourceIdKey];\n  } else if (typeof sourceIdKey === \"function\") {\n    return sourceIdKey;\n  } else {\n    throw new Error(\n      `sourceIdKey should be null, a string or a function, got ${typeof sourceIdKey}`\n    );\n  }\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport const _isBaseDocumentLoader = (arg: any): arg is BaseDocumentLoader => {\n  if (\n    \"load\" in arg &&\n    typeof arg.load === \"function\" &&\n    \"loadAndSplit\" in arg &&\n    typeof arg.loadAndSplit === \"function\"\n  ) {\n    return true;\n  }\n  return false;\n};\n\ninterface IndexArgs {\n  docsSource: BaseDocumentLoader | DocumentInterface[];\n  recordManager: RecordManagerInterface;\n  vectorStore: VectorStore;\n  options?: IndexOptions;\n}\n\n/**\n * Index data from the doc source into the vector store.\n *\n * Indexing functionality uses a manager to keep track of which documents\n * are in the vector store.\n *\n * This allows us to keep track of which documents were updated, and which\n * documents were deleted, which documents should be skipped.\n *\n * For the time being, documents are indexed using their hashes, and users\n *  are not able to specify the uid of the document.\n *\n * @param {IndexArgs} args\n * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.\n * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.\n * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.\n * @param {IndexOptions | undefined} args.options Options for indexing.\n * @returns {Promise<IndexingResult>}\n */\nexport async function index(args: IndexArgs): Promise<IndexingResult> {\n  const { docsSource, recordManager, vectorStore, options } = args;\n  const {\n    batchSize = 100,\n    cleanup,\n    sourceIdKey,\n    cleanupBatchSize = 1000,\n    forceUpdate = false,\n  } = options ?? {};\n\n  if (cleanup === \"incremental\" && !sourceIdKey) {\n    throw new Error(\n      \"sourceIdKey is required when cleanup mode is incremental. Please provide through 'options.sourceIdKey'.\"\n    );\n  }\n\n  const docs = _isBaseDocumentLoader(docsSource)\n    ? await docsSource.load()\n    : docsSource;\n\n  const sourceIdAssigner = _getSourceIdAssigner(sourceIdKey ?? null);\n\n  const indexStartDt = await recordManager.getTime();\n  let numAdded = 0;\n  let numDeleted = 0;\n  let numUpdated = 0;\n  let numSkipped = 0;\n\n  const batches = _batch<DocumentInterface>(batchSize ?? 100, docs);\n\n  for (const batch of batches) {\n    const hashedDocs = _deduplicateInOrder(\n      batch.map((doc) => _HashedDocument.fromDocument(doc))\n    );\n\n    const sourceIds = hashedDocs.map((doc) => sourceIdAssigner(doc));\n\n    if (cleanup === \"incremental\") {\n      hashedDocs.forEach((_hashedDoc, index) => {\n        const source = sourceIds[index];\n        if (source === null) {\n          throw new Error(\n            \"sourceIdKey must be provided when cleanup is incremental\"\n          );\n        }\n      });\n    }\n\n    const batchExists = await recordManager.exists(\n      hashedDocs.map((doc) => doc.uid)\n    );\n\n    const uids: string[] = [];\n    const docsToIndex: DocumentInterface[] = [];\n    const docsToUpdate: string[] = [];\n    const seenDocs = new Set<string>();\n    hashedDocs.forEach((hashedDoc, i) => {\n      const docExists = batchExists[i];\n      if (docExists) {\n        if (forceUpdate) {\n          seenDocs.add(hashedDoc.uid);\n        } else {\n          docsToUpdate.push(hashedDoc.uid);\n          return;\n        }\n      }\n      uids.push(hashedDoc.uid);\n      docsToIndex.push(hashedDoc.toDocument());\n    });\n\n    if (docsToUpdate.length > 0) {\n      await recordManager.update(docsToUpdate, { timeAtLeast: indexStartDt });\n      numSkipped += docsToUpdate.length;\n    }\n\n    if (docsToIndex.length > 0) {\n      await vectorStore.addDocuments(docsToIndex, { ids: uids });\n      numAdded += docsToIndex.length - seenDocs.size;\n      numUpdated += seenDocs.size;\n    }\n\n    await recordManager.update(\n      hashedDocs.map((doc) => doc.uid),\n      { timeAtLeast: indexStartDt, groupIds: sourceIds }\n    );\n\n    if (cleanup === \"incremental\") {\n      sourceIds.forEach((sourceId) => {\n        if (!sourceId) throw new Error(\"Source id cannot be null\");\n      });\n      const uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        groupIds: sourceIds,\n      });\n\n      if (uidsToDelete.length > 0) {\n        await vectorStore.delete({ ids: uidsToDelete });\n        await recordManager.deleteKeys(uidsToDelete);\n        numDeleted += uidsToDelete.length;\n      }\n    }\n  }\n\n  if (cleanup === \"full\") {\n    let uidsToDelete = await recordManager.listKeys({\n      before: indexStartDt,\n      limit: cleanupBatchSize,\n    });\n    while (uidsToDelete.length > 0) {\n      await vectorStore.delete({ ids: uidsToDelete });\n      await recordManager.deleteKeys(uidsToDelete);\n      numDeleted += uidsToDelete.length;\n      uidsToDelete = await recordManager.listKeys({\n        before: indexStartDt,\n        limit: cleanupBatchSize,\n      });\n    }\n  }\n\n  return {\n    numAdded,\n    numDeleted,\n    numUpdated,\n    numSkipped,\n  };\n}\n"],"mappings":";;;;;;;;;;;AAwCA,IAAa,kBAAb,MAAgE;CAC9D;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA,aAAqC;CAErC,YAAY,QAA4B;EACtC,KAAK,MAAM,OAAO;EAClB,KAAK,cAAc,OAAO;EAC1B,KAAK,WAAW,OAAO;CACzB;CAEA,sBAAsB,cAAoC;EACxD,KAAK,aAAa;CACpB;CAEA,kBAAwB;EACtB,MAAM,gBAAgB;GAAC;GAAS;GAAgB;EAAe;EAE/D,KAAK,MAAM,OAAO,eAChB,IAAI,OAAO,KAAK,UACd,MAAM,IAAI,MACR,+BAA+B,IAAI,yDAAyD,cAAc,KACxG,IACF,EAAE,EACJ;EAIJ,MAAM,cAAc,KAAK,kBAAkB,KAAK,WAAW;EAE3D,IAAI;GACF,MAAM,eAAe,KAAK,sBAAsB,KAAK,QAAQ;GAC7D,KAAK,cAAc;GACnB,KAAK,eAAe;EACtB,SAAS,GAAG;GACV,MAAM,IAAI,MACR,4BAA4B,EAAE,uDAChC;EACF;EAEA,KAAK,QAAQ,KAAK,kBAAkB,KAAK,cAAc,KAAK,YAAY;EAExE,IAAI,CAAC,KAAK,KACR,KAAK,MAAM,KAAK;CAEpB;CAEA,aAAgC;EAC9B,OAAO,IAAI,SAAS;GAClB,aAAa,KAAK;GAClB,UAAU,KAAK;EACjB,CAAC;CACH;CAEA,OAAO,aACL,UACA,KACiB;EACjB,MAAM,MAAM,IAAI,KAAK;GACnB,aAAa,SAAS;GACtB,UAAU,SAAS;GACnB,KAAK,OAAQ,SAAiD;EAChE,CAAC;EACD,IAAI,gBAAgB;EACpB,OAAO;CACT;CAEA,kBAA0B,aAA6B;EACrD,MAAM,aAAa,KAAK,WAAW,WAAW;EAC9C,OAAOA,GAAO,YAAY,gBAAgB;CAC5C;CAEA,sBAA8B,MAAuC;EACnE,MAAM,kBAAkB,KAAK,UAAU,MAAM,OAAO,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC;EACrE,MAAM,aAAa,KAAK,WAAW,eAAe;EAClD,OAAOA,GAAO,YAAY,gBAAgB;CAC5C;AACF;AAyCA,SAAgB,OAAU,MAAc,UAAsB;CAC5D,MAAM,UAAiB,CAAC;CACxB,IAAI,eAAoB,CAAC;CAEzB,SAAS,SAAS,SAAS;EACzB,aAAa,KAAK,IAAI;EAEtB,IAAI,aAAa,UAAU,MAAM;GAC/B,QAAQ,KAAK,YAAY;GACzB,eAAe,CAAC;EAClB;CACF,CAAC;CAED,IAAI,aAAa,SAAS,GACxB,QAAQ,KAAK,YAAY;CAG3B,OAAO;AACT;AAEA,SAAgB,oBACd,iBAC2B;CAC3B,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,eAA0C,CAAC;CAEjD,KAAK,MAAM,aAAa,iBAAiB;EACvC,IAAI,CAAC,UAAU,OACb,MAAM,IAAI,MAAM,sCAAsC;EAGxD,IAAI,CAAC,KAAK,IAAI,UAAU,KAAK,GAAG;GAC9B,KAAK,IAAI,UAAU,KAAK;GACxB,aAAa,KAAK,SAAS;EAC7B;CACF;CACA,OAAO;AACT;AAEA,SAAgB,qBACd,aAC2C;CAC3C,IAAI,gBAAgB,MAClB,QAAQ,SAA4B;MAC/B,IAAI,OAAO,gBAAgB,UAChC,QAAQ,QAA2B,IAAI,SAAS;MAC3C,IAAI,OAAO,gBAAgB,YAChC,OAAO;MAEP,MAAM,IAAI,MACR,2DAA2D,OAAO,aACpE;AAEJ;AAGA,MAAa,yBAAyB,QAAwC;CAC5E,IACE,UAAU,OACV,OAAO,IAAI,SAAS,cACpB,kBAAkB,OAClB,OAAO,IAAI,iBAAiB,YAE5B,OAAO;CAET,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AA4BA,eAAsB,MAAM,MAA0C;CACpE,MAAM,EAAE,YAAY,eAAe,aAAa,YAAY;CAC5D,MAAM,EACJ,YAAY,KACZ,SACA,aACA,mBAAmB,KACnB,cAAc,UACZ,WAAW,CAAC;CAEhB,IAAI,YAAY,iBAAiB,CAAC,aAChC,MAAM,IAAI,MACR,yGACF;CAGF,MAAM,OAAO,sBAAsB,UAAU,IACzC,MAAM,WAAW,KAAK,IACtB;CAEJ,MAAM,mBAAmB,qBAAqB,eAAe,IAAI;CAEjE,MAAM,eAAe,MAAM,cAAc,QAAQ;CACjD,IAAI,WAAW;CACf,IAAI,aAAa;CACjB,IAAI,aAAa;CACjB,IAAI,aAAa;CAEjB,MAAM,UAAU,OAA0B,aAAa,KAAK,IAAI;CAEhE,KAAK,MAAM,SAAS,SAAS;EAC3B,MAAM,aAAa,oBACjB,MAAM,KAAK,QAAQ,gBAAgB,aAAa,GAAG,CAAC,CACtD;EAEA,MAAM,YAAY,WAAW,KAAK,QAAQ,iBAAiB,GAAG,CAAC;EAE/D,IAAI,YAAY,eACd,WAAW,SAAS,YAAY,UAAU;GAExC,IADe,UAAU,WACV,MACb,MAAM,IAAI,MACR,0DACF;EAEJ,CAAC;EAGH,MAAM,cAAc,MAAM,cAAc,OACtC,WAAW,KAAK,QAAQ,IAAI,GAAG,CACjC;EAEA,MAAM,OAAiB,CAAC;EACxB,MAAM,cAAmC,CAAC;EAC1C,MAAM,eAAyB,CAAC;EAChC,MAAM,2BAAW,IAAI,IAAY;EACjC,WAAW,SAAS,WAAW,MAAM;GAEnC,IADkB,YAAY,IACf;IACb,IAAI,aACF,SAAS,IAAI,UAAU,GAAG;SACrB;KACL,aAAa,KAAK,UAAU,GAAG;KAC/B;IACF;GACF;GACA,KAAK,KAAK,UAAU,GAAG;GACvB,YAAY,KAAK,UAAU,WAAW,CAAC;EACzC,CAAC;EAED,IAAI,aAAa,SAAS,GAAG;GAC3B,MAAM,cAAc,OAAO,cAAc,EAAE,aAAa,aAAa,CAAC;GACtE,cAAc,aAAa;EAC7B;EAEA,IAAI,YAAY,SAAS,GAAG;GAC1B,MAAM,YAAY,aAAa,aAAa,EAAE,KAAK,KAAK,CAAC;GACzD,YAAY,YAAY,SAAS,SAAS;GAC1C,cAAc,SAAS;EACzB;EAEA,MAAM,cAAc,OAClB,WAAW,KAAK,QAAQ,IAAI,GAAG,GAC/B;GAAE,aAAa;GAAc,UAAU;EAAU,CACnD;EAEA,IAAI,YAAY,eAAe;GAC7B,UAAU,SAAS,aAAa;IAC9B,IAAI,CAAC,UAAU,MAAM,IAAI,MAAM,0BAA0B;GAC3D,CAAC;GACD,MAAM,eAAe,MAAM,cAAc,SAAS;IAChD,QAAQ;IACR,UAAU;GACZ,CAAC;GAED,IAAI,aAAa,SAAS,GAAG;IAC3B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;IAC9C,MAAM,cAAc,WAAW,YAAY;IAC3C,cAAc,aAAa;GAC7B;EACF;CACF;CAEA,IAAI,YAAY,QAAQ;EACtB,IAAI,eAAe,MAAM,cAAc,SAAS;GAC9C,QAAQ;GACR,OAAO;EACT,CAAC;EACD,OAAO,aAAa,SAAS,GAAG;GAC9B,MAAM,YAAY,OAAO,EAAE,KAAK,aAAa,CAAC;GAC9C,MAAM,cAAc,WAAW,YAAY;GAC3C,cAAc,aAAa;GAC3B,eAAe,MAAM,cAAc,SAAS;IAC1C,QAAQ;IACR,OAAO;GACT,CAAC;EACH;CACF;CAEA,OAAO;EACL;EACA;EACA;EACA;CACF;AACF"}

package/dist/language_models/base.cjs

@@ -139,7 +139,7 @@ var BaseLangChain = class extends require_base.Runnable {
 		this.callbacks = params.callbacks;
 		this.tags = params.tags ?? [];
 		this.metadata = params.metadata ?? {};
-		this._addVersion("@langchain/core", "1.2.9");
+		this._addVersion("@langchain/core", "1.2.11");
 	}
 	_addVersion(pkg, version) {
 		const existing = this.metadata?.versions;

package/dist/language_models/base.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"base.cjs","names":["encodingForModel","Runnable","InMemoryCache","AsyncCaller","StringPromptValue","ChatPromptValue","coerceMessageLikeToMessage"],"sources":["../../src/language_models/base.ts"],"sourcesContent":["import type { Tiktoken, TiktokenModel } from \"js-tiktoken/lite\";\nimport type { ZodV3Like, ZodV4Like } from \"../utils/types/zod.js\";\n\nimport { type BaseCache, InMemoryCache } from \"../caches/index.js\";\nimport {\n  type BasePromptValueInterface,\n  StringPromptValue,\n  ChatPromptValue,\n} from \"../prompt_values.js\";\nimport {\n  type BaseMessage,\n  type BaseMessageLike,\n  type MessageContent,\n} from \"../messages/base.js\";\nimport { coerceMessageLikeToMessage } from \"../messages/utils.js\";\nimport { type LLMResult } from \"../outputs.js\";\nimport { CallbackManager, Callbacks } from \"../callbacks/manager.js\";\nimport { AsyncCaller, AsyncCallerParams } from \"../utils/async_caller.js\";\nimport { encodingForModel } from \"../utils/tiktoken.js\";\nimport { Runnable, type RunnableInterface } from \"../runnables/base.js\";\nimport { RunnableConfig } from \"../runnables/config.js\";\nimport { JSONSchema } from \"../utils/json_schema.js\";\nimport {\n  InferInteropZodOutput,\n  InteropZodObject,\n  InteropZodType,\n} from \"../utils/types/zod.js\";\nimport { ModelProfile } from \"./profile.js\";\nimport { type SerializableSchema } from \"../utils/standard_schema.js\";\n\n// https://www.npmjs.com/package/js-tiktoken\n\nexport const getModelNameForTiktoken = (modelName: string): TiktokenModel => {\n  if (modelName.startsWith(\"gpt-5\")) {\n    return \"gpt-5\" as TiktokenModel;\n  }\n\n  if (modelName.startsWith(\"gpt-3.5-turbo-16k\")) {\n    return \"gpt-3.5-turbo-16k\";\n  }\n\n  if (modelName.startsWith(\"gpt-3.5-turbo-\")) {\n    return \"gpt-3.5-turbo\";\n  }\n\n  if (modelName.startsWith(\"gpt-4-32k\")) {\n    return \"gpt-4-32k\";\n  }\n\n  if (modelName.startsWith(\"gpt-4-\")) {\n    return \"gpt-4\";\n  }\n\n  if (modelName.startsWith(\"gpt-4o\")) {\n    return \"gpt-4o\";\n  }\n\n  return modelName as TiktokenModel;\n};\n\nexport const getEmbeddingContextSize = (modelName?: string): number => {\n  switch (modelName) {\n    case \"text-embedding-ada-002\":\n      return 8191;\n    default:\n      return 2046;\n  }\n};\n\n/**\n * Get the context window size (max input tokens) for a given model.\n *\n * Context window sizes are sourced from official model documentation:\n * - OpenAI: https://platform.openai.com/docs/models\n * - Anthropic: https://docs.anthropic.com/claude/docs/models-overview\n * - Google: https://ai.google.dev/gemini/docs/models/gemini\n *\n * @param modelName - The name of the model\n * @returns The context window size in tokens\n */\nexport const getModelContextSize = (modelName: string): number => {\n  const normalizedName = getModelNameForTiktoken(modelName) as string;\n\n  switch (normalizedName) {\n    // GPT-5 series\n    case \"gpt-5\":\n    case \"gpt-5-turbo\":\n    case \"gpt-5-turbo-preview\":\n      return 400000;\n\n    // GPT-4o series\n    case \"gpt-4o\":\n    case \"gpt-4o-mini\":\n    case \"gpt-4o-2024-05-13\":\n    case \"gpt-4o-2024-08-06\":\n      return 128000;\n\n    // GPT-4 Turbo series\n    case \"gpt-4-turbo\":\n    case \"gpt-4-turbo-preview\":\n    case \"gpt-4-turbo-2024-04-09\":\n    case \"gpt-4-0125-preview\":\n    case \"gpt-4-1106-preview\":\n      return 128000;\n\n    // GPT-4 series\n    case \"gpt-4-32k\":\n    case \"gpt-4-32k-0314\":\n    case \"gpt-4-32k-0613\":\n      return 32768;\n    case \"gpt-4\":\n    case \"gpt-4-0314\":\n    case \"gpt-4-0613\":\n      return 8192;\n\n    // GPT-3.5 Turbo series\n    case \"gpt-3.5-turbo-16k\":\n    case \"gpt-3.5-turbo-16k-0613\":\n      return 16384;\n    case \"gpt-3.5-turbo\":\n    case \"gpt-3.5-turbo-0301\":\n    case \"gpt-3.5-turbo-0613\":\n    case \"gpt-3.5-turbo-1106\":\n    case \"gpt-3.5-turbo-0125\":\n      return 4096;\n\n    // Legacy GPT-3 models\n    case \"text-davinci-003\":\n    case \"text-davinci-002\":\n      return 4097;\n    case \"text-davinci-001\":\n      return 2049;\n    case \"text-curie-001\":\n    case \"text-babbage-001\":\n    case \"text-ada-001\":\n      return 2048;\n\n    // Code models\n    case \"code-davinci-002\":\n    case \"code-davinci-001\":\n      return 8000;\n    case \"code-cushman-001\":\n      return 2048;\n\n    // Claude models (Anthropic)\n    case \"claude-3-5-sonnet-20241022\":\n    case \"claude-3-5-sonnet-20240620\":\n    case \"claude-3-opus-20240229\":\n    case \"claude-3-sonnet-20240229\":\n    case \"claude-3-haiku-20240307\":\n    case \"claude-2.1\":\n      return 200000;\n    case \"claude-2.0\":\n    case \"claude-instant-1.2\":\n      return 100000;\n\n    // Gemini models (Google)\n    case \"gemini-1.5-pro\":\n    case \"gemini-1.5-pro-latest\":\n    case \"gemini-1.5-flash\":\n    case \"gemini-1.5-flash-latest\":\n      return 1000000; // 1M tokens\n    case \"gemini-pro\":\n    case \"gemini-pro-vision\":\n      return 32768;\n\n    default:\n      return 4097;\n  }\n};\n\n/**\n * Whether or not the input matches the OpenAI tool definition.\n * @param {unknown} tool The input to check.\n * @returns {boolean} Whether the input is an OpenAI tool definition.\n */\nexport function isOpenAITool(tool: unknown): tool is ToolDefinition {\n  if (typeof tool !== \"object\" || !tool) return false;\n  if (\n    \"type\" in tool &&\n    tool.type === \"function\" &&\n    \"function\" in tool &&\n    typeof tool.function === \"object\" &&\n    tool.function &&\n    \"name\" in tool.function &&\n    \"parameters\" in tool.function\n  ) {\n    return true;\n  }\n  return false;\n}\n\ninterface CalculateMaxTokenProps {\n  prompt: string;\n  modelName: TiktokenModel;\n}\n\nexport const calculateMaxTokens = async ({\n  prompt,\n  modelName,\n}: CalculateMaxTokenProps) => {\n  let numTokens;\n\n  try {\n    numTokens = (\n      await encodingForModel(getModelNameForTiktoken(modelName))\n    ).encode(prompt).length;\n  } catch {\n    console.warn(\n      \"Failed to calculate number of tokens, falling back to approximate count\"\n    );\n\n    // fallback to approximate calculation if tiktoken is not available\n    // each token is ~4 characters: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them#\n    numTokens = Math.ceil(prompt.length / 4);\n  }\n\n  const maxTokens = getModelContextSize(modelName);\n  return maxTokens - numTokens;\n};\n\nconst getVerbosity = () => false;\n\nexport type SerializedLLM = {\n  _model: string;\n  _type: string;\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n} & Record<string, any>;\n\nexport interface BaseLangChainParams {\n  verbose?: boolean;\n  callbacks?: Callbacks;\n  tags?: string[];\n  metadata?: Record<string, unknown>;\n}\n\n/**\n * Base class for language models, chains, tools.\n */\nexport abstract class BaseLangChain<\n  RunInput,\n  RunOutput,\n  CallOptions extends RunnableConfig = RunnableConfig,\n>\n  extends Runnable<RunInput, RunOutput, CallOptions>\n  implements BaseLangChainParams\n{\n  /**\n   * Whether to print out response text.\n   */\n  verbose: boolean;\n\n  callbacks?: Callbacks;\n\n  tags?: string[];\n\n  metadata?: Record<string, unknown>;\n\n  get lc_attributes(): { [key: string]: undefined } | undefined {\n    return {\n      callbacks: undefined,\n      verbose: undefined,\n    };\n  }\n\n  constructor(params: BaseLangChainParams) {\n    super(params);\n    this.verbose = params.verbose ?? getVerbosity();\n    this.callbacks = params.callbacks;\n    this.tags = params.tags ?? [];\n    this.metadata = params.metadata ?? {};\n    this._addVersion(\"@langchain/core\", __PKG_VERSION__);\n  }\n\n  protected _addVersion(pkg: string, version: string) {\n    const existing = this.metadata?.versions;\n    this.metadata = {\n      ...this.metadata,\n      versions: {\n        ...(typeof existing === \"object\" && existing !== null ? existing : {}),\n        [pkg]: version,\n      },\n    };\n  }\n}\n\n/**\n * Base interface for language model parameters.\n * A subclass of {@link BaseLanguageModel} should have a constructor that\n * takes in a parameter that extends this interface.\n */\nexport interface BaseLanguageModelParams\n  extends AsyncCallerParams, BaseLangChainParams {\n  /**\n   * @deprecated Use `callbacks` instead\n   */\n  callbackManager?: CallbackManager;\n\n  cache?: BaseCache | boolean;\n}\n\nexport interface BaseLanguageModelTracingCallOptions {\n  /**\n   * Describes the format of structured outputs.\n   * This should be provided if an output is considered to be structured\n   */\n  ls_structured_output_format?: {\n    /**\n     * An object containing the method used for structured output (e.g., \"jsonMode\").\n     */\n    kwargs: { method: string };\n    /**\n     * The JSON schema describing the expected output structure.\n     */\n    schema?: JSONSchema;\n  };\n}\n\nexport interface BaseLanguageModelCallOptions\n  extends RunnableConfig, BaseLanguageModelTracingCallOptions {\n  /**\n   * Stop tokens to use for this call.\n   * If not provided, the default stop tokens for the model will be used.\n   */\n  stop?: string[];\n  /**\n   * Overrides the model's configured `maxRetries` for this call only.\n   */\n  maxRetries?: number;\n}\n\nexport interface FunctionDefinition {\n  /**\n   * The name of the function to be called. Must be a-z, A-Z, 0-9, or contain\n   * underscores and dashes, with a maximum length of 64.\n   */\n  name: string;\n\n  /**\n   * The parameters the functions accepts, described as a JSON Schema object. See the\n   * [guide](https://platform.openai.com/docs/guides/gpt/function-calling) for\n   * examples, and the\n   * [JSON Schema reference](https://json-schema.org/understanding-json-schema/) for\n   * documentation about the format.\n   *\n   * To describe a function that accepts no parameters, provide the value\n   * `{\"type\": \"object\", \"properties\": {}}`.\n   */\n  parameters: Record<string, unknown> | JSONSchema;\n\n  /**\n   * A description of what the function does, used by the model to choose when and\n   * how to call the function.\n   */\n  description?: string;\n}\n\nexport interface ToolDefinition {\n  type: \"function\";\n  function: FunctionDefinition;\n}\n\nexport type FunctionCallOption = {\n  name: string;\n};\n\nexport interface BaseFunctionCallOptions extends BaseLanguageModelCallOptions {\n  function_call?: FunctionCallOption;\n  functions?: FunctionDefinition[];\n}\n\nexport type BaseLanguageModelInput =\n  | BasePromptValueInterface\n  | string\n  | BaseMessageLike[];\n\nexport type StructuredOutputType = InferInteropZodOutput<InteropZodObject>;\n\nexport type StructuredOutputMethodOptions<IncludeRaw extends boolean = false> =\n  {\n    name?: string;\n    method?: \"functionCalling\" | \"jsonMode\" | \"jsonSchema\" | string;\n    includeRaw?: IncludeRaw;\n    /** Whether to use strict mode. Currently only supported by OpenAI models. */\n    strict?: boolean;\n  };\n\n/** @deprecated Use StructuredOutputMethodOptions instead */\nexport type StructuredOutputMethodParams<\n  RunOutput,\n  IncludeRaw extends boolean = false,\n> = {\n  /** @deprecated Pass schema in as the first argument */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  schema: InteropZodType<RunOutput> | Record<string, any>;\n  name?: string;\n  method?: \"functionCalling\" | \"jsonMode\";\n  includeRaw?: IncludeRaw;\n};\n\nexport interface BaseLanguageModelInterface<\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  RunOutput = any,\n  CallOptions extends BaseLanguageModelCallOptions =\n    BaseLanguageModelCallOptions,\n> extends RunnableInterface<BaseLanguageModelInput, RunOutput, CallOptions> {\n  get callKeys(): string[];\n\n  generatePrompt(\n    promptValues: BasePromptValueInterface[],\n    options?: string[] | Partial<CallOptions>,\n    callbacks?: Callbacks\n  ): Promise<LLMResult>;\n\n  _modelType(): string;\n\n  _llmType(): string;\n\n  getNumTokens(content: MessageContent): Promise<number>;\n\n  /**\n   * Get the identifying parameters of the LLM.\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  _identifyingParams(): Record<string, any>;\n\n  serialize(): SerializedLLM;\n}\n\nexport type LanguageModelOutput = BaseMessage | string;\n\nexport type LanguageModelLike = RunnableInterface<\n  BaseLanguageModelInput,\n  LanguageModelOutput\n>;\n\n/**\n * Base class for language models.\n */\nexport abstract class BaseLanguageModel<\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  RunOutput = any,\n  CallOptions extends BaseLanguageModelCallOptions =\n    BaseLanguageModelCallOptions,\n>\n  extends BaseLangChain<BaseLanguageModelInput, RunOutput, CallOptions>\n  implements\n    BaseLanguageModelParams,\n    BaseLanguageModelInterface<RunOutput, CallOptions>\n{\n  /**\n   * Keys that the language model accepts as call options.\n   */\n  get callKeys(): string[] {\n    return [\"stop\", \"timeout\", \"signal\", \"tags\", \"metadata\", \"callbacks\"];\n  }\n\n  /**\n   * The async caller should be used by subclasses to make any async calls,\n   * which will thus benefit from the concurrency and retry logic.\n   */\n  caller: AsyncCaller;\n\n  cache?: BaseCache;\n\n  constructor({\n    callbacks,\n    callbackManager,\n    ...params\n  }: BaseLanguageModelParams) {\n    const { cache, ...rest } = params;\n    super({\n      callbacks: callbacks ?? callbackManager,\n      ...rest,\n    });\n    if (typeof cache === \"object\") {\n      this.cache = cache;\n    } else if (cache) {\n      this.cache = InMemoryCache.global();\n    } else {\n      this.cache = undefined;\n    }\n    this.caller = new AsyncCaller(params ?? {});\n  }\n\n  abstract generatePrompt(\n    promptValues: BasePromptValueInterface[],\n    options?: string[] | CallOptions,\n    callbacks?: Callbacks\n  ): Promise<LLMResult>;\n\n  abstract _modelType(): string;\n\n  abstract _llmType(): string;\n\n  private _encoding?: Tiktoken;\n\n  /**\n   * Get the number of tokens in the content.\n   * @param content The content to get the number of tokens for.\n   * @returns The number of tokens in the content.\n   */\n  async getNumTokens(content: MessageContent) {\n    // Extract text content from MessageContent\n    let textContent: string;\n    if (typeof content === \"string\") {\n      textContent = content;\n    } else {\n      /**\n       * Content is an array of ContentBlock\n       *\n       * ToDo(@christian-bromann): This is a temporary fix to get the number of tokens for the content.\n       * We need to find a better way to do this.\n       * @see https://github.com/langchain-ai/langchainjs/pull/8341#pullrequestreview-2933713116\n       */\n      textContent = content\n        .map((item) => {\n          if (typeof item === \"string\") return item;\n          if (item.type === \"text\" && \"text\" in item) return item.text;\n          return \"\";\n        })\n        .join(\"\");\n    }\n\n    // fallback to approximate calculation if tiktoken is not available\n    let numTokens = Math.ceil(textContent.length / 4);\n\n    if (!this._encoding) {\n      try {\n        this._encoding = await encodingForModel(\n          \"modelName\" in this\n            ? getModelNameForTiktoken(this.modelName as string)\n            : \"gpt2\"\n        );\n      } catch (error) {\n        console.warn(\n          \"Failed to calculate number of tokens, falling back to approximate count\",\n          error\n        );\n      }\n    }\n\n    if (this._encoding) {\n      try {\n        numTokens = this._encoding.encode(textContent).length;\n      } catch (error) {\n        console.warn(\n          \"Failed to calculate number of tokens, falling back to approximate count\",\n          error\n        );\n      }\n    }\n\n    return numTokens;\n  }\n\n  protected static _convertInputToPromptValue(\n    input: BaseLanguageModelInput\n  ): BasePromptValueInterface {\n    if (typeof input === \"string\") {\n      return new StringPromptValue(input);\n    } else if (Array.isArray(input)) {\n      return new ChatPromptValue(input.map(coerceMessageLikeToMessage));\n    } else {\n      return input;\n    }\n  }\n\n  /**\n   * Get the identifying parameters of the LLM.\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  _identifyingParams(): Record<string, any> {\n    return {};\n  }\n\n  /**\n   * Create a unique cache key for a specific call to a specific language model.\n   * @param callOptions Call options for the model\n   * @returns A unique cache key.\n   */\n  _getSerializedCacheKeyParametersForCall(\n    // TODO: Fix when we remove the RunnableLambda backwards compatibility shim.\n    {\n      config,\n      ...callOptions\n    }: CallOptions & { config?: RunnableConfig }\n  ): string {\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    const params: Record<string, any> = {\n      ...this._identifyingParams(),\n      ...callOptions,\n      _type: this._llmType(),\n      _model: this._modelType(),\n    };\n    const filteredEntries = Object.entries(params).filter(\n      ([_, value]) => value !== undefined\n    );\n    const serializedEntries = filteredEntries\n      .map(([key, value]) => `${key}:${JSON.stringify(value)}`)\n      .sort()\n      .join(\",\");\n    return serializedEntries;\n  }\n\n  /**\n   * @deprecated\n   * Return a json-like object representing this LLM.\n   */\n  serialize(): SerializedLLM {\n    return {\n      ...this._identifyingParams(),\n      _type: this._llmType(),\n      _model: this._modelType(),\n    };\n  }\n\n  /**\n   * @deprecated\n   * Load an LLM from a json-like object describing it.\n   */\n  static async deserialize(_data: SerializedLLM): Promise<BaseLanguageModel> {\n    throw new Error(\"Use .toJSON() instead\");\n  }\n\n  /**\n   * Return profiling information for the model.\n   *\n   * @returns {ModelProfile} An object describing the model's capabilities and constraints\n   */\n  get profile(): ModelProfile {\n    return {};\n  }\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema: SerializableSchema<RunOutput>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema: SerializableSchema<RunOutput>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV3Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV3Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV4Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV4Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  /**\n   * Model wrapper that returns outputs formatted to match the given schema.\n   *\n   * @template {BaseLanguageModelInput} RunInput The input type for the Runnable, expected to be the same input for the LLM.\n   * @template {Record<string, any>} RunOutput The output type for the Runnable, expected to be a Zod schema object for structured output validation.\n   *\n   * @param {InteropZodType<RunOutput>} schema The schema for the structured output. Either as a Zod schema or a valid JSON schema object.\n   *   If a Zod schema is passed, the returned attributes will be validated, whereas with JSON schema they will not be.\n   * @param {string} name The name of the function to call.\n   * @param {\"functionCalling\" | \"jsonMode\"} [method=functionCalling] The method to use for getting the structured output. Defaults to \"functionCalling\".\n   * @param {boolean | undefined} [includeRaw=false] Whether to include the raw output in the result. Defaults to false.\n   * @returns {Runnable<RunInput, RunOutput> | Runnable<RunInput, { raw: BaseMessage; parsed: RunOutput }>} A new runnable that calls the LLM with structured output.\n   */\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | InteropZodType<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<boolean>\n  ):\n    | Runnable<BaseLanguageModelInput, RunOutput>\n    | Runnable<\n        BaseLanguageModelInput,\n        {\n          raw: BaseMessage;\n          parsed: RunOutput;\n        }\n      >;\n\n  /**\n   * Filter out large/inappropriate fields from invocation params for tracing metadata.\n   * Removes fields like tools, functions, messages, response_format that can be large.\n   */\n  protected _filterInvocationParamsForTracing(\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    params: Record<string, any>\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  ): Record<string, any> {\n    const { tools, functions, messages, response_format, ...rest } = params;\n    return rest;\n  }\n}\n\n/**\n * Shared interface for token usage\n * return type from LLM calls.\n */\nexport interface TokenUsage {\n  completionTokens?: number;\n  promptTokens?: number;\n  totalTokens?: number;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAgCA,MAAa,2BAA2B,cAAqC;CAC3E,IAAI,UAAU,WAAW,OAAO,GAC9B,OAAO;CAGT,IAAI,UAAU,WAAW,mBAAmB,GAC1C,OAAO;CAGT,IAAI,UAAU,WAAW,gBAAgB,GACvC,OAAO;CAGT,IAAI,UAAU,WAAW,WAAW,GAClC,OAAO;CAGT,IAAI,UAAU,WAAW,QAAQ,GAC/B,OAAO;CAGT,IAAI,UAAU,WAAW,QAAQ,GAC/B,OAAO;CAGT,OAAO;AACT;AAEA,MAAa,2BAA2B,cAA+B;CACrE,QAAQ,WAAR;EACE,KAAK,0BACH,OAAO;EACT,SACE,OAAO;CACX;AACF;;;;;;;;;;;;AAaA,MAAa,uBAAuB,cAA8B;CAGhE,QAFuB,wBAAwB,SAE1B,GAArB;EAEE,KAAK;EACL,KAAK;EACL,KAAK,uBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,qBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK,kBACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,cACH,OAAO;EAGT,KAAK;EACL,KAAK,0BACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK,oBACH,OAAO;EACT,KAAK,oBACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,gBACH,OAAO;EAGT,KAAK;EACL,KAAK,oBACH,OAAO;EACT,KAAK,oBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,cACH,OAAO;EACT,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,2BACH,OAAO;EACT,KAAK;EACL,KAAK,qBACH,OAAO;EAET,SACE,OAAO;CACX;AACF;;;;;;AAOA,SAAgB,aAAa,MAAuC;CAClE,IAAI,OAAO,SAAS,YAAY,CAAC,MAAM,OAAO;CAC9C,IACE,UAAU,QACV,KAAK,SAAS,cACd,cAAc,QACd,OAAO,KAAK,aAAa,YACzB,KAAK,YACL,UAAU,KAAK,YACf,gBAAgB,KAAK,UAErB,OAAO;CAET,OAAO;AACT;AAOA,MAAa,qBAAqB,OAAO,EACvC,QACA,gBAC4B;CAC5B,IAAI;CAEJ,IAAI;EACF,aACE,MAAMA,uBAAAA,iBAAiB,wBAAwB,SAAS,CAAC,EAAA,CACzD,OAAO,MAAM,CAAC,CAAC;CACnB,QAAQ;EACN,QAAQ,KACN,yEACF;EAIA,YAAY,KAAK,KAAK,OAAO,SAAS,CAAC;CACzC;CAGA,OADkB,oBAAoB,SACvB,IAAI;AACrB;AAEA,MAAM,qBAAqB;;;;AAkB3B,IAAsB,gBAAtB,cAKUC,aAAAA,SAEV;;;;CAIE;CAEA;CAEA;CAEA;CAEA,IAAI,gBAA0D;EAC5D,OAAO;GACL,WAAW,KAAA;GACX,SAAS,KAAA;EACX;CACF;CAEA,YAAY,QAA6B;EACvC,MAAM,MAAM;EACZ,KAAK,UAAU,OAAO,WAAW,aAAa;EAC9C,KAAK,YAAY,OAAO;EACxB,KAAK,OAAO,OAAO,QAAQ,CAAC;EAC5B,KAAK,WAAW,OAAO,YAAY,CAAC;EACpC,KAAK,YAAY,mBAAA,OAAkC;CACrD;CAEA,YAAsB,KAAa,SAAiB;EAClD,MAAM,WAAW,KAAK,UAAU;EAChC,KAAK,WAAW;GACd,GAAG,KAAK;GACR,UAAU;IACR,GAAI,OAAO,aAAa,YAAY,aAAa,OAAO,WAAW,CAAC;KACnE,MAAM;GACT;EACF;CACF;AACF;;;;AA2JA,IAAsB,oBAAtB,cAMU,cAIV;;;;CAIE,IAAI,WAAqB;EACvB,OAAO;GAAC;GAAQ;GAAW;GAAU;GAAQ;GAAY;EAAW;CACtE;;;;;CAMA;CAEA;CAEA,YAAY,EACV,WACA,iBACA,GAAG,UACuB;EAC1B,MAAM,EAAE,OAAO,GAAG,SAAS;EAC3B,MAAM;GACJ,WAAW,aAAa;GACxB,GAAG;EACL,CAAC;EACD,IAAI,OAAO,UAAU,UACnB,KAAK,QAAQ;OACR,IAAI,OACT,KAAK,QAAQC,qBAAAA,cAAc,OAAO;OAElC,KAAK,QAAQ,KAAA;EAEf,KAAK,SAAS,IAAIC,2BAAAA,YAAY,UAAU,CAAC,CAAC;CAC5C;CAYA;;;;;;CAOA,MAAM,aAAa,SAAyB;EAE1C,IAAI;EACJ,IAAI,OAAO,YAAY,UACrB,cAAc;;;;;;;;;EASd,cAAc,QACX,KAAK,SAAS;GACb,IAAI,OAAO,SAAS,UAAU,OAAO;GACrC,IAAI,KAAK,SAAS,UAAU,UAAU,MAAM,OAAO,KAAK;GACxD,OAAO;EACT,CAAC,CAAC,CACD,KAAK,EAAE;EAIZ,IAAI,YAAY,KAAK,KAAK,YAAY,SAAS,CAAC;EAEhD,IAAI,CAAC,KAAK,WACR,IAAI;GACF,KAAK,YAAY,MAAMH,uBAAAA,iBACrB,eAAe,OACX,wBAAwB,KAAK,SAAmB,IAChD,MACN;EACF,SAAS,OAAO;GACd,QAAQ,KACN,2EACA,KACF;EACF;EAGF,IAAI,KAAK,WACP,IAAI;GACF,YAAY,KAAK,UAAU,OAAO,WAAW,CAAC,CAAC;EACjD,SAAS,OAAO;GACd,QAAQ,KACN,2EACA,KACF;EACF;EAGF,OAAO;CACT;CAEA,OAAiB,2BACf,OAC0B;EAC1B,IAAI,OAAO,UAAU,UACnB,OAAO,IAAII,sBAAAA,kBAAkB,KAAK;OAC7B,IAAI,MAAM,QAAQ,KAAK,GAC5B,OAAO,IAAIC,sBAAAA,gBAAgB,MAAM,IAAIC,cAAAA,0BAA0B,CAAC;OAEhE,OAAO;CAEX;;;;CAMA,qBAA0C;EACxC,OAAO,CAAC;CACV;;;;;;CAOA,wCAEE,EACE,QACA,GAAG,eAEG;EAER,MAAM,SAA8B;GAClC,GAAG,KAAK,mBAAmB;GAC3B,GAAG;GACH,OAAO,KAAK,SAAS;GACrB,QAAQ,KAAK,WAAW;EAC1B;EAQA,OAPwB,OAAO,QAAQ,MAAM,CAAC,CAAC,QAC5C,CAAC,GAAG,WAAW,UAAU,KAAA,CAEY,CAAC,CACtC,KAAK,CAAC,KAAK,WAAW,GAAG,IAAI,GAAG,KAAK,UAAU,KAAK,GAAG,CAAC,CACxD,KAAK,CAAC,CACN,KAAK,GACe;CACzB;;;;;CAMA,YAA2B;EACzB,OAAO;GACL,GAAG,KAAK,mBAAmB;GAC3B,OAAO,KAAK,SAAS;GACrB,QAAQ,KAAK,WAAW;EAC1B;CACF;;;;;CAMA,aAAa,YAAY,OAAkD;EACzE,MAAM,IAAI,MAAM,uBAAuB;CACzC;;;;;;CAOA,IAAI,UAAwB;EAC1B,OAAO,CAAC;CACV;;;;;CAkGA,kCAEE,QAEqB;EACrB,MAAM,EAAE,OAAO,WAAW,UAAU,iBAAiB,GAAG,SAAS;EACjE,OAAO;CACT;AACF"}
+{"version":3,"file":"base.cjs","names":["encodingForModel","Runnable","InMemoryCache","AsyncCaller","StringPromptValue","ChatPromptValue","coerceMessageLikeToMessage"],"sources":["../../src/language_models/base.ts"],"sourcesContent":["import type { Tiktoken, TiktokenModel } from \"js-tiktoken/lite\";\nimport type { ZodV3Like, ZodV4Like } from \"../utils/types/zod.js\";\n\nimport { type BaseCache, InMemoryCache } from \"../caches/index.js\";\nimport {\n  type BasePromptValueInterface,\n  StringPromptValue,\n  ChatPromptValue,\n} from \"../prompt_values.js\";\nimport {\n  type BaseMessage,\n  type BaseMessageLike,\n  type MessageContent,\n} from \"../messages/base.js\";\nimport { coerceMessageLikeToMessage } from \"../messages/utils.js\";\nimport { type LLMResult } from \"../outputs.js\";\nimport { CallbackManager, Callbacks } from \"../callbacks/manager.js\";\nimport { AsyncCaller, AsyncCallerParams } from \"../utils/async_caller.js\";\nimport { encodingForModel } from \"../utils/tiktoken.js\";\nimport { Runnable, type RunnableInterface } from \"../runnables/base.js\";\nimport { RunnableConfig } from \"../runnables/config.js\";\nimport { JSONSchema } from \"../utils/json_schema.js\";\nimport {\n  InferInteropZodOutput,\n  InteropZodObject,\n  InteropZodType,\n} from \"../utils/types/zod.js\";\nimport { ModelProfile } from \"./profile.js\";\nimport { type SerializableSchema } from \"../utils/standard_schema.js\";\n\n// https://www.npmjs.com/package/js-tiktoken\n\nexport const getModelNameForTiktoken = (modelName: string): TiktokenModel => {\n  if (modelName.startsWith(\"gpt-5\")) {\n    return \"gpt-5\" as TiktokenModel;\n  }\n\n  if (modelName.startsWith(\"gpt-3.5-turbo-16k\")) {\n    return \"gpt-3.5-turbo-16k\";\n  }\n\n  if (modelName.startsWith(\"gpt-3.5-turbo-\")) {\n    return \"gpt-3.5-turbo\";\n  }\n\n  if (modelName.startsWith(\"gpt-4-32k\")) {\n    return \"gpt-4-32k\";\n  }\n\n  if (modelName.startsWith(\"gpt-4-\")) {\n    return \"gpt-4\";\n  }\n\n  if (modelName.startsWith(\"gpt-4o\")) {\n    return \"gpt-4o\";\n  }\n\n  return modelName as TiktokenModel;\n};\n\nexport const getEmbeddingContextSize = (modelName?: string): number => {\n  switch (modelName) {\n    case \"text-embedding-ada-002\":\n      return 8191;\n    default:\n      return 2046;\n  }\n};\n\n/**\n * Get the context window size (max input tokens) for a given model.\n *\n * Context window sizes are sourced from official model documentation:\n * - OpenAI: https://platform.openai.com/docs/models\n * - Anthropic: https://docs.anthropic.com/claude/docs/models-overview\n * - Google: https://ai.google.dev/gemini/docs/models/gemini\n *\n * @param modelName - The name of the model\n * @returns The context window size in tokens\n */\nexport const getModelContextSize = (modelName: string): number => {\n  const normalizedName = getModelNameForTiktoken(modelName) as string;\n\n  switch (normalizedName) {\n    // GPT-5 series\n    case \"gpt-5\":\n    case \"gpt-5-turbo\":\n    case \"gpt-5-turbo-preview\":\n      return 400000;\n\n    // GPT-4o series\n    case \"gpt-4o\":\n    case \"gpt-4o-mini\":\n    case \"gpt-4o-2024-05-13\":\n    case \"gpt-4o-2024-08-06\":\n      return 128000;\n\n    // GPT-4 Turbo series\n    case \"gpt-4-turbo\":\n    case \"gpt-4-turbo-preview\":\n    case \"gpt-4-turbo-2024-04-09\":\n    case \"gpt-4-0125-preview\":\n    case \"gpt-4-1106-preview\":\n      return 128000;\n\n    // GPT-4 series\n    case \"gpt-4-32k\":\n    case \"gpt-4-32k-0314\":\n    case \"gpt-4-32k-0613\":\n      return 32768;\n    case \"gpt-4\":\n    case \"gpt-4-0314\":\n    case \"gpt-4-0613\":\n      return 8192;\n\n    // GPT-3.5 Turbo series\n    case \"gpt-3.5-turbo-16k\":\n    case \"gpt-3.5-turbo-16k-0613\":\n      return 16384;\n    case \"gpt-3.5-turbo\":\n    case \"gpt-3.5-turbo-0301\":\n    case \"gpt-3.5-turbo-0613\":\n    case \"gpt-3.5-turbo-1106\":\n    case \"gpt-3.5-turbo-0125\":\n      return 4096;\n\n    // Legacy GPT-3 models\n    case \"text-davinci-003\":\n    case \"text-davinci-002\":\n      return 4097;\n    case \"text-davinci-001\":\n      return 2049;\n    case \"text-curie-001\":\n    case \"text-babbage-001\":\n    case \"text-ada-001\":\n      return 2048;\n\n    // Code models\n    case \"code-davinci-002\":\n    case \"code-davinci-001\":\n      return 8000;\n    case \"code-cushman-001\":\n      return 2048;\n\n    // Claude models (Anthropic)\n    case \"claude-3-5-sonnet-20241022\":\n    case \"claude-3-5-sonnet-20240620\":\n    case \"claude-3-opus-20240229\":\n    case \"claude-3-sonnet-20240229\":\n    case \"claude-3-haiku-20240307\":\n    case \"claude-2.1\":\n      return 200000;\n    case \"claude-2.0\":\n    case \"claude-instant-1.2\":\n      return 100000;\n\n    // Gemini models (Google)\n    case \"gemini-1.5-pro\":\n    case \"gemini-1.5-pro-latest\":\n    case \"gemini-1.5-flash\":\n    case \"gemini-1.5-flash-latest\":\n      return 1000000; // 1M tokens\n    case \"gemini-pro\":\n    case \"gemini-pro-vision\":\n      return 32768;\n\n    default:\n      return 4097;\n  }\n};\n\n/**\n * Whether or not the input matches the OpenAI tool definition.\n * @param {unknown} tool The input to check.\n * @returns {boolean} Whether the input is an OpenAI tool definition.\n */\nexport function isOpenAITool(tool: unknown): tool is ToolDefinition {\n  if (typeof tool !== \"object\" || !tool) return false;\n  if (\n    \"type\" in tool &&\n    tool.type === \"function\" &&\n    \"function\" in tool &&\n    typeof tool.function === \"object\" &&\n    tool.function &&\n    \"name\" in tool.function &&\n    \"parameters\" in tool.function\n  ) {\n    return true;\n  }\n  return false;\n}\n\ninterface CalculateMaxTokenProps {\n  prompt: string;\n  modelName: TiktokenModel;\n}\n\nexport const calculateMaxTokens = async ({\n  prompt,\n  modelName,\n}: CalculateMaxTokenProps) => {\n  let numTokens;\n\n  try {\n    numTokens = (\n      await encodingForModel(getModelNameForTiktoken(modelName))\n    ).encode(prompt).length;\n  } catch {\n    console.warn(\n      \"Failed to calculate number of tokens, falling back to approximate count\"\n    );\n\n    // fallback to approximate calculation if tiktoken is not available\n    // each token is ~4 characters: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them#\n    numTokens = Math.ceil(prompt.length / 4);\n  }\n\n  const maxTokens = getModelContextSize(modelName);\n  return maxTokens - numTokens;\n};\n\nconst getVerbosity = () => false;\n\nexport type SerializedLLM = {\n  _model: string;\n  _type: string;\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n} & Record<string, any>;\n\nexport interface BaseLangChainParams {\n  verbose?: boolean;\n  callbacks?: Callbacks;\n  tags?: string[];\n  metadata?: Record<string, unknown>;\n}\n\n/**\n * Base class for language models, chains, tools.\n */\nexport abstract class BaseLangChain<\n  RunInput,\n  RunOutput,\n  CallOptions extends RunnableConfig = RunnableConfig,\n>\n  extends Runnable<RunInput, RunOutput, CallOptions>\n  implements BaseLangChainParams\n{\n  /**\n   * Whether to print out response text.\n   */\n  verbose: boolean;\n\n  callbacks?: Callbacks;\n\n  tags?: string[];\n\n  metadata?: Record<string, unknown>;\n\n  get lc_attributes(): { [key: string]: undefined } | undefined {\n    return {\n      callbacks: undefined,\n      verbose: undefined,\n    };\n  }\n\n  constructor(params: BaseLangChainParams) {\n    super(params);\n    this.verbose = params.verbose ?? getVerbosity();\n    this.callbacks = params.callbacks;\n    this.tags = params.tags ?? [];\n    this.metadata = params.metadata ?? {};\n    this._addVersion(\"@langchain/core\", __PKG_VERSION__);\n  }\n\n  protected _addVersion(pkg: string, version: string) {\n    const existing = this.metadata?.versions;\n    this.metadata = {\n      ...this.metadata,\n      versions: {\n        ...(typeof existing === \"object\" && existing !== null ? existing : {}),\n        [pkg]: version,\n      },\n    };\n  }\n}\n\n/**\n * Base interface for language model parameters.\n * A subclass of {@link BaseLanguageModel} should have a constructor that\n * takes in a parameter that extends this interface.\n */\nexport interface BaseLanguageModelParams\n  extends AsyncCallerParams, BaseLangChainParams {\n  /**\n   * @deprecated Use `callbacks` instead\n   */\n  callbackManager?: CallbackManager;\n\n  cache?: BaseCache | boolean;\n}\n\nexport interface BaseLanguageModelTracingCallOptions {\n  /**\n   * Describes the format of structured outputs.\n   * This should be provided if an output is considered to be structured\n   */\n  ls_structured_output_format?: {\n    /**\n     * An object containing the method used for structured output (e.g., \"jsonMode\").\n     */\n    kwargs: { method: string };\n    /**\n     * The JSON schema describing the expected output structure.\n     */\n    schema?: JSONSchema;\n  };\n}\n\nexport interface BaseLanguageModelCallOptions\n  extends RunnableConfig, BaseLanguageModelTracingCallOptions {\n  /**\n   * Stop tokens to use for this call.\n   * If not provided, the default stop tokens for the model will be used.\n   */\n  stop?: string[];\n  /**\n   * Overrides the model's configured `maxRetries` for this call only.\n   */\n  maxRetries?: number;\n}\n\nexport interface FunctionDefinition {\n  /**\n   * The name of the function to be called. Must be a-z, A-Z, 0-9, or contain\n   * underscores and dashes, with a maximum length of 64.\n   */\n  name: string;\n\n  /**\n   * The parameters the functions accepts, described as a JSON Schema object. See the\n   * [guide](https://platform.openai.com/docs/guides/gpt/function-calling) for\n   * examples, and the\n   * [JSON Schema reference](https://json-schema.org/understanding-json-schema/) for\n   * documentation about the format.\n   *\n   * To describe a function that accepts no parameters, provide the value\n   * `{\"type\": \"object\", \"properties\": {}}`.\n   */\n  parameters: Record<string, unknown> | JSONSchema;\n\n  /**\n   * A description of what the function does, used by the model to choose when and\n   * how to call the function.\n   */\n  description?: string;\n}\n\nexport interface ToolDefinition {\n  type: \"function\";\n  function: FunctionDefinition;\n}\n\nexport type FunctionCallOption = {\n  name: string;\n};\n\nexport interface BaseFunctionCallOptions extends BaseLanguageModelCallOptions {\n  function_call?: FunctionCallOption;\n  functions?: FunctionDefinition[];\n}\n\nexport type BaseLanguageModelInput =\n  | BasePromptValueInterface\n  | string\n  | BaseMessageLike[];\n\nexport type StructuredOutputType = InferInteropZodOutput<InteropZodObject>;\n\nexport type StructuredOutputMethodOptions<IncludeRaw extends boolean = false> =\n  {\n    name?: string;\n    method?: \"functionCalling\" | \"jsonMode\" | \"jsonSchema\" | string;\n    includeRaw?: IncludeRaw;\n    /** Whether to use strict mode. Currently only supported by OpenAI models. */\n    strict?: boolean;\n  };\n\n/** @deprecated Use StructuredOutputMethodOptions instead */\nexport type StructuredOutputMethodParams<\n  RunOutput,\n  IncludeRaw extends boolean = false,\n> = {\n  /** @deprecated Pass schema in as the first argument */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  schema: InteropZodType<RunOutput> | Record<string, any>;\n  name?: string;\n  method?: \"functionCalling\" | \"jsonMode\";\n  includeRaw?: IncludeRaw;\n};\n\nexport interface BaseLanguageModelInterface<\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  RunOutput = any,\n  CallOptions extends BaseLanguageModelCallOptions =\n    BaseLanguageModelCallOptions,\n> extends RunnableInterface<BaseLanguageModelInput, RunOutput, CallOptions> {\n  get callKeys(): string[];\n\n  generatePrompt(\n    promptValues: BasePromptValueInterface[],\n    options?: string[] | Partial<CallOptions>,\n    callbacks?: Callbacks\n  ): Promise<LLMResult>;\n\n  _modelType(): string;\n\n  _llmType(): string;\n\n  getNumTokens(content: MessageContent): Promise<number>;\n\n  /**\n   * Get the identifying parameters of the LLM.\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  _identifyingParams(): Record<string, any>;\n\n  serialize(): SerializedLLM;\n}\n\nexport type LanguageModelOutput = BaseMessage | string;\n\nexport type LanguageModelLike = RunnableInterface<\n  BaseLanguageModelInput,\n  LanguageModelOutput\n>;\n\n/**\n * Base class for language models.\n */\nexport abstract class BaseLanguageModel<\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  RunOutput = any,\n  CallOptions extends BaseLanguageModelCallOptions =\n    BaseLanguageModelCallOptions,\n>\n  extends BaseLangChain<BaseLanguageModelInput, RunOutput, CallOptions>\n  implements\n    BaseLanguageModelParams,\n    BaseLanguageModelInterface<RunOutput, CallOptions>\n{\n  /**\n   * Keys that the language model accepts as call options.\n   */\n  get callKeys(): string[] {\n    return [\"stop\", \"timeout\", \"signal\", \"tags\", \"metadata\", \"callbacks\"];\n  }\n\n  /**\n   * The async caller should be used by subclasses to make any async calls,\n   * which will thus benefit from the concurrency and retry logic.\n   */\n  caller: AsyncCaller;\n\n  cache?: BaseCache;\n\n  constructor({\n    callbacks,\n    callbackManager,\n    ...params\n  }: BaseLanguageModelParams) {\n    const { cache, ...rest } = params;\n    super({\n      callbacks: callbacks ?? callbackManager,\n      ...rest,\n    });\n    if (typeof cache === \"object\") {\n      this.cache = cache;\n    } else if (cache) {\n      this.cache = InMemoryCache.global();\n    } else {\n      this.cache = undefined;\n    }\n    this.caller = new AsyncCaller(params ?? {});\n  }\n\n  abstract generatePrompt(\n    promptValues: BasePromptValueInterface[],\n    options?: string[] | CallOptions,\n    callbacks?: Callbacks\n  ): Promise<LLMResult>;\n\n  abstract _modelType(): string;\n\n  abstract _llmType(): string;\n\n  private _encoding?: Tiktoken;\n\n  /**\n   * Get the number of tokens in the content.\n   * @param content The content to get the number of tokens for.\n   * @returns The number of tokens in the content.\n   */\n  async getNumTokens(content: MessageContent) {\n    // Extract text content from MessageContent\n    let textContent: string;\n    if (typeof content === \"string\") {\n      textContent = content;\n    } else {\n      /**\n       * Content is an array of ContentBlock\n       *\n       * ToDo(@christian-bromann): This is a temporary fix to get the number of tokens for the content.\n       * We need to find a better way to do this.\n       * @see https://github.com/langchain-ai/langchainjs/pull/8341#pullrequestreview-2933713116\n       */\n      textContent = content\n        .map((item) => {\n          if (typeof item === \"string\") return item;\n          if (item.type === \"text\" && \"text\" in item) return item.text;\n          return \"\";\n        })\n        .join(\"\");\n    }\n\n    // fallback to approximate calculation if tiktoken is not available\n    let numTokens = Math.ceil(textContent.length / 4);\n\n    if (!this._encoding) {\n      try {\n        this._encoding = await encodingForModel(\n          \"modelName\" in this\n            ? getModelNameForTiktoken(this.modelName as string)\n            : \"gpt2\"\n        );\n      } catch (error) {\n        console.warn(\n          \"Failed to calculate number of tokens, falling back to approximate count\",\n          error\n        );\n      }\n    }\n\n    if (this._encoding) {\n      try {\n        numTokens = this._encoding.encode(textContent).length;\n      } catch (error) {\n        console.warn(\n          \"Failed to calculate number of tokens, falling back to approximate count\",\n          error\n        );\n      }\n    }\n\n    return numTokens;\n  }\n\n  protected static _convertInputToPromptValue(\n    input: BaseLanguageModelInput\n  ): BasePromptValueInterface {\n    if (typeof input === \"string\") {\n      return new StringPromptValue(input);\n    } else if (Array.isArray(input)) {\n      return new ChatPromptValue(input.map(coerceMessageLikeToMessage));\n    } else {\n      return input;\n    }\n  }\n\n  /**\n   * Get the identifying parameters of the LLM.\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  _identifyingParams(): Record<string, any> {\n    return {};\n  }\n\n  /**\n   * Create a unique cache key for a specific call to a specific language model.\n   * @param callOptions Call options for the model\n   * @returns A unique cache key.\n   */\n  _getSerializedCacheKeyParametersForCall(\n    // TODO: Fix when we remove the RunnableLambda backwards compatibility shim.\n    {\n      config,\n      ...callOptions\n    }: CallOptions & { config?: RunnableConfig }\n  ): string {\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    const params: Record<string, any> = {\n      ...this._identifyingParams(),\n      ...callOptions,\n      _type: this._llmType(),\n      _model: this._modelType(),\n    };\n    const filteredEntries = Object.entries(params).filter(\n      ([_, value]) => value !== undefined\n    );\n    const serializedEntries = filteredEntries\n      .map(([key, value]) => `${key}:${JSON.stringify(value)}`)\n      .sort()\n      .join(\",\");\n    return serializedEntries;\n  }\n\n  /**\n   * @deprecated\n   * Return a json-like object representing this LLM.\n   */\n  serialize(): SerializedLLM {\n    return {\n      ...this._identifyingParams(),\n      _type: this._llmType(),\n      _model: this._modelType(),\n    };\n  }\n\n  /**\n   * @deprecated\n   * Load an LLM from a json-like object describing it.\n   */\n  static async deserialize(_data: SerializedLLM): Promise<BaseLanguageModel> {\n    throw new Error(\"Use .toJSON() instead\");\n  }\n\n  /**\n   * Return profiling information for the model.\n   *\n   * @returns {ModelProfile} An object describing the model's capabilities and constraints\n   */\n  get profile(): ModelProfile {\n    return {};\n  }\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema: SerializableSchema<RunOutput>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema: SerializableSchema<RunOutput>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV3Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV3Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV4Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<false>\n  ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | ZodV4Like<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<true>\n  ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n  /**\n   * Model wrapper that returns outputs formatted to match the given schema.\n   *\n   * @template {BaseLanguageModelInput} RunInput The input type for the Runnable, expected to be the same input for the LLM.\n   * @template {Record<string, any>} RunOutput The output type for the Runnable, expected to be a Zod schema object for structured output validation.\n   *\n   * @param {InteropZodType<RunOutput>} schema The schema for the structured output. Either as a Zod schema or a valid JSON schema object.\n   *   If a Zod schema is passed, the returned attributes will be validated, whereas with JSON schema they will not be.\n   * @param {string} name The name of the function to call.\n   * @param {\"functionCalling\" | \"jsonMode\"} [method=functionCalling] The method to use for getting the structured output. Defaults to \"functionCalling\".\n   * @param {boolean | undefined} [includeRaw=false] Whether to include the raw output in the result. Defaults to false.\n   * @returns {Runnable<RunInput, RunOutput> | Runnable<RunInput, { raw: BaseMessage; parsed: RunOutput }>} A new runnable that calls the LLM with structured output.\n   */\n  withStructuredOutput?<\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    RunOutput extends Record<string, any> = Record<string, any>,\n  >(\n    schema:\n      | InteropZodType<RunOutput>\n      // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n      | Record<string, any>,\n    config?: StructuredOutputMethodOptions<boolean>\n  ):\n    | Runnable<BaseLanguageModelInput, RunOutput>\n    | Runnable<\n        BaseLanguageModelInput,\n        {\n          raw: BaseMessage;\n          parsed: RunOutput;\n        }\n      >;\n\n  /**\n   * Filter out large/inappropriate fields from invocation params for tracing metadata.\n   * Removes fields like tools, functions, messages, response_format that can be large.\n   */\n  protected _filterInvocationParamsForTracing(\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    params: Record<string, any>\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  ): Record<string, any> {\n    const { tools, functions, messages, response_format, ...rest } = params;\n    return rest;\n  }\n}\n\n/**\n * Shared interface for token usage\n * return type from LLM calls.\n */\nexport interface TokenUsage {\n  completionTokens?: number;\n  promptTokens?: number;\n  totalTokens?: number;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAgCA,MAAa,2BAA2B,cAAqC;CAC3E,IAAI,UAAU,WAAW,OAAO,GAC9B,OAAO;CAGT,IAAI,UAAU,WAAW,mBAAmB,GAC1C,OAAO;CAGT,IAAI,UAAU,WAAW,gBAAgB,GACvC,OAAO;CAGT,IAAI,UAAU,WAAW,WAAW,GAClC,OAAO;CAGT,IAAI,UAAU,WAAW,QAAQ,GAC/B,OAAO;CAGT,IAAI,UAAU,WAAW,QAAQ,GAC/B,OAAO;CAGT,OAAO;AACT;AAEA,MAAa,2BAA2B,cAA+B;CACrE,QAAQ,WAAR;EACE,KAAK,0BACH,OAAO;EACT,SACE,OAAO;CACX;AACF;;;;;;;;;;;;AAaA,MAAa,uBAAuB,cAA8B;CAGhE,QAFuB,wBAAwB,SAE1B,GAArB;EAEE,KAAK;EACL,KAAK;EACL,KAAK,uBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,qBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK,kBACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,cACH,OAAO;EAGT,KAAK;EACL,KAAK,0BACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK,oBACH,OAAO;EACT,KAAK,oBACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,gBACH,OAAO;EAGT,KAAK;EACL,KAAK,oBACH,OAAO;EACT,KAAK,oBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,cACH,OAAO;EACT,KAAK;EACL,KAAK,sBACH,OAAO;EAGT,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK,2BACH,OAAO;EACT,KAAK;EACL,KAAK,qBACH,OAAO;EAET,SACE,OAAO;CACX;AACF;;;;;;AAOA,SAAgB,aAAa,MAAuC;CAClE,IAAI,OAAO,SAAS,YAAY,CAAC,MAAM,OAAO;CAC9C,IACE,UAAU,QACV,KAAK,SAAS,cACd,cAAc,QACd,OAAO,KAAK,aAAa,YACzB,KAAK,YACL,UAAU,KAAK,YACf,gBAAgB,KAAK,UAErB,OAAO;CAET,OAAO;AACT;AAOA,MAAa,qBAAqB,OAAO,EACvC,QACA,gBAC4B;CAC5B,IAAI;CAEJ,IAAI;EACF,aACE,MAAMA,uBAAAA,iBAAiB,wBAAwB,SAAS,CAAC,EAAA,CACzD,OAAO,MAAM,CAAC,CAAC;CACnB,QAAQ;EACN,QAAQ,KACN,yEACF;EAIA,YAAY,KAAK,KAAK,OAAO,SAAS,CAAC;CACzC;CAGA,OADkB,oBAAoB,SACvB,IAAI;AACrB;AAEA,MAAM,qBAAqB;;;;AAkB3B,IAAsB,gBAAtB,cAKUC,aAAAA,SAEV;;;;CAIE;CAEA;CAEA;CAEA;CAEA,IAAI,gBAA0D;EAC5D,OAAO;GACL,WAAW,KAAA;GACX,SAAS,KAAA;EACX;CACF;CAEA,YAAY,QAA6B;EACvC,MAAM,MAAM;EACZ,KAAK,UAAU,OAAO,WAAW,aAAa;EAC9C,KAAK,YAAY,OAAO;EACxB,KAAK,OAAO,OAAO,QAAQ,CAAC;EAC5B,KAAK,WAAW,OAAO,YAAY,CAAC;EACpC,KAAK,YAAY,mBAAA,QAAkC;CACrD;CAEA,YAAsB,KAAa,SAAiB;EAClD,MAAM,WAAW,KAAK,UAAU;EAChC,KAAK,WAAW;GACd,GAAG,KAAK;GACR,UAAU;IACR,GAAI,OAAO,aAAa,YAAY,aAAa,OAAO,WAAW,CAAC;KACnE,MAAM;GACT;EACF;CACF;AACF;;;;AA2JA,IAAsB,oBAAtB,cAMU,cAIV;;;;CAIE,IAAI,WAAqB;EACvB,OAAO;GAAC;GAAQ;GAAW;GAAU;GAAQ;GAAY;EAAW;CACtE;;;;;CAMA;CAEA;CAEA,YAAY,EACV,WACA,iBACA,GAAG,UACuB;EAC1B,MAAM,EAAE,OAAO,GAAG,SAAS;EAC3B,MAAM;GACJ,WAAW,aAAa;GACxB,GAAG;EACL,CAAC;EACD,IAAI,OAAO,UAAU,UACnB,KAAK,QAAQ;OACR,IAAI,OACT,KAAK,QAAQC,qBAAAA,cAAc,OAAO;OAElC,KAAK,QAAQ,KAAA;EAEf,KAAK,SAAS,IAAIC,2BAAAA,YAAY,UAAU,CAAC,CAAC;CAC5C;CAYA;;;;;;CAOA,MAAM,aAAa,SAAyB;EAE1C,IAAI;EACJ,IAAI,OAAO,YAAY,UACrB,cAAc;;;;;;;;;EASd,cAAc,QACX,KAAK,SAAS;GACb,IAAI,OAAO,SAAS,UAAU,OAAO;GACrC,IAAI,KAAK,SAAS,UAAU,UAAU,MAAM,OAAO,KAAK;GACxD,OAAO;EACT,CAAC,CAAC,CACD,KAAK,EAAE;EAIZ,IAAI,YAAY,KAAK,KAAK,YAAY,SAAS,CAAC;EAEhD,IAAI,CAAC,KAAK,WACR,IAAI;GACF,KAAK,YAAY,MAAMH,uBAAAA,iBACrB,eAAe,OACX,wBAAwB,KAAK,SAAmB,IAChD,MACN;EACF,SAAS,OAAO;GACd,QAAQ,KACN,2EACA,KACF;EACF;EAGF,IAAI,KAAK,WACP,IAAI;GACF,YAAY,KAAK,UAAU,OAAO,WAAW,CAAC,CAAC;EACjD,SAAS,OAAO;GACd,QAAQ,KACN,2EACA,KACF;EACF;EAGF,OAAO;CACT;CAEA,OAAiB,2BACf,OAC0B;EAC1B,IAAI,OAAO,UAAU,UACnB,OAAO,IAAII,sBAAAA,kBAAkB,KAAK;OAC7B,IAAI,MAAM,QAAQ,KAAK,GAC5B,OAAO,IAAIC,sBAAAA,gBAAgB,MAAM,IAAIC,cAAAA,0BAA0B,CAAC;OAEhE,OAAO;CAEX;;;;CAMA,qBAA0C;EACxC,OAAO,CAAC;CACV;;;;;;CAOA,wCAEE,EACE,QACA,GAAG,eAEG;EAER,MAAM,SAA8B;GAClC,GAAG,KAAK,mBAAmB;GAC3B,GAAG;GACH,OAAO,KAAK,SAAS;GACrB,QAAQ,KAAK,WAAW;EAC1B;EAQA,OAPwB,OAAO,QAAQ,MAAM,CAAC,CAAC,QAC5C,CAAC,GAAG,WAAW,UAAU,KAAA,CAEY,CAAC,CACtC,KAAK,CAAC,KAAK,WAAW,GAAG,IAAI,GAAG,KAAK,UAAU,KAAK,GAAG,CAAC,CACxD,KAAK,CAAC,CACN,KAAK,GACe;CACzB;;;;;CAMA,YAA2B;EACzB,OAAO;GACL,GAAG,KAAK,mBAAmB;GAC3B,OAAO,KAAK,SAAS;GACrB,QAAQ,KAAK,WAAW;EAC1B;CACF;;;;;CAMA,aAAa,YAAY,OAAkD;EACzE,MAAM,IAAI,MAAM,uBAAuB;CACzC;;;;;;CAOA,IAAI,UAAwB;EAC1B,OAAO,CAAC;CACV;;;;;CAkGA,kCAEE,QAEqB;EACrB,MAAM,EAAE,OAAO,WAAW,UAAU,iBAAiB,GAAG,SAAS;EACjE,OAAO;CACT;AACF"}

package/dist/language_models/base.js

@@ -138,7 +138,7 @@ var BaseLangChain = class extends Runnable {
 		this.callbacks = params.callbacks;
 		this.tags = params.tags ?? [];
 		this.metadata = params.metadata ?? {};
-		this._addVersion("@langchain/core", "1.2.9");
+		this._addVersion("@langchain/core", "1.2.11");
 	}
 	_addVersion(pkg, version) {
 		const existing = this.metadata?.versions;