langchain 0.1.6 → 0.1.8

package/dist/agents/react/prompt.cjs

@@ -1,7 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.FORMAT_INSTRUCTIONS = void 0;
-exports.FORMAT_INSTRUCTIONS /* #__PURE__ */ = `Use the following format:
+exports.FORMAT_INSTRUCTIONS = `Use the following format:
 
 Question: the input question you must answer
 Thought: you should always think about what to do

package/dist/agents/react/prompt.js

@@ -1,4 +1,4 @@
-export const FORMAT_INSTRUCTIONS /* #__PURE__ */ = `Use the following format:
+export const FORMAT_INSTRUCTIONS = `Use the following format:
 
 Question: the input question you must answer
 Thought: you should always think about what to do

package/dist/agents/toolkits/conversational_retrieval/openai_functions.cjs

@@ -15,13 +15,14 @@ const token_buffer_memory_js_1 = require("./token_buffer_memory.cjs");
  * @returns A Promise that resolves to an initialized AgentExecutor.
  */
 async function createConversationalRetrievalAgent(llm, tools, options) {
-    const { rememberIntermediateSteps = true, memoryKey = "chat_history", outputKey = "output", prefix, verbose, } = options ?? {};
+    const { rememberIntermediateSteps = true, memoryKey = "chat_history", outputKey = "output", inputKey = "input", prefix, verbose, } = options ?? {};
     let memory;
     if (rememberIntermediateSteps) {
         memory = new token_buffer_memory_js_1.OpenAIAgentTokenBufferMemory({
             memoryKey,
             llm,
             outputKey,
+            inputKey,
         });
     }
     else {
@@ -31,6 +32,7 @@ async function createConversationalRetrievalAgent(llm, tools, options) {
             maxTokenLimit: 12000,
             returnMessages: true,
             outputKey,
+            inputKey,
         });
     }
     const executor = await (0, initialize_js_1.initializeAgentExecutorWithOptions)(tools, llm, {

package/dist/agents/toolkits/conversational_retrieval/openai_functions.d.ts

@@ -4,6 +4,7 @@ export type ConversationalRetrievalAgentOptions = {
     rememberIntermediateSteps?: boolean;
     memoryKey?: string;
     outputKey?: string;
+    inputKey?: string;
     prefix?: string;
     verbose?: boolean;
 };

package/dist/agents/toolkits/conversational_retrieval/openai_functions.js

@@ -12,13 +12,14 @@ import { OpenAIAgentTokenBufferMemory } from "./token_buffer_memory.js";
  * @returns A Promise that resolves to an initialized AgentExecutor.
  */
 export async function createConversationalRetrievalAgent(llm, tools, options) {
-    const { rememberIntermediateSteps = true, memoryKey = "chat_history", outputKey = "output", prefix, verbose, } = options ?? {};
+    const { rememberIntermediateSteps = true, memoryKey = "chat_history", outputKey = "output", inputKey = "input", prefix, verbose, } = options ?? {};
     let memory;
     if (rememberIntermediateSteps) {
         memory = new OpenAIAgentTokenBufferMemory({
             memoryKey,
             llm,
             outputKey,
+            inputKey,
         });
     }
     else {
@@ -28,6 +29,7 @@ export async function createConversationalRetrievalAgent(llm, tools, options) {
             maxTokenLimit: 12000,
             returnMessages: true,
             outputKey,
+            inputKey,
         });
     }
     const executor = await initializeAgentExecutorWithOptions(tools, llm, {

package/dist/chains/constitutional_ai/constitutional_principle.js

@@ -56,7 +56,7 @@ export class ConstitutionalPrinciple {
         };
     }
 }
-export const PRINCIPLES = /* #__PURE__ */ {
+export const PRINCIPLES = {
     harmful1: /* #__PURE__ */ new ConstitutionalPrinciple({
         name: "harmful1",
         critiqueRequest: "Identify specific ways in which the assistant’s last response is harmful, unethical, racist, sexist, toxic, dangerous, or illegal.",

package/dist/document_loaders/web/github.cjs

@@ -12,7 +12,7 @@ const async_caller_1 = require("@langchain/core/utils/async_caller");
 const base_js_1 = require("../base.cjs");
 const directory_js_1 = require("../fs/directory.cjs");
 const extname_js_1 = require("../../util/extname.cjs");
-const extensions = new Set(binary_extensions_1.default);
+const extensions = /* #__PURE__ */ new Set(binary_extensions_1.default);
 /**
  * A function that checks if a file path is a binary file based on its
  * extension.

package/dist/document_loaders/web/github.js

@@ -6,7 +6,7 @@ import { AsyncCaller, } from "@langchain/core/utils/async_caller";
 import { BaseDocumentLoader } from "../base.js";
 import { UnknownHandling } from "../fs/directory.js";
 import { extname } from "../../util/extname.js";
-const extensions = new Set(binaryExtensions);
+const extensions = /* #__PURE__ */ new Set(binaryExtensions);
 /**
  * A function that checks if a file path is a binary file based on its
  * extension.

package/dist/evaluation/comparison/pairwise.cjs

@@ -6,7 +6,7 @@ const outputs_1 = require("@langchain/core/outputs");
 const base_js_1 = require("../base.cjs");
 const prompt_js_1 = require("./prompt.cjs");
 const index_js_1 = require("../../chains/index.cjs");
-const SUPPORTED_CRITERIA = /* #__PURE__ */ {
+const SUPPORTED_CRITERIA = {
     conciseness: "Is the submission concise and to the point?",
     relevance: "Is the submission referring to a real quote from the text?",
     correctness: "Is the submission correct, accurate, and factual?",

package/dist/evaluation/comparison/pairwise.js

@@ -3,7 +3,7 @@ import { RUN_KEY } from "@langchain/core/outputs";
 import { eqSet, LLMPairwiseStringEvaluator, } from "../base.js";
 import { PROMPT, PROMPT_WITH_REFERENCES } from "./prompt.js";
 import { ConstitutionalPrinciple } from "../../chains/index.js";
-const SUPPORTED_CRITERIA = /* #__PURE__ */ {
+const SUPPORTED_CRITERIA = {
     conciseness: "Is the submission concise and to the point?",
     relevance: "Is the submission referring to a real quote from the text?",
     correctness: "Is the submission correct, accurate, and factual?",

package/dist/evaluation/criteria/criteria.cjs

@@ -6,7 +6,7 @@ const outputs_1 = require("@langchain/core/outputs");
 const base_js_1 = require("../base.cjs");
 const prompt_js_1 = require("./prompt.cjs");
 const constitutional_principle_js_1 = require("../../chains/constitutional_ai/constitutional_principle.cjs");
-const SUPPORTED_CRITERIA = /* #__PURE__ */ {
+const SUPPORTED_CRITERIA = {
     conciseness: "Is the submission concise and to the point?",
     relevance: "Is the submission referring to a real quote from the text?",
     correctness: "Is the submission correct, accurate, and factual?",

package/dist/evaluation/criteria/criteria.js

@@ -3,7 +3,7 @@ import { RUN_KEY } from "@langchain/core/outputs";
 import { eqSet, LLMStringEvaluator, } from "../base.js";
 import { CRITERIA_PROMPT, PROMPT_WITH_REFERENCES } from "./prompt.js";
 import { ConstitutionalPrinciple } from "../../chains/constitutional_ai/constitutional_principle.js";
-const SUPPORTED_CRITERIA = /* #__PURE__ */ {
+const SUPPORTED_CRITERIA = {
     conciseness: "Is the submission concise and to the point?",
     relevance: "Is the submission referring to a real quote from the text?",
     correctness: "Is the submission correct, accurate, and factual?",

package/dist/indexes/index.cjs

@@ -0,0 +1,5 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.index = void 0;
+var indexing_js_1 = require("./indexing.cjs");
+Object.defineProperty(exports, "index", { enumerable: true, get: function () { return indexing_js_1.index; } });

package/dist/indexes/index.d.ts

@@ -0,0 +1 @@
+export { type CleanupMode, type IndexOptions, index } from "./indexing.js";

package/dist/indexes/index.js

@@ -0,0 +1 @@
+export { index } from "./indexing.js";

package/dist/indexes/indexing.cjs

@@ -0,0 +1,265 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.index = void 0;
+const uuid_1 = require("uuid");
+const base_1 = require("@langchain/community/indexes/base");
+const hash_1 = require("@langchain/core/utils/hash");
+const document_js_1 = require("../document.cjs");
+/**
+ * HashedDocument is a Document with hashes calculated.
+ * Hashes are calculated based on page content and metadata.
+ * It is used for indexing.
+ */
+class HashedDocument {
+    constructor(fields) {
+        Object.defineProperty(this, "uid", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "hash_", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "contentHash", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "metadataHash", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "pageContent", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "metadata", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.uid = fields.uid;
+        this.pageContent = fields.pageContent;
+        this.metadata = fields.metadata;
+    }
+    calculateHashes() {
+        const forbiddenKeys = ["hash_", "content_hash", "metadata_hash"];
+        for (const key of forbiddenKeys) {
+            if (key in this.metadata) {
+                throw new Error(`Metadata cannot contain key ${key} as it is reserved for internal use. Restricted keys: [${forbiddenKeys.join(", ")}]`);
+            }
+        }
+        const contentHash = this._hashStringToUUID(this.pageContent);
+        try {
+            const metadataHash = this._hashNestedDictToUUID(this.metadata);
+            this.contentHash = contentHash;
+            this.metadataHash = metadataHash;
+        }
+        catch (e) {
+            throw new Error(`Failed to hash metadata: ${e}. Please use a dict that can be serialized using json.`);
+        }
+        this.hash_ = this._hashStringToUUID(this.contentHash + this.metadataHash);
+        if (!this.uid) {
+            this.uid = this.hash_;
+        }
+    }
+    toDocument() {
+        return new document_js_1.Document({
+            pageContent: this.pageContent,
+            metadata: this.metadata,
+        });
+    }
+    static fromDocument(document, uid) {
+        const doc = new this({
+            pageContent: document.pageContent,
+            metadata: document.metadata,
+            uid: uid || document.uid,
+        });
+        doc.calculateHashes();
+        return doc;
+    }
+    _hashStringToUUID(inputString) {
+        const hash_value = (0, hash_1.insecureHash)(inputString);
+        return (0, uuid_1.v5)(hash_value, base_1.UUIDV5_NAMESPACE);
+    }
+    _hashNestedDictToUUID(data) {
+        const serialized_data = JSON.stringify(data, Object.keys(data).sort());
+        const hash_value = (0, hash_1.insecureHash)(serialized_data);
+        return (0, uuid_1.v5)(hash_value, base_1.UUIDV5_NAMESPACE);
+    }
+}
+function batch(size, iterable) {
+    const batches = [];
+    let currentBatch = [];
+    iterable.forEach((item) => {
+        currentBatch.push(item);
+        if (currentBatch.length >= size) {
+            batches.push(currentBatch);
+            currentBatch = [];
+        }
+    });
+    if (currentBatch.length > 0) {
+        batches.push(currentBatch);
+    }
+    return batches;
+}
+function deduplicateInOrder(hashedDocuments) {
+    const seen = new Set();
+    const deduplicated = [];
+    for (const hashedDoc of hashedDocuments) {
+        if (!hashedDoc.hash_) {
+            throw new Error("Hashed document does not have a hash");
+        }
+        if (!seen.has(hashedDoc.hash_)) {
+            seen.add(hashedDoc.hash_);
+            deduplicated.push(hashedDoc);
+        }
+    }
+    return deduplicated;
+}
+function getSourceIdAssigner(sourceIdKey) {
+    if (sourceIdKey === null) {
+        return (_doc) => null;
+    }
+    else if (typeof sourceIdKey === "string") {
+        return (doc) => doc.metadata[sourceIdKey];
+    }
+    else if (typeof sourceIdKey === "function") {
+        return sourceIdKey;
+    }
+    else {
+        throw new Error(`sourceIdKey should be null, a string or a function, got ${typeof sourceIdKey}`);
+    }
+}
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+const _isBaseDocumentLoader = (arg) => {
+    if ("load" in arg &&
+        typeof arg.load === "function" &&
+        "loadAndSplit" in arg &&
+        typeof arg.loadAndSplit === "function") {
+        return true;
+    }
+    return false;
+};
+/**
+ * Index data from the doc source into the vector store.
+ *
+ * Indexing functionality uses a manager to keep track of which documents
+ * are in the vector store.
+ *
+ * This allows us to keep track of which documents were updated, and which
+ * documents were deleted, which documents should be skipped.
+ *
+ * For the time being, documents are indexed using their hashes, and users
+ *  are not able to specify the uid of the document.
+ *
+ * @param {IndexArgs} args
+ * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.
+ * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.
+ * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.
+ * @param {IndexOptions | undefined} args.options Options for indexing.
+ * @returns {Promise<IndexingResult>}
+ */
+async function index(args) {
+    const { docsSource, recordManager, vectorStore, options } = args;
+    const { batchSize = 100, cleanup, sourceIdKey, cleanupBatchSize = 1000, forceUpdate = false, } = options ?? {};
+    if (cleanup === "incremental" && !sourceIdKey) {
+        throw new Error("sourceIdKey is required when cleanup mode is incremental. Please provide through 'options.sourceIdKey'.");
+    }
+    const docs = _isBaseDocumentLoader(docsSource)
+        ? await docsSource.load()
+        : docsSource;
+    const sourceIdAssigner = getSourceIdAssigner(sourceIdKey ?? null);
+    const indexStartDt = await recordManager.getTime();
+    let numAdded = 0;
+    let numDeleted = 0;
+    let numUpdated = 0;
+    let numSkipped = 0;
+    const batches = batch(batchSize ?? 100, docs);
+    for (const batch of batches) {
+        const hashedDocs = deduplicateInOrder(batch.map((doc) => HashedDocument.fromDocument(doc)));
+        const sourceIds = hashedDocs.map((doc) => sourceIdAssigner(doc));
+        if (cleanup === "incremental") {
+            hashedDocs.forEach((_hashedDoc, index) => {
+                const source = sourceIds[index];
+                if (source === null) {
+                    throw new Error("sourceIdKey must be provided when cleanup is incremental");
+                }
+            });
+        }
+        const batchExists = await recordManager.exists(hashedDocs.map((doc) => doc.uid));
+        const uids = [];
+        const docsToIndex = [];
+        const docsToUpdate = [];
+        const seenDocs = new Set();
+        hashedDocs.forEach((hashedDoc, i) => {
+            const docExists = batchExists[i];
+            if (docExists) {
+                if (forceUpdate) {
+                    seenDocs.add(hashedDoc.uid);
+                }
+                else {
+                    docsToUpdate.push(hashedDoc.uid);
+                    return;
+                }
+            }
+            uids.push(hashedDoc.uid);
+            docsToIndex.push(hashedDoc.toDocument());
+        });
+        if (docsToUpdate.length > 0) {
+            await recordManager.update(docsToUpdate, { timeAtLeast: indexStartDt });
+            numSkipped += docsToUpdate.length;
+        }
+        if (docsToIndex.length > 0) {
+            await vectorStore.addDocuments(docsToIndex, { ids: uids });
+            numAdded += docsToIndex.length - seenDocs.size;
+            numUpdated += seenDocs.size;
+        }
+        await recordManager.update(hashedDocs.map((doc) => doc.uid), { timeAtLeast: indexStartDt, groupIds: sourceIds });
+        if (cleanup === "incremental") {
+            sourceIds.forEach((sourceId) => {
+                if (!sourceId)
+                    throw new Error("Source id cannot be null");
+            });
+            const uidsToDelete = await recordManager.listKeys({
+                before: indexStartDt,
+                groupIds: sourceIds,
+            });
+            await vectorStore.delete({ ids: uidsToDelete });
+            await recordManager.deleteKeys(uidsToDelete);
+            numDeleted += uidsToDelete.length;
+        }
+    }
+    if (cleanup === "full") {
+        let uidsToDelete = await recordManager.listKeys({
+            before: indexStartDt,
+            limit: cleanupBatchSize,
+        });
+        while (uidsToDelete.length > 0) {
+            await vectorStore.delete({ ids: uidsToDelete });
+            await recordManager.deleteKeys(uidsToDelete);
+            numDeleted += uidsToDelete.length;
+            uidsToDelete = await recordManager.listKeys({
+                before: indexStartDt,
+                limit: cleanupBatchSize,
+            });
+        }
+    }
+    return {
+        numAdded,
+        numDeleted,
+        numUpdated,
+        numSkipped,
+    };
+}
+exports.index = index;

package/dist/indexes/indexing.d.ts

@@ -0,0 +1,75 @@
+import { VectorStore } from "@langchain/core/vectorstores";
+import { RecordManagerInterface } from "@langchain/community/indexes/base";
+import { DocumentInterface } from "@langchain/core/documents";
+import { BaseDocumentLoader } from "../document_loaders/base.js";
+type IndexingResult = {
+    numAdded: number;
+    numDeleted: number;
+    numUpdated: number;
+    numSkipped: number;
+};
+type StringOrDocFunc = string | ((doc: DocumentInterface) => string);
+export type CleanupMode = "full" | "incremental";
+export type IndexOptions = {
+    /**
+     * The number of documents to index in one batch.
+     */
+    batchSize?: number;
+    /**
+     * The cleanup mode to use. Can be "full", "incremental" or undefined.
+     * - **Incremental**: Cleans up all documents that haven't been updated AND
+     *   that are associated with source ids that were seen
+     *   during indexing.
+     *   Clean up is done continuously during indexing helping
+     *   to minimize the probability of users seeing duplicated
+     *   content.
+     * - **Full**: Delete all documents that haven to been returned by the loader.
+     *   Clean up runs after all documents have been indexed.
+     *   This means that users may see duplicated content during indexing.
+     * - **undefined**: Do not delete any documents.
+     */
+    cleanup?: CleanupMode;
+    /**
+     * Optional key that helps identify the original source of the document.
+     * Must either be a string representing the key of the source in the metadata
+     * or a function that takes a document and returns a string representing the source.
+     * **Required when cleanup is incremental**.
+     */
+    sourceIdKey?: StringOrDocFunc;
+    /**
+     * Batch size to use when cleaning up documents.
+     */
+    cleanupBatchSize?: number;
+    /**
+     * Force update documents even if they are present in the
+     * record manager. Useful if you are re-indexing with updated embeddings.
+     */
+    forceUpdate?: boolean;
+};
+interface IndexArgs {
+    docsSource: BaseDocumentLoader | DocumentInterface[];
+    recordManager: RecordManagerInterface;
+    vectorStore: VectorStore;
+    options?: IndexOptions;
+}
+/**
+ * Index data from the doc source into the vector store.
+ *
+ * Indexing functionality uses a manager to keep track of which documents
+ * are in the vector store.
+ *
+ * This allows us to keep track of which documents were updated, and which
+ * documents were deleted, which documents should be skipped.
+ *
+ * For the time being, documents are indexed using their hashes, and users
+ *  are not able to specify the uid of the document.
+ *
+ * @param {IndexArgs} args
+ * @param {BaseDocumentLoader | DocumentInterface[]} args.docsSource The source of documents to index. Can be a DocumentLoader or a list of Documents.
+ * @param {RecordManagerInterface} args.recordManager The record manager to use for keeping track of indexed documents.
+ * @param {VectorStore} args.vectorStore The vector store to use for storing the documents.
+ * @param {IndexOptions | undefined} args.options Options for indexing.
+ * @returns {Promise<IndexingResult>}
+ */
+export declare function index(args: IndexArgs): Promise<IndexingResult>;
+export {};