@llamaindex/core - npm Package Compare versions

Comparing version 0.2.1 to 0.2.2

dist/memory/index.cjs

+Object.defineProperty(exports, '__esModule', { value: true });
+
+var index_cjs$1 = require('../global/index.cjs');
+var index_cjs$2 = require('../storage/chat-store/index.cjs');
+var index_cjs = require('../utils/index.cjs');
+var env = require('@llamaindex/env');
+var index_cjs$3 = require('../prompts/index.cjs');
+
+const DEFAULT_TOKEN_LIMIT_RATIO = 0.75;
+const DEFAULT_CHAT_STORE_KEY = "chat_history";
+/**
+ * A ChatMemory is used to keep the state of back and forth chat messages
+ */ class BaseMemory {
+    _tokenCountForMessages(messages) {
+        if (messages.length === 0) {
+            return 0;
+        }
+        const tokenizer = index_cjs$1.Settings.tokenizer;
+        const str = messages.map((m)=>index_cjs.extractText(m.content)).join(" ");
+        return tokenizer.encode(str).length;
+    }
+}
+class BaseChatStoreMemory extends BaseMemory {
+    constructor(chatStore = new index_cjs$2.SimpleChatStore(), chatStoreKey = DEFAULT_CHAT_STORE_KEY){
+        super();
+        this.chatStore = chatStore;
+        this.chatStoreKey = chatStoreKey;
+    }
+    getAllMessages() {
+        return this.chatStore.getMessages(this.chatStoreKey);
+    }
+    put(messages) {
+        this.chatStore.addMessage(this.chatStoreKey, messages);
+    }
+    set(messages) {
+        this.chatStore.setMessages(this.chatStoreKey, messages);
+    }
+    reset() {
+        this.chatStore.deleteMessages(this.chatStoreKey);
+    }
+}
+
+class ChatMemoryBuffer extends BaseChatStoreMemory {
+    constructor(options){
+        super(options?.chatStore, options?.chatStoreKey);
+        const llm = options?.llm ?? index_cjs$1.Settings.llm;
+        const contextWindow = llm.metadata.contextWindow;
+        this.tokenLimit = options?.tokenLimit ?? Math.ceil(contextWindow * DEFAULT_TOKEN_LIMIT_RATIO);
+        if (options?.chatHistory) {
+            this.chatStore.setMessages(this.chatStoreKey, options.chatHistory);
+        }
+    }
+    getMessages(input, initialTokenCount = 0) {
+        const messages = this.getAllMessages();
+        if (initialTokenCount > this.tokenLimit) {
+            throw new Error("Initial token count exceeds token limit");
+        }
+        let messageCount = messages.length;
+        let currentMessages = messages.slice(-messageCount);
+        let tokenCount = this._tokenCountForMessages(messages) + initialTokenCount;
+        while(tokenCount > this.tokenLimit && messageCount > 1){
+            messageCount -= 1;
+            if (messages.at(-messageCount).role === "assistant") {
+                messageCount -= 1;
+            }
+            currentMessages = messages.slice(-messageCount);
+            tokenCount = this._tokenCountForMessages(currentMessages) + initialTokenCount;
+        }
+        if (tokenCount > this.tokenLimit && messageCount <= 0) {
+            return [];
+        }
+        return messages.slice(-messageCount);
+    }
+}
+
+class ChatSummaryMemoryBuffer extends BaseMemory {
+    constructor(options){
+        super();
+        this.messages = options?.messages ?? [];
+        this.summaryPrompt = options?.summaryPrompt ?? index_cjs$3.defaultSummaryPrompt;
+        this.llm = options?.llm ?? index_cjs$1.Settings.llm;
+        if (!this.llm.metadata.maxTokens) {
+            throw new Error("LLM maxTokens is not set. Needed so the summarizer ensures the context window size of the LLM.");
+        }
+        this.tokenizer = options?.tokenizer ?? env.tokenizers.tokenizer();
+        this.tokensToSummarize = this.llm.metadata.contextWindow - this.llm.metadata.maxTokens;
+        if (this.tokensToSummarize < this.llm.metadata.contextWindow * 0.25) {
+            throw new Error("The number of tokens that trigger the summarize process are less than 25% of the context window. Try lowering maxTokens or use a model with a larger context window.");
+        }
+    }
+    async summarize() {
+        // get the conversation messages to create summary
+        const messagesToSummarize = this.calcConversationMessages();
+        let promptMessages;
+        do {
+            promptMessages = [
+                {
+                    content: this.summaryPrompt.format({
+                        context: index_cjs.messagesToHistory(messagesToSummarize)
+                    }),
+                    role: "user",
+                    options: {}
+                }
+            ];
+            // remove oldest message until the chat history is short enough for the context window
+            messagesToSummarize.shift();
+        }while (this.tokenizer.encode(promptMessages[0].content).length > this.tokensToSummarize)
+        const response = await this.llm.chat({
+            messages: promptMessages
+        });
+        return {
+            content: response.message.content,
+            role: "memory"
+        };
+    }
+    // Find last summary message
+    get lastSummaryIndex() {
+        const reversedMessages = this.messages.slice().reverse();
+        const index = reversedMessages.findIndex((message)=>message.role === "memory");
+        if (index === -1) {
+            return null;
+        }
+        return this.messages.length - 1 - index;
+    }
+    getLastSummary() {
+        const lastSummaryIndex = this.lastSummaryIndex;
+        return lastSummaryIndex ? this.messages[lastSummaryIndex] : null;
+    }
+    get systemMessages() {
+        // get array of all system messages
+        return this.messages.filter((message)=>message.role === "system");
+    }
+    get nonSystemMessages() {
+        // get array of all non-system messages
+        return this.messages.filter((message)=>message.role !== "system");
+    }
+    /**
+   * Calculates the messages that describe the conversation so far.
+   * If there's no memory, all non-system messages are used.
+   * If there's a memory, uses all messages after the last summary message.
+   */ calcConversationMessages(transformSummary) {
+        const lastSummaryIndex = this.lastSummaryIndex;
+        if (!lastSummaryIndex) {
+            // there's no memory, so just use all non-system messages
+            return this.nonSystemMessages;
+        } else {
+            // there's a memory, so use all messages after the last summary message
+            // and convert summary message so it can be send to the LLM
+            const summaryMessage = transformSummary ? {
+                content: `Summary of the conversation so far: ${this.messages[lastSummaryIndex].content}`,
+                role: "system"
+            } : this.messages[lastSummaryIndex];
+            return [
+                summaryMessage,
+                ...this.messages.slice(lastSummaryIndex + 1)
+            ];
+        }
+    }
+    calcCurrentRequestMessages() {
+        // TODO: check order: currently, we're sending:
+        // system messages first, then transient messages and then the messages that describe the conversation so far
+        return [
+            ...this.systemMessages,
+            ...this.calcConversationMessages(true)
+        ];
+    }
+    reset() {
+        this.messages = [];
+    }
+    async getMessages() {
+        const requestMessages = this.calcCurrentRequestMessages();
+        // get tokens of current request messages and the transient messages
+        const tokens = requestMessages.reduce((count, message)=>count + this.tokenizer.encode(index_cjs.extractText(message.content)).length, 0);
+        if (tokens > this.tokensToSummarize) {
+            // if there are too many tokens for the next request, call summarize
+            const memoryMessage = await this.summarize();
+            const lastMessage = this.messages.at(-1);
+            if (lastMessage && lastMessage.role === "user") {
+                // if last message is a user message, ensure that it's sent after the new memory message
+                this.messages.pop();
+                this.messages.push(memoryMessage);
+                this.messages.push(lastMessage);
+            } else {
+                // otherwise just add the memory message
+                this.messages.push(memoryMessage);
+            }
+            // TODO: we still might have too many tokens
+            // e.g. too large system messages or transient messages
+            // how should we deal with that?
+            return this.calcCurrentRequestMessages();
+        }
+        return requestMessages;
+    }
+    async getAllMessages() {
+        return this.getMessages();
+    }
+    put(message) {
+        this.messages.push(message);
+    }
+}
+
+exports.BaseMemory = BaseMemory;
+exports.ChatMemoryBuffer = ChatMemoryBuffer;
+exports.ChatSummaryMemoryBuffer = ChatSummaryMemoryBuffer;

dist/memory/index.d.cts

+import { MessageContent, ChatMessage, LLM } from '../llms/index.cjs';
+import { BaseChatStore } from '../storage/chat-store/index.cjs';
+import { Tokenizer } from '@llamaindex/env';
+import { SummaryPrompt } from '../prompts/index.cjs';
+
+/**
+ * A ChatMemory is used to keep the state of back and forth chat messages
+ */
+declare abstract class BaseMemory<AdditionalMessageOptions extends object = object> {
+    abstract getMessages(input?: MessageContent | undefined): ChatMessage<AdditionalMessageOptions>[] | Promise<ChatMessage<AdditionalMessageOptions>[]>;
+    abstract getAllMessages(): ChatMessage<AdditionalMessageOptions>[] | Promise<ChatMessage<AdditionalMessageOptions>[]>;
+    abstract put(messages: ChatMessage<AdditionalMessageOptions>): void;
+    abstract reset(): void;
+    protected _tokenCountForMessages(messages: ChatMessage[]): number;
+}
+declare abstract class BaseChatStoreMemory<AdditionalMessageOptions extends object = object> extends BaseMemory<AdditionalMessageOptions> {
+    chatStore: BaseChatStore<AdditionalMessageOptions>;
+    chatStoreKey: string;
+    protected constructor(chatStore?: BaseChatStore<AdditionalMessageOptions>, chatStoreKey?: string);
+    getAllMessages(): ChatMessage<AdditionalMessageOptions>[];
+    put(messages: ChatMessage<AdditionalMessageOptions>): void;
+    set(messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    reset(): void;
+}
+
+type ChatMemoryBufferOptions<AdditionalMessageOptions extends object = object> = {
+    tokenLimit?: number | undefined;
+    chatStore?: BaseChatStore<AdditionalMessageOptions> | undefined;
+    chatStoreKey?: string | undefined;
+    chatHistory?: ChatMessage<AdditionalMessageOptions>[] | undefined;
+    llm?: LLM<object, AdditionalMessageOptions> | undefined;
+};
+declare class ChatMemoryBuffer<AdditionalMessageOptions extends object = object> extends BaseChatStoreMemory<AdditionalMessageOptions> {
+    tokenLimit: number;
+    constructor(options?: Partial<ChatMemoryBufferOptions<AdditionalMessageOptions>>);
+    getMessages(input?: MessageContent | undefined, initialTokenCount?: number): ChatMessage<AdditionalMessageOptions>[];
+}
+
+declare class ChatSummaryMemoryBuffer extends BaseMemory {
+    /**
+     * Tokenizer function that converts text to tokens,
+     *  this is used to calculate the number of tokens in a message.
+     */
+    tokenizer: Tokenizer;
+    tokensToSummarize: number;
+    messages: ChatMessage[];
+    summaryPrompt: SummaryPrompt;
+    llm: LLM;
+    constructor(options?: Partial<ChatSummaryMemoryBuffer>);
+    private summarize;
+    private get lastSummaryIndex();
+    getLastSummary(): ChatMessage | null;
+    private get systemMessages();
+    private get nonSystemMessages();
+    /**
+     * Calculates the messages that describe the conversation so far.
+     * If there's no memory, all non-system messages are used.
+     * If there's a memory, uses all messages after the last summary message.
+     */
+    private calcConversationMessages;
+    private calcCurrentRequestMessages;
+    reset(): void;
+    getMessages(): Promise<ChatMessage[]>;
+    getAllMessages(): Promise<ChatMessage[]>;
+    put(message: ChatMessage): void;
+}
+
+export { BaseMemory, ChatMemoryBuffer, ChatSummaryMemoryBuffer };

dist/memory/index.d.ts

+import { MessageContent, ChatMessage, LLM } from '../llms/index.js';
+import { BaseChatStore } from '../storage/chat-store/index.js';
+import { Tokenizer } from '@llamaindex/env';
+import { SummaryPrompt } from '../prompts/index.js';
+
+/**
+ * A ChatMemory is used to keep the state of back and forth chat messages
+ */
+declare abstract class BaseMemory<AdditionalMessageOptions extends object = object> {
+    abstract getMessages(input?: MessageContent | undefined): ChatMessage<AdditionalMessageOptions>[] | Promise<ChatMessage<AdditionalMessageOptions>[]>;
+    abstract getAllMessages(): ChatMessage<AdditionalMessageOptions>[] | Promise<ChatMessage<AdditionalMessageOptions>[]>;
+    abstract put(messages: ChatMessage<AdditionalMessageOptions>): void;
+    abstract reset(): void;
+    protected _tokenCountForMessages(messages: ChatMessage[]): number;
+}
+declare abstract class BaseChatStoreMemory<AdditionalMessageOptions extends object = object> extends BaseMemory<AdditionalMessageOptions> {
+    chatStore: BaseChatStore<AdditionalMessageOptions>;
+    chatStoreKey: string;
+    protected constructor(chatStore?: BaseChatStore<AdditionalMessageOptions>, chatStoreKey?: string);
+    getAllMessages(): ChatMessage<AdditionalMessageOptions>[];
+    put(messages: ChatMessage<AdditionalMessageOptions>): void;
+    set(messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    reset(): void;
+}
+
+type ChatMemoryBufferOptions<AdditionalMessageOptions extends object = object> = {
+    tokenLimit?: number | undefined;
+    chatStore?: BaseChatStore<AdditionalMessageOptions> | undefined;
+    chatStoreKey?: string | undefined;
+    chatHistory?: ChatMessage<AdditionalMessageOptions>[] | undefined;
+    llm?: LLM<object, AdditionalMessageOptions> | undefined;
+};
+declare class ChatMemoryBuffer<AdditionalMessageOptions extends object = object> extends BaseChatStoreMemory<AdditionalMessageOptions> {
+    tokenLimit: number;
+    constructor(options?: Partial<ChatMemoryBufferOptions<AdditionalMessageOptions>>);
+    getMessages(input?: MessageContent | undefined, initialTokenCount?: number): ChatMessage<AdditionalMessageOptions>[];
+}
+
+declare class ChatSummaryMemoryBuffer extends BaseMemory {
+    /**
+     * Tokenizer function that converts text to tokens,
+     *  this is used to calculate the number of tokens in a message.
+     */
+    tokenizer: Tokenizer;
+    tokensToSummarize: number;
+    messages: ChatMessage[];
+    summaryPrompt: SummaryPrompt;
+    llm: LLM;
+    constructor(options?: Partial<ChatSummaryMemoryBuffer>);
+    private summarize;
+    private get lastSummaryIndex();
+    getLastSummary(): ChatMessage | null;
+    private get systemMessages();
+    private get nonSystemMessages();
+    /**
+     * Calculates the messages that describe the conversation so far.
+     * If there's no memory, all non-system messages are used.
+     * If there's a memory, uses all messages after the last summary message.
+     */
+    private calcConversationMessages;
+    private calcCurrentRequestMessages;
+    reset(): void;
+    getMessages(): Promise<ChatMessage[]>;
+    getAllMessages(): Promise<ChatMessage[]>;
+    put(message: ChatMessage): void;
+}
+
+export { BaseMemory, ChatMemoryBuffer, ChatSummaryMemoryBuffer };

dist/memory/index.js

+import { Settings } from '../global/index.js';
+import { SimpleChatStore } from '../storage/chat-store/index.js';
+import { extractText, messagesToHistory } from '../utils/index.js';
+import { tokenizers } from '@llamaindex/env';
+import { defaultSummaryPrompt } from '../prompts/index.js';
+
+const DEFAULT_TOKEN_LIMIT_RATIO = 0.75;
+const DEFAULT_CHAT_STORE_KEY = "chat_history";
+/**
+ * A ChatMemory is used to keep the state of back and forth chat messages
+ */ class BaseMemory {
+    _tokenCountForMessages(messages) {
+        if (messages.length === 0) {
+            return 0;
+        }
+        const tokenizer = Settings.tokenizer;
+        const str = messages.map((m)=>extractText(m.content)).join(" ");
+        return tokenizer.encode(str).length;
+    }
+}
+class BaseChatStoreMemory extends BaseMemory {
+    constructor(chatStore = new SimpleChatStore(), chatStoreKey = DEFAULT_CHAT_STORE_KEY){
+        super();
+        this.chatStore = chatStore;
+        this.chatStoreKey = chatStoreKey;
+    }
+    getAllMessages() {
+        return this.chatStore.getMessages(this.chatStoreKey);
+    }
+    put(messages) {
+        this.chatStore.addMessage(this.chatStoreKey, messages);
+    }
+    set(messages) {
+        this.chatStore.setMessages(this.chatStoreKey, messages);
+    }
+    reset() {
+        this.chatStore.deleteMessages(this.chatStoreKey);
+    }
+}
+
+class ChatMemoryBuffer extends BaseChatStoreMemory {
+    constructor(options){
+        super(options?.chatStore, options?.chatStoreKey);
+        const llm = options?.llm ?? Settings.llm;
+        const contextWindow = llm.metadata.contextWindow;
+        this.tokenLimit = options?.tokenLimit ?? Math.ceil(contextWindow * DEFAULT_TOKEN_LIMIT_RATIO);
+        if (options?.chatHistory) {
+            this.chatStore.setMessages(this.chatStoreKey, options.chatHistory);
+        }
+    }
+    getMessages(input, initialTokenCount = 0) {
+        const messages = this.getAllMessages();
+        if (initialTokenCount > this.tokenLimit) {
+            throw new Error("Initial token count exceeds token limit");
+        }
+        let messageCount = messages.length;
+        let currentMessages = messages.slice(-messageCount);
+        let tokenCount = this._tokenCountForMessages(messages) + initialTokenCount;
+        while(tokenCount > this.tokenLimit && messageCount > 1){
+            messageCount -= 1;
+            if (messages.at(-messageCount).role === "assistant") {
+                messageCount -= 1;
+            }
+            currentMessages = messages.slice(-messageCount);
+            tokenCount = this._tokenCountForMessages(currentMessages) + initialTokenCount;
+        }
+        if (tokenCount > this.tokenLimit && messageCount <= 0) {
+            return [];
+        }
+        return messages.slice(-messageCount);
+    }
+}
+
+class ChatSummaryMemoryBuffer extends BaseMemory {
+    constructor(options){
+        super();
+        this.messages = options?.messages ?? [];
+        this.summaryPrompt = options?.summaryPrompt ?? defaultSummaryPrompt;
+        this.llm = options?.llm ?? Settings.llm;
+        if (!this.llm.metadata.maxTokens) {
+            throw new Error("LLM maxTokens is not set. Needed so the summarizer ensures the context window size of the LLM.");
+        }
+        this.tokenizer = options?.tokenizer ?? tokenizers.tokenizer();
+        this.tokensToSummarize = this.llm.metadata.contextWindow - this.llm.metadata.maxTokens;
+        if (this.tokensToSummarize < this.llm.metadata.contextWindow * 0.25) {
+            throw new Error("The number of tokens that trigger the summarize process are less than 25% of the context window. Try lowering maxTokens or use a model with a larger context window.");
+        }
+    }
+    async summarize() {
+        // get the conversation messages to create summary
+        const messagesToSummarize = this.calcConversationMessages();
+        let promptMessages;
+        do {
+            promptMessages = [
+                {
+                    content: this.summaryPrompt.format({
+                        context: messagesToHistory(messagesToSummarize)
+                    }),
+                    role: "user",
+                    options: {}
+                }
+            ];
+            // remove oldest message until the chat history is short enough for the context window
+            messagesToSummarize.shift();
+        }while (this.tokenizer.encode(promptMessages[0].content).length > this.tokensToSummarize)
+        const response = await this.llm.chat({
+            messages: promptMessages
+        });
+        return {
+            content: response.message.content,
+            role: "memory"
+        };
+    }
+    // Find last summary message
+    get lastSummaryIndex() {
+        const reversedMessages = this.messages.slice().reverse();
+        const index = reversedMessages.findIndex((message)=>message.role === "memory");
+        if (index === -1) {
+            return null;
+        }
+        return this.messages.length - 1 - index;
+    }
+    getLastSummary() {
+        const lastSummaryIndex = this.lastSummaryIndex;
+        return lastSummaryIndex ? this.messages[lastSummaryIndex] : null;
+    }
+    get systemMessages() {
+        // get array of all system messages
+        return this.messages.filter((message)=>message.role === "system");
+    }
+    get nonSystemMessages() {
+        // get array of all non-system messages
+        return this.messages.filter((message)=>message.role !== "system");
+    }
+    /**
+   * Calculates the messages that describe the conversation so far.
+   * If there's no memory, all non-system messages are used.
+   * If there's a memory, uses all messages after the last summary message.
+   */ calcConversationMessages(transformSummary) {
+        const lastSummaryIndex = this.lastSummaryIndex;
+        if (!lastSummaryIndex) {
+            // there's no memory, so just use all non-system messages
+            return this.nonSystemMessages;
+        } else {
+            // there's a memory, so use all messages after the last summary message
+            // and convert summary message so it can be send to the LLM
+            const summaryMessage = transformSummary ? {
+                content: `Summary of the conversation so far: ${this.messages[lastSummaryIndex].content}`,
+                role: "system"
+            } : this.messages[lastSummaryIndex];
+            return [
+                summaryMessage,
+                ...this.messages.slice(lastSummaryIndex + 1)
+            ];
+        }
+    }
+    calcCurrentRequestMessages() {
+        // TODO: check order: currently, we're sending:
+        // system messages first, then transient messages and then the messages that describe the conversation so far
+        return [
+            ...this.systemMessages,
+            ...this.calcConversationMessages(true)
+        ];
+    }
+    reset() {
+        this.messages = [];
+    }
+    async getMessages() {
+        const requestMessages = this.calcCurrentRequestMessages();
+        // get tokens of current request messages and the transient messages
+        const tokens = requestMessages.reduce((count, message)=>count + this.tokenizer.encode(extractText(message.content)).length, 0);
+        if (tokens > this.tokensToSummarize) {
+            // if there are too many tokens for the next request, call summarize
+            const memoryMessage = await this.summarize();
+            const lastMessage = this.messages.at(-1);
+            if (lastMessage && lastMessage.role === "user") {
+                // if last message is a user message, ensure that it's sent after the new memory message
+                this.messages.pop();
+                this.messages.push(memoryMessage);
+                this.messages.push(lastMessage);
+            } else {
+                // otherwise just add the memory message
+                this.messages.push(memoryMessage);
+            }
+            // TODO: we still might have too many tokens
+            // e.g. too large system messages or transient messages
+            // how should we deal with that?
+            return this.calcCurrentRequestMessages();
+        }
+        return requestMessages;
+    }
+    async getAllMessages() {
+        return this.getMessages();
+    }
+    put(message) {
+        this.messages.push(message);
+    }
+}
+
+export { BaseMemory, ChatMemoryBuffer, ChatSummaryMemoryBuffer };

dist/storage/chat-store/index.cjs

+Object.defineProperty(exports, '__esModule', { value: true });
+
+class BaseChatStore {
+}
+
+class SimpleChatStore extends BaseChatStore {
+    #store;
+    setMessages(key, messages) {
+        this.#store.set(key, messages);
+    }
+    getMessages(key) {
+        return this.#store.get(key) ?? [];
+    }
+    addMessage(key, message, idx) {
+        const messages = this.#store.get(key) ?? [];
+        if (idx === undefined) {
+            messages.push(message);
+        } else {
+            messages.splice(idx, 0, message);
+        }
+        this.#store.set(key, messages);
+    }
+    deleteMessages(key) {
+        this.#store.delete(key);
+    }
+    deleteMessage(key, idx) {
+        const messages = this.#store.get(key) ?? [];
+        messages.splice(idx, 1);
+        this.#store.set(key, messages);
+    }
+    getKeys() {
+        return this.#store.keys();
+    }
+    constructor(...args){
+        super(...args);
+        this.#store = new Map();
+    }
+}
+
+exports.BaseChatStore = BaseChatStore;
+exports.SimpleChatStore = SimpleChatStore;

dist/storage/chat-store/index.d.cts

+import { ChatMessage } from '../../llms/index.cjs';
+
+declare abstract class BaseChatStore<AdditionalMessageOptions extends object = object> {
+    abstract setMessages(key: string, messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    abstract getMessages(key: string): ChatMessage<AdditionalMessageOptions>[];
+    abstract addMessage(key: string, message: ChatMessage<AdditionalMessageOptions>, idx?: number): void;
+    abstract deleteMessages(key: string): void;
+    abstract deleteMessage(key: string, idx: number): void;
+    abstract getKeys(): IterableIterator<string>;
+}
+
+declare class SimpleChatStore<AdditionalMessageOptions extends object = object> extends BaseChatStore<AdditionalMessageOptions> {
+    #private;
+    setMessages(key: string, messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    getMessages(key: string): ChatMessage<AdditionalMessageOptions>[];
+    addMessage(key: string, message: ChatMessage<AdditionalMessageOptions>, idx?: number): void;
+    deleteMessages(key: string): void;
+    deleteMessage(key: string, idx: number): void;
+    getKeys(): MapIterator<string>;
+}
+
+export { BaseChatStore, SimpleChatStore };

dist/storage/chat-store/index.d.ts

+import { ChatMessage } from '../../llms/index.js';
+
+declare abstract class BaseChatStore<AdditionalMessageOptions extends object = object> {
+    abstract setMessages(key: string, messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    abstract getMessages(key: string): ChatMessage<AdditionalMessageOptions>[];
+    abstract addMessage(key: string, message: ChatMessage<AdditionalMessageOptions>, idx?: number): void;
+    abstract deleteMessages(key: string): void;
+    abstract deleteMessage(key: string, idx: number): void;
+    abstract getKeys(): IterableIterator<string>;
+}
+
+declare class SimpleChatStore<AdditionalMessageOptions extends object = object> extends BaseChatStore<AdditionalMessageOptions> {
+    #private;
+    setMessages(key: string, messages: ChatMessage<AdditionalMessageOptions>[]): void;
+    getMessages(key: string): ChatMessage<AdditionalMessageOptions>[];
+    addMessage(key: string, message: ChatMessage<AdditionalMessageOptions>, idx?: number): void;
+    deleteMessages(key: string): void;
+    deleteMessage(key: string, idx: number): void;
+    getKeys(): MapIterator<string>;
+}
+
+export { BaseChatStore, SimpleChatStore };

dist/storage/chat-store/index.js

+class BaseChatStore {
+}
+
+class SimpleChatStore extends BaseChatStore {
+    #store;
+    setMessages(key, messages) {
+        this.#store.set(key, messages);
+    }
+    getMessages(key) {
+        return this.#store.get(key) ?? [];
+    }
+    addMessage(key, message, idx) {
+        const messages = this.#store.get(key) ?? [];
+        if (idx === undefined) {
+            messages.push(message);
+        } else {
+            messages.splice(idx, 0, message);
+        }
+        this.#store.set(key, messages);
+    }
+    deleteMessages(key) {
+        this.#store.delete(key);
+    }
+    deleteMessage(key, idx) {
+        const messages = this.#store.get(key) ?? [];
+        messages.splice(idx, 1);
+        this.#store.set(key, messages);
+    }
+    getKeys() {
+        return this.#store.keys();
+    }
+    constructor(...args){
+        super(...args);
+        this.#store = new Map();
+    }
+}
+
+export { BaseChatStore, SimpleChatStore };

dist/embeddings/index.cjs

@@ -76,23 +76,36 @@ Object.defineProperty(exports, '__esModule', { value: true });
 class BaseEmbedding extends index_cjs.TransformComponent {
-    constructor(){
-        super(async (nodes, options)=>{
-            const texts = nodes.map((node)=>node.getContent(index_cjs.MetadataMode.EMBED));
-            const embeddings = await this.getTextEmbeddingsBatch(texts, options);
-            for(let i = 0; i < nodes.length; i++){
-                nodes[i].embedding = embeddings[i];
-            }
-            return nodes;
-        });
-        this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
-        /**
+    constructor(transformFn){
+        if (transformFn) {
+            super(transformFn);
+            this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
+            /**
    * Optionally override this method to retrieve multiple embeddings in a single request
    * @param texts
    */ this.getTextEmbeddings = async (texts)=>{
-            const embeddings = [];
-            for (const text of texts){
-                const embedding = await this.getTextEmbedding(text);
-                embeddings.push(embedding);
-            }
-            return embeddings;
-        };
+                const embeddings = [];
+                for (const text of texts){
+                    const embedding = await this.getTextEmbedding(text);
+                    embeddings.push(embedding);
+                }
+                return embeddings;
+            };
+        } else {
+            super(async (nodes, options)=>{
+                const texts = nodes.map((node)=>node.getContent(index_cjs.MetadataMode.EMBED));
+                const embeddings = await this.getTextEmbeddingsBatch(texts, options);
+                for(let i = 0; i < nodes.length; i++){
+                    nodes[i].embedding = embeddings[i];
+                }
+                return nodes;
+            });
+            this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
+            this.getTextEmbeddings = async (texts)=>{
+                const embeddings = [];
+                for (const text of texts){
+                    const embedding = await this.getTextEmbedding(text);
+                    embeddings.push(embedding);
+                }
+                return embeddings;
+            };
+        }
     }
@@ -142,6 +155,45 @@ similarity(embedding1, embedding2, mode = exports.SimilarityType.DEFAULT) {
 
+/*
+ * Base class for Multi Modal embeddings.
+ */ class MultiModalEmbedding extends BaseEmbedding {
+    constructor(){
+        super(async (nodes, options)=>{
+            const nodeMap = index_cjs.splitNodesByType(nodes);
+            const imageNodes = nodeMap[index_cjs.ModalityType.IMAGE] ?? [];
+            const textNodes = nodeMap[index_cjs.ModalityType.TEXT] ?? [];
+            const embeddings = await batchEmbeddings(textNodes.map((node)=>node.getContent(index_cjs.MetadataMode.EMBED)), this.getTextEmbeddings.bind(this), this.embedBatchSize, options);
+            for(let i = 0; i < textNodes.length; i++){
+                textNodes[i].embedding = embeddings[i];
+            }
+            const imageEmbeddings = await batchEmbeddings(imageNodes.map((n)=>n.image), this.getImageEmbeddings.bind(this), this.embedBatchSize, options);
+            for(let i = 0; i < imageNodes.length; i++){
+                imageNodes[i].embedding = imageEmbeddings[i];
+            }
+            return nodes;
+        });
+    }
+    /**
+   * Optionally override this method to retrieve multiple image embeddings in a single request
+   * @param images
+   */ async getImageEmbeddings(images) {
+        return Promise.all(images.map((imgFilePath)=>this.getImageEmbedding(imgFilePath)));
+    }
+    async getQueryEmbedding(query) {
+        const image = index_cjs$1.extractImage(query);
+        if (image) {
+            return await this.getImageEmbedding(image);
+        }
+        const text = index_cjs$1.extractSingleText(query);
+        if (text) {
+            return await this.getTextEmbedding(text);
+        }
+        return null;
+    }
+}
+
 exports.BaseEmbedding = BaseEmbedding;
 exports.DEFAULT_SIMILARITY_TOP_K = DEFAULT_SIMILARITY_TOP_K;
+exports.MultiModalEmbedding = MultiModalEmbedding;
 exports.batchEmbeddings = batchEmbeddings;
 exports.similarity = similarity;
 exports.truncateMaxTokens = truncateMaxTokens;

dist/embeddings/index.d.cts

 import { Tokenizers } from '@llamaindex/env';
 import { MessageContentDetail } from '../llms/index.cjs';
-import { TransformComponent } from '../schema/index.cjs';
+import { TransformComponent, BaseNode, ImageType } from '../schema/index.cjs';
 
@@ -36,3 +36,3 @@ declare const DEFAULT_SIMILARITY_TOP_K = 2;
     embedInfo?: EmbeddingInfo;
-    constructor();
+    protected constructor(transformFn?: (nodes: BaseNode[], options?: BaseEmbeddingOptions) => Promise<BaseNode[]>);
     similarity(embedding1: number[], embedding2: number[], mode?: SimilarityType): number;
@@ -56,4 +56,15 @@ abstract getTextEmbedding(text: string): Promise<number[]>;
 
+declare abstract class MultiModalEmbedding extends BaseEmbedding {
+    abstract getImageEmbedding(images: ImageType): Promise<number[]>;
+    protected constructor();
+    /**
+     * Optionally override this method to retrieve multiple image embeddings in a single request
+     * @param images
+     */
+    getImageEmbeddings(images: ImageType[]): Promise<number[][]>;
+    getQueryEmbedding(query: MessageContentDetail): Promise<number[] | null>;
+}
+
 declare function truncateMaxTokens(tokenizer: Tokenizers, value: string, maxTokens: number): string;
 
-export { BaseEmbedding, type BaseEmbeddingOptions, DEFAULT_SIMILARITY_TOP_K, type EmbeddingInfo, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };
+export { BaseEmbedding, type BaseEmbeddingOptions, DEFAULT_SIMILARITY_TOP_K, type EmbeddingInfo, MultiModalEmbedding, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };

dist/embeddings/index.d.ts

 import { Tokenizers } from '@llamaindex/env';
 import { MessageContentDetail } from '../llms/index.js';
-import { TransformComponent } from '../schema/index.js';
+import { TransformComponent, BaseNode, ImageType } from '../schema/index.js';
 
@@ -36,3 +36,3 @@ declare const DEFAULT_SIMILARITY_TOP_K = 2;
     embedInfo?: EmbeddingInfo;
-    constructor();
+    protected constructor(transformFn?: (nodes: BaseNode[], options?: BaseEmbeddingOptions) => Promise<BaseNode[]>);
     similarity(embedding1: number[], embedding2: number[], mode?: SimilarityType): number;
@@ -56,4 +56,15 @@ abstract getTextEmbedding(text: string): Promise<number[]>;
 
+declare abstract class MultiModalEmbedding extends BaseEmbedding {
+    abstract getImageEmbedding(images: ImageType): Promise<number[]>;
+    protected constructor();
+    /**
+     * Optionally override this method to retrieve multiple image embeddings in a single request
+     * @param images
+     */
+    getImageEmbeddings(images: ImageType[]): Promise<number[][]>;
+    getQueryEmbedding(query: MessageContentDetail): Promise<number[] | null>;
+}
+
 declare function truncateMaxTokens(tokenizer: Tokenizers, value: string, maxTokens: number): string;
 
-export { BaseEmbedding, type BaseEmbeddingOptions, DEFAULT_SIMILARITY_TOP_K, type EmbeddingInfo, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };
+export { BaseEmbedding, type BaseEmbeddingOptions, DEFAULT_SIMILARITY_TOP_K, type EmbeddingInfo, MultiModalEmbedding, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };

dist/embeddings/index.js

@@ -1,3 +0,3 @@
-import { TransformComponent, MetadataMode } from '../schema/index.js';
-import { extractSingleText } from '../utils/index.js';
+import { TransformComponent, MetadataMode, splitNodesByType, ModalityType } from '../schema/index.js';
+import { extractSingleText, extractImage } from '../utils/index.js';
 import { tokenizers } from '@llamaindex/env';
@@ -74,23 +74,36 @@
 class BaseEmbedding extends TransformComponent {
-    constructor(){
-        super(async (nodes, options)=>{
-            const texts = nodes.map((node)=>node.getContent(MetadataMode.EMBED));
-            const embeddings = await this.getTextEmbeddingsBatch(texts, options);
-            for(let i = 0; i < nodes.length; i++){
-                nodes[i].embedding = embeddings[i];
-            }
-            return nodes;
-        });
-        this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
-        /**
+    constructor(transformFn){
+        if (transformFn) {
+            super(transformFn);
+            this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
+            /**
    * Optionally override this method to retrieve multiple embeddings in a single request
    * @param texts
    */ this.getTextEmbeddings = async (texts)=>{
-            const embeddings = [];
-            for (const text of texts){
-                const embedding = await this.getTextEmbedding(text);
-                embeddings.push(embedding);
-            }
-            return embeddings;
-        };
+                const embeddings = [];
+                for (const text of texts){
+                    const embedding = await this.getTextEmbedding(text);
+                    embeddings.push(embedding);
+                }
+                return embeddings;
+            };
+        } else {
+            super(async (nodes, options)=>{
+                const texts = nodes.map((node)=>node.getContent(MetadataMode.EMBED));
+                const embeddings = await this.getTextEmbeddingsBatch(texts, options);
+                for(let i = 0; i < nodes.length; i++){
+                    nodes[i].embedding = embeddings[i];
+                }
+                return nodes;
+            });
+            this.embedBatchSize = DEFAULT_EMBED_BATCH_SIZE;
+            this.getTextEmbeddings = async (texts)=>{
+                const embeddings = [];
+                for (const text of texts){
+                    const embedding = await this.getTextEmbedding(text);
+                    embeddings.push(embedding);
+                }
+                return embeddings;
+            };
+        }
     }
@@ -140,2 +153,40 @@ similarity(embedding1, embedding2, mode = SimilarityType.DEFAULT) {
 
-export { BaseEmbedding, DEFAULT_SIMILARITY_TOP_K, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };
+/*
+ * Base class for Multi Modal embeddings.
+ */ class MultiModalEmbedding extends BaseEmbedding {
+    constructor(){
+        super(async (nodes, options)=>{
+            const nodeMap = splitNodesByType(nodes);
+            const imageNodes = nodeMap[ModalityType.IMAGE] ?? [];
+            const textNodes = nodeMap[ModalityType.TEXT] ?? [];
+            const embeddings = await batchEmbeddings(textNodes.map((node)=>node.getContent(MetadataMode.EMBED)), this.getTextEmbeddings.bind(this), this.embedBatchSize, options);
+            for(let i = 0; i < textNodes.length; i++){
+                textNodes[i].embedding = embeddings[i];
+            }
+            const imageEmbeddings = await batchEmbeddings(imageNodes.map((n)=>n.image), this.getImageEmbeddings.bind(this), this.embedBatchSize, options);
+            for(let i = 0; i < imageNodes.length; i++){
+                imageNodes[i].embedding = imageEmbeddings[i];
+            }
+            return nodes;
+        });
+    }
+    /**
+   * Optionally override this method to retrieve multiple image embeddings in a single request
+   * @param images
+   */ async getImageEmbeddings(images) {
+        return Promise.all(images.map((imgFilePath)=>this.getImageEmbedding(imgFilePath)));
+    }
+    async getQueryEmbedding(query) {
+        const image = extractImage(query);
+        if (image) {
+            return await this.getImageEmbedding(image);
+        }
+        const text = extractSingleText(query);
+        if (text) {
+            return await this.getTextEmbedding(text);
+        }
+        return null;
+    }
+}
+
+export { BaseEmbedding, DEFAULT_SIMILARITY_TOP_K, MultiModalEmbedding, SimilarityType, batchEmbeddings, similarity, truncateMaxTokens };

package.json

 {
   "name": "@llamaindex/core",
   "type": "module",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "LlamaIndex Core Module",
@@ -160,2 +160,30 @@ "exports": {
       }
+    },
+    "./memory": {
+      "require": {
+        "types": "./dist/memory/index.d.cts",
+        "default": "./dist/memory/index.cjs"
+      },
+      "import": {
+        "types": "./dist/memory/index.d.ts",
+        "default": "./dist/memory/index.js"
+      },
+      "default": {
+        "types": "./dist/memory/index.d.ts",
+        "default": "./dist/memory/index.js"
+      }
+    },
+    "./storage/chat-store": {
+      "require": {
+        "types": "./dist/storage/chat-store/index.d.cts",
+        "default": "./dist/storage/chat-store/index.cjs"
+      },
+      "import": {
+        "types": "./dist/storage/chat-store/index.d.ts",
+        "default": "./dist/storage/chat-store/index.js"
+      },
+      "default": {
+        "types": "./dist/storage/chat-store/index.d.ts",
+        "default": "./dist/storage/chat-store/index.js"
+      }
     }
@@ -162,0 +190,0 @@ },

New alerts

Long strings

Supply chain risk

Contains long string literals, which may be a sign of obfuscated or packed code.

Found 1 instance in 1 package

Fixed alerts

Long strings

Supply chain risk

Contains long string literals, which may be a sign of obfuscated or packed code.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

364741

10.25%

Number of package files

54

17.39%

Lines of code

7945

9.1%

No dependency changes