feat: VectoryMemoryBlock (#2110)

Co-authored-by: Marcus Schiesser <mail@marcusschiesser.de>

.changeset/heavy-trees-report.md

@@ -0,0 +1,6 @@
+---
+"@llamaindex/doc": patch
+"@llamaindex/core": patch
+---
+
+feat: VectoryMemoryBlock

apps/next/src/content/docs/llamaindex/modules/data/memory/index.mdx

@@ -106,34 +106,40 @@ const memory = createMemory({
 
 Long-term memory is represented as `Memory Block` objects. These objects contain information that are from previous user sessions or from the beginning of the current conversation. When memory is retrieved (by calling `getLLM`), the short-term and long-term memories are merged together within the given `tokenLimit`. 
 
-Currently, there are two predefined memory blocks:
+Currently, there are three predefined memory blocks:
 
 - `staticBlock`: A memory block that stores a static piece of information.
 - `factExtractionBlock`: A memory block that extracts facts from the chat history.
+- `vectorBlock`: A memory block that stores and retrieves chat messages from a vector database using semantic similarity search. Messages are stored individually and retrieved based on their relevance to recent conversation context. Here we've passed in the `vectorStore` to use to store and retrieve the chat messages.
 
 This sounds a bit complicated, but it's actually quite simple. Let's look at an example:
 
 ```ts
-import { createMemory, factExtractionBlock, staticBlock } from "llamaindex";
+import { createMemory, factExtractionBlock, staticBlock, vectorBlock } from "llamaindex";
+import { QdrantVectorStore } from "@llamaindex/qdrant";
+import { OpenAIEmbedding } from "@llamaindex/openai";
 
 const memoryBlocks= [
   staticBlock({
-    id: "core_info",
     content: "My name is Logan, and I live in Saskatoon. I work at LlamaIndex.",
   }),
   factExtractionBlock({
-    id: "user-extracted_info",
     priority: 1,
     llm: llm,
     maxFacts: 50,
   }),
+  vectorBlock({
+    vectorStore: new QdrantVectorStore({ url: "http://localhost:6333" }),
+    priority: 2,
+  }),
 ];
 ```
 
-Here, we've setup two memory blocks:
+Here, we've setup three memory blocks:
 
-- `core_info`: A static memory block that stores some core information about the user. This information will always be inserted into the memory. The type used is `MessageContent` to support multi-modal content.
-- `extracted_info`: An extracted memory block that will extract information from the chat history. Here we've passed in the `llm` to use to extract facts from the chat history, and set the `maxFacts` to 50. If the number of extracted facts exceeds this limit, the `maxFacts` will be automatically summarized and reduced to leave room for new information.
+- `staticBlock`: A static memory block that stores some core information about the user. This information will always be inserted into the memory. The type used is `MessageContent` to support multi-modal content.
+- `factExtractionBlock`: An extracted memory block that will extract information from the chat history. Here we've passed in the `llm` to use to extract facts from the chat history, and set the `maxFacts` to 50. If the number of extracted facts exceeds this limit, the `maxFacts` will be automatically summarized and reduced to leave room for new information.
+- `vectorBlock`: A vector memory block that will store in a vector database and retrieve them from there. Messages are stored individually and retrieved based on their relevance to recent conversation context. Here we've passed in the `vectorStore` to use to store and retrieve the chat messages.
 
 You'll also notice that we've set the `priority` for the `factExtractionBlock` block. This is used to determine the handling when the memory blocks content (i.e. long-term memory) + short-term memory exceeds the token limit on the `Memory` object.
 
@@ -158,6 +164,46 @@ When memory is retrieved (using `getLLM`), the short-term and long-term memories
 
 The amount of short-term memory included is specified by the `shortTermTokenLimitRatio`. If it's set to `0.7`, 70% of the `tokenLimit` is used for short-term memory (not including the static memory block).
 
+
+#### VectorBlock Configuration Options
+
+The `vectorBlock` offers several configuration options to customize its behavior:
+
+```ts
+vectorBlock({
+  vectorStore: new QdrantVectorStore({ url: "http://localhost:6333" }),
+  priority: 2,
+  retrievalContextWindow: 5, // Number of recent messages to use for context when retrieving
+  formatTemplate: new PromptTemplate({ template: "Context: {{ context }}" }), // Custom formatting template
+  nodePostprocessors: [/* custom postprocessors */], // Apply processing to retrieved nodes
+  queryOptions: {
+    similarityTopK: 3, // Number of top similar results to return (default: 2)
+    mode: VectorStoreQueryMode.DEFAULT, // Query mode for the vector store
+    sessionFilterKey: "session_id", // Metadata key for session filtering (default: "session_id")
+    // Custom filters can be added here - session filter is automatically included
+    filters: {
+      filters: [
+        { key: "custom_field", value: "custom_value", operator: "==" }
+      ],
+      condition: "and"
+    }
+  }
+})
+```
+
+**Key Configuration Options:**
+
+- **`retrievalContextWindow`**: Number of recent messages to consider when creating the retrieval query (default: 5). A larger window provides more context but may be less precise.
+- **`formatTemplate`**: Template for formatting retrieved information before adding to memory. Defaults to a simple context template.
+- **`nodePostprocessors`**: Array of postprocessors to apply to retrieved nodes, useful for filtering or transforming results.
+- **`queryOptions.similarityTopK`**: Number of most similar messages to retrieve from the vector store (default: 2).
+- **`queryOptions.sessionFilterKey`**: Metadata key used to isolate memory between different sessions (default: "session_id").
+- **`queryOptions.filters`**: Additional metadata filters for retrieval. The session filter is automatically added to ensure memory isolation.
+
+**Session Isolation:**
+
+The vectorBlock automatically adds a session filter using the block's ID to ensure that memories from different sessions don't interfere with each other. This filter uses the `sessionFilterKey` (default: "session_id") and can be customized if needed.
+
 ## Persistence with Snapshots
 
 Save and restore memory state:

e2e/node/vector-store/pinecone.e2e.ts

@@ -23,7 +23,7 @@ await test("pinecone", async (t) => {
   });
 
   const vectorStore = new PineconeVectorStore({
-    embeddingModel: openaiEmbedding,
+    embedModel: openaiEmbedding,
   });
 
   t.after(async () => {

examples/agents/memory/vector-block.ts

@@ -0,0 +1,150 @@
+/**
+ * Example: Vector Memory Block
+ *
+ * This example demonstrates how to use the VectorMemoryBlock to store and retrieve
+ * conversation history using vector similarity search. The vector memory block
+ * stores messages in a vector store and can retrieve relevant context based on
+ * semantic similarity to recent messages.
+ */
+
+import { OpenAI, OpenAIEmbedding } from "@llamaindex/openai";
+import { QdrantVectorStore } from "@llamaindex/qdrant";
+import { createMemory, vectorBlock } from "llamaindex";
+
+// Set up the LLM and embedding model
+const llm = new OpenAI({ model: "gpt-4.1-mini" });
+const embedModel = new OpenAIEmbedding({ model: "text-embedding-3-small" });
+
+// Simulate a conversation with some context
+// This conversation has 8 messages, which is more than the token limit of 100 tokens (set below)
+// The last 4 messages are kept in to short term memory block (as their tokens are in the limit)
+// Whereas the first 5 messages are added to long term memory block (in here we will use the vector memory block with Qdrant)
+const CONVERSATION_TURNS = [
+  //// This is the first 5 messages that are added to long term memory block (vector memory block)
+  {
+    role: "user",
+    content: "Hi, I'm Sarah and I work as a data scientist at Google.",
+  },
+  {
+    role: "assistant",
+    content:
+      "Hello Sarah! It's great to meet you. Data science at Google must be exciting!",
+  },
+  {
+    role: "user",
+    content:
+      "Yes, I specialize in machine learning and natural language processing.",
+  },
+  {
+    role: "assistant",
+    content: "That's impressive! ML and NLP are fascinating fields.",
+  },
+  {
+    role: "user",
+    content:
+      "I have a PhD in Computer Science from Stanford, and I love hiking on weekends.",
+  },
+
+  //// This is the last 4 messages that are added to short term memory block
+  {
+    role: "assistant",
+    content:
+      "Wow, Stanford PhD! And hiking is a great way to unwind from tech work.",
+  },
+  {
+    role: "user",
+    content: "I also have two cats named Whiskers and Mittens.",
+  },
+  {
+    role: "assistant",
+    content:
+      "Cats make wonderful companions! Whiskers and Mittens are cute names.",
+  },
+  {
+    role: "user",
+    content: "Summary information about Sarah and her cats",
+  },
+];
+
+async function main() {
+  console.log("=== Vector Memory Block Example ===\n");
+
+  /**
+   * Create a vector store. You can quickly get a local instance of Qdrant running with Docker:
+   * ```bash
+   * docker pull qdrant/qdrant
+   * docker run -p 6333:6333 qdrant/qdrant
+   * ```
+   *
+   * Go to http://localhost:6333/dashboard#/collections to see your data
+   */
+  const vectorStore = new QdrantVectorStore({
+    url: "http://localhost:6333",
+    embedModel,
+  });
+
+  // Create a vector memory block using the factory function
+  const vectorMemoryBlock = vectorBlock({
+    vectorStore,
+    priority: 5,
+  });
+
+  // Create a memory store with the vector memory block
+  const memory = createMemory([], {
+    llm,
+    memoryBlocks: [vectorMemoryBlock],
+    tokenLimit: 100,
+    shortTermTokenLimitRatio: 0.7,
+  });
+
+  // Store the conversation history in the vector memory
+  console.log(`Adding ${CONVERSATION_TURNS.length} messages to the memory...`);
+  for (const message of CONVERSATION_TURNS) {
+    await memory.add(message);
+  }
+
+  // Retrieve relevant context for the current user request
+  console.log("Retrieving relevant context...");
+  const chatHistory = await memory.getLLM();
+
+  // You will see there's 1 generated context message from vector memory block, and 4 messages from short term memory block
+  console.log("Chat memory:", chatHistory);
+
+  // Now simulate the assistant responding with context
+  console.log("\nAssistant response with context:");
+  const response = await llm.chat({
+    messages: chatHistory,
+  });
+  console.log(response.message.content);
+
+  // Try adding more messages to the memory
+  const newMessages = [
+    {
+      role: "user",
+      content: "Write a long paragraph about weather in Tokyo",
+    },
+    {
+      role: "assistant",
+      content:
+        "The weather in Tokyo is sunny and warm. The temperature is around 20 degrees Celsius. The weather is very nice and the people are friendly.",
+    },
+    {
+      role: "user",
+      content: "What is the weather in Tokyo?",
+    },
+  ];
+  // Add the new messages to the memory
+  for (const message of newMessages) {
+    await memory.add(message);
+  }
+
+  // Try retrieving the new messages
+  const newChatHistory = await memory.getLLM();
+  // You can see now that new chat history will contain the nodes (separated by `\n`) in the
+  // context message that is generated by the vector memory block
+  // The number of retrieved nodes is set by `similarityTopK` in `queryOptions` of `vectorBlock`
+  // (default `similarityTopK` is 2)
+  console.log("New chat history:", newChatHistory);
+}
+
+main().catch(console.error);

examples/storage/qdrantdb/with-gemini.ts

@@ -15,7 +15,7 @@ async function main() {
   const vectorStore = new QdrantVectorStore({
     url: process.env.QDRANT_URL,
     apiKey: process.env.QDRANT_API_KEY,
-    embeddingModel: embedding,
+    embedModel: embedding,
     collectionName: "gemini_test",
   });
   const storageContext = await storageContextFromDefaults({ vectorStore });

examples/storage/qdrantdb/with-jina.ts

@@ -16,7 +16,7 @@ async function main() {
   const vectorStore = new QdrantVectorStore({
     url: process.env.QDRANT_URL,
     apiKey: process.env.QDRANT_API_KEY,
-    embeddingModel: embedding,
+    embedModel: embedding,
     collectionName: "jina_test",
   });
   const storageContext = await storageContextFromDefaults({ vectorStore });

packages/core/src/memory/block/base.ts

@@ -39,7 +39,9 @@ export abstract class BaseMemoryBlock<
    *
    * @returns The memory block content as an array of ChatMessage.
    */
-  abstract get(): Promise<MemoryMessage<TAdditionalMessageOptions>[]>;
+  abstract get(
+    messages?: MemoryMessage<TAdditionalMessageOptions>[],
+  ): Promise<MemoryMessage<TAdditionalMessageOptions>[]>;
 
   /**
    * Store the messages in the memory block.

packages/core/src/memory/block/index.ts

@@ -1,3 +1,4 @@
 export { BaseMemoryBlock } from "./base";
 export { FactExtractionMemoryBlock } from "./fact";
 export { StaticMemoryBlock } from "./static";
+export { VectorMemoryBlock } from "./vector";

packages/core/src/memory/block/vector.ts

@@ -0,0 +1,250 @@
+import type { BaseEmbedding } from "../../embeddings";
+import type { BaseNodePostprocessor } from "../../postprocessor";
+import { BasePromptTemplate, defaultContextSystemPrompt } from "../../prompts";
+import type { NodeWithScore } from "../../schema";
+import { MetadataMode, TextNode } from "../../schema";
+import { extractText } from "../../utils/llms";
+import type {
+  BaseVectorStore,
+  MetadataFilter,
+  VectorStoreQuery,
+} from "../../vector-store";
+import { VectorStoreQueryMode } from "../../vector-store";
+import type { MemoryMessage } from "../types";
+import { BaseMemoryBlock, type MemoryBlockOptions } from "./base";
+
+/**
+ * The options for the vector memory block.
+ */
+export type VectorMemoryBlockOptions = {
+  /**
+   * The vector store to use for retrieval.
+   */
+  vectorStore: BaseVectorStore;
+
+  /**
+   * Maximum number of messages to include for context when retrieving.
+   * @default 5
+   */
+  retrievalContextWindow?: number;
+
+  /**
+   * Template for formatting the retrieved information.
+   * @default new PromptTemplate({ template: "{{ text }}" })
+   */
+  formatTemplate?: BasePromptTemplate;
+
+  /**
+   * List of node postprocessors to apply to the retrieved nodes containing messages.
+   *
+   * @default []
+   */
+  nodePostprocessors?: BaseNodePostprocessor[];
+
+  /**
+   * Configuration options for vector store queries when retrieving memory.
+   *
+   * @default
+   * ```typescript
+   * {
+   *   similarityTopK: 2,                    // Number of top similar results to return
+   *   mode: VectorStoreQueryMode.DEFAULT,   // Query mode for the vector store
+   *   sessionFilterKey: "session_id",       // Metadata key for session filtering
+   *   filters: {
+   *     filters: [
+   *       { key: "session_id", value: "<current block id>", operator: "==" }
+   *     ],
+   *     condition: "and"
+   *   }
+   * }
+   * ```
+   *
+   * Note: A session filter is automatically added to ensure memory isolation between blocks.
+   * If custom filters are provided, the session filter will be merged with them.
+   */
+  queryOptions?: Partial<VectorMemoryBlockQueryOptions>;
+} & MemoryBlockOptions;
+
+export type VectorMemoryBlockQueryOptions = Omit<
+  VectorStoreQuery,
+  "queryEmbedding" | "queryStr"
+> & {
+  sessionFilterKey: string;
+};
+
+/**
+ * A memory block that retrieves relevant information from a vector store.
+ *
+ * This block stores conversation history in a vector store and retrieves
+ * relevant information based on the most recent messages.
+ */
+export class VectorMemoryBlock<
+  TAdditionalMessageOptions extends object = object,
+> extends BaseMemoryBlock<TAdditionalMessageOptions> {
+  private readonly vectorStore: BaseVectorStore;
+  private readonly retrievalContextWindow: number;
+  private readonly formatTemplate: BasePromptTemplate;
+  private readonly nodePostprocessors: BaseNodePostprocessor[];
+  private readonly queryOptions: VectorMemoryBlockQueryOptions;
+
+  constructor(options: VectorMemoryBlockOptions) {
+    super(options);
+
+    // Validate vector store
+    if (!options.vectorStore.storesText) {
+      throw new Error(
+        "vectorStore must store text to be used as a retrieval memory block",
+      );
+    }
+
+    this.vectorStore = options.vectorStore;
+    this.retrievalContextWindow = options.retrievalContextWindow ?? 5;
+    this.queryOptions = this.buildDefaultQueryOptions(options.queryOptions);
+    this.formatTemplate = options.formatTemplate ?? defaultContextSystemPrompt;
+    this.nodePostprocessors = options.nodePostprocessors ?? [];
+  }
+
+  get embedModel(): BaseEmbedding {
+    return this.vectorStore.embedModel;
+  }
+
+  async get(
+    messages: MemoryMessage<TAdditionalMessageOptions>[] = [],
+  ): Promise<MemoryMessage<TAdditionalMessageOptions>[]> {
+    if (messages?.length === 0) return [];
+
+    // Use the last message or a context window of messages for the query
+    let context: MemoryMessage<TAdditionalMessageOptions>[];
+    if (
+      this.retrievalContextWindow > 1 &&
+      messages.length >= this.retrievalContextWindow
+    ) {
+      context = messages.slice(-this.retrievalContextWindow);
+    } else {
+      context = messages;
+    }
+    const queryText = context
+      .map((message) => extractText(message.content))
+      .join("\n\n");
+    if (!queryText) return [];
+
+    // Create and execute the query
+    const queryEmbedding = await this.embedModel.getTextEmbedding(queryText);
+    const query: VectorStoreQuery = {
+      queryStr: queryText,
+      queryEmbedding,
+      ...this.queryOptions,
+    };
+    const results = await this.vectorStore.query(query);
+    if (!results.nodes?.length) return [];
+
+    // Create nodes with scores
+    const nodesWithScores: NodeWithScore[] = results.nodes.map(
+      (node, index) => ({
+        node,
+        score: results.similarities?.[index] ?? undefined,
+      }),
+    );
+
+    // Apply postprocessors
+    let processedNodes = nodesWithScores;
+    for (const postprocessor of this.nodePostprocessors) {
+      processedNodes = await postprocessor.postprocessNodes(
+        processedNodes,
+        queryText,
+      );
+    }
+
+    // Format the results
+    const retrievedText = processedNodes
+      .map(({ node }) => node.getContent(MetadataMode.NONE))
+      .join("\n\n");
+
+    const formattedText = this.formatTemplate.format({
+      context: retrievedText,
+    });
+
+    // Return as memory message
+    return [
+      {
+        id: this.id,
+        role: "memory",
+        content: formattedText,
+      } as MemoryMessage<TAdditionalMessageOptions>,
+    ];
+  }
+
+  async put(
+    messages: MemoryMessage<TAdditionalMessageOptions>[],
+  ): Promise<void> {
+    if (messages.length === 0) return;
+
+    // Format messages with role, text content, and additional info
+    const texts: string[] = [];
+
+    for (const message of messages) {
+      const text = extractText(message.content);
+      if (!text) continue;
+
+      let messageText = text;
+
+      // Add additional info if present
+      const additionalInfo = (message.options ?? {}) as Record<string, unknown>;
+      if (Object.keys(additionalInfo).length > 0) {
+        messageText += `\nAdditional Info: (${JSON.stringify(additionalInfo)})`;
+      }
+
+      texts.push(`<message role='${message.role}'>${messageText}</message>`);
+    }
+
+    if (texts.length === 0) return;
+
+    // Create text node with session metadata
+    const textNode = new TextNode({
+      text: texts.join("\n"),
+      metadata: { [this.queryOptions.sessionFilterKey]: this.id },
+    });
+
+    // Get embedding for the text
+    textNode.embedding = await this.embedModel.getTextEmbedding(textNode.text);
+
+    // Add to vector store
+    await this.vectorStore.add([textNode]);
+  }
+
+  private buildDefaultQueryOptions(
+    options: Partial<VectorMemoryBlockQueryOptions> | undefined,
+  ): VectorMemoryBlockQueryOptions {
+    const {
+      similarityTopK = 2,
+      mode = VectorStoreQueryMode.DEFAULT,
+      sessionFilterKey = "session_id",
+    } = options ?? {};
+
+    let filters = options?.filters;
+
+    const sessionFilter: MetadataFilter = {
+      key: sessionFilterKey,
+      value: this.id,
+      operator: "==",
+    };
+
+    if (filters) {
+      // Only add session_id filter if it doesn't exist in the filters list
+      const sessionIdFilterExists = filters.filters.some(
+        (filter) => filter.key === sessionFilterKey,
+      );
+      if (!sessionIdFilterExists) {
+        filters.filters.push(sessionFilter);
+      }
+    } else {
+      // If no filters are provided, add the session_id filter
+      filters = {
+        filters: [sessionFilter],
+        condition: "and",
+      };
+    }
+
+    return { ...options, similarityTopK, mode, sessionFilterKey, filters };
+  }
+}

packages/core/src/memory/factories.ts

@@ -8,6 +8,10 @@ import {
   StaticMemoryBlock,
   type StaticMemoryBlockOptions,
 } from "./block/static";
+import {
+  VectorMemoryBlock,
+  type VectorMemoryBlockOptions,
+} from "./block/vector";
 import { DEFAULT_TOKEN_LIMIT, Memory, type MemoryOptions } from "./memory";
 import type { MemoryMessage } from "./types";
 
@@ -115,6 +119,17 @@ export function factExtractionBlock<TMessageOptions extends object = object>(
   return new FactExtractionMemoryBlock<TMessageOptions>(options);
 }
 
+/**
+ * create a VectorMemoryBlock
+ * @param options - Configuration options for the vector memory block
+ * @returns A new VectorMemoryBlock instance
+ */
+export function vectorBlock<TMessageOptions extends object = object>(
+  options: VectorMemoryBlockOptions,
+): VectorMemoryBlock<TMessageOptions> {
+  return new VectorMemoryBlock<TMessageOptions>(options);
+}
+
 /**
  * Creates a new Memory instance from a snapshot
  * @param snapshot The snapshot to load from

packages/core/src/memory/memory.ts

@@ -31,6 +31,13 @@ export type MemoryOptions<TMessageOptions extends object = object> = {
    * Used internally for memory restoration from snapshots.
    */
   memoryCursor?: number;
+
+  /**
+   * The default LLM to use for memory retrieval.
+   * If not provided, the default `Settings.llm` will be used.
+   * This default LLM can be overridden by the LLM passed in the `getLLM` method.
+   */
+  llm?: LLM | undefined;
 };
 
 export class Memory<
@@ -65,6 +72,10 @@ export class Memory<
    * The cursor for the messages that have been processed into long-term memory.
    */
   private memoryCursor: number = 0;
+  /**
+   * The default LLM to use for memory retrieval.
+   */
+  private llm: LLM | undefined;
 
   constructor(
     messages: MemoryMessage<TMessageOptions>[] = [],
@@ -76,6 +87,7 @@ export class Memory<
       options.shortTermTokenLimitRatio ?? DEFAULT_SHORT_TERM_TOKEN_LIMIT_RATIO;
     this.memoryBlocks = options.memoryBlocks ?? [];
     this.memoryCursor = options.memoryCursor ?? 0;
+    this.initLLM(options.llm);
 
     this.adapters = {
       ...options.customAdapters,
@@ -84,6 +96,15 @@ export class Memory<
     } as TAdapters & BuiltinAdapters<TMessageOptions>;
   }
 
+  private initLLM(llm: LLM | undefined) {
+    // safe initialize LLM without throwing error if Settings.llm hasn't been set yet
+    try {
+      this.llm = llm ?? Settings.llm;
+    } catch (error) {
+      this.llm = undefined;
+    }
+  }
+
   /**
    * Add a message to the memory
    * @param message - The message to add to the memory
@@ -160,12 +181,13 @@ export class Memory<
   /**
    * Get the messages from the memory, optionally including transient messages.
    * only return messages that are within context window of the LLM
-   * @param llm - To fit the result messages to the context window of the LLM. If not provided, the default token limit will be used.
+   * @param llm - To fit the result messages to the context window of the LLM  (fallback to default llm if not provided).
+   * If llm is not specified in both the constructor and the method, the default token limit will be used.
    * @param transientMessages - Optional transient messages to include.
    * @returns The messages from the memory, optionally including transient messages.
    */
   async getLLM(
-    llm?: LLM,
+    llm: LLM | undefined = this.llm,
     transientMessages?: ChatMessage<TMessageOptions>[],
   ): Promise<ChatMessage[]> {
     // Priority of result messages:
@@ -176,11 +198,20 @@ export class Memory<
       ? Math.ceil(contextWindow * DEFAULT_TOKEN_LIMIT_RATIO)
       : this.tokenLimit;
 
+    let blockInputMessages = this.messages;
+    if (transientMessages && transientMessages.length > 0) {
+      blockInputMessages = [
+        ...this.messages,
+        ...transientMessages.map((m) => this.adapters.llamaindex.toMemory(m)),
+      ];
+    }
+
     // Start with fixed block messages (priority=0)
     // as it must always be included in the retrieval result
     const messages = await this.getMemoryBlockMessages(
       this.memoryBlocks.filter((block) => block.priority === 0),
       tokenLimit,
+      blockInputMessages,
     );
     // remaining token limit for short-term and memory blocks content
     const remainingTokenLimit =
@@ -207,6 +238,7 @@ export class Memory<
     const longTermBlockMessages = await this.getMemoryBlockMessages(
       longTermBlocks,
       memoryBlocksTokenLimit,
+      blockInputMessages,
     );
     messages.push(...longTermBlockMessages);
 
@@ -252,6 +284,7 @@ export class Memory<
   private async getMemoryBlockMessages(
     blocks: BaseMemoryBlock<TMessageOptions>[],
     tokenLimit?: number,
+    messages?: MemoryMessage<TMessageOptions>[],
   ): Promise<ChatMessage<TMessageOptions>[]> {
     if (blocks.length === 0) {
       return [];
@@ -265,7 +298,7 @@ export class Memory<
     let addedTokenCount = 0;
     for (const block of sortedBlocks) {
       try {
-        const content = await block.get();
+        const content = await block.get(messages);
         for (const message of content) {
           const chatMessage = this.adapters.llamaindex.fromMemory(message);
           const messageTokenCount = this.countMessagesToken([chatMessage]);

packages/core/src/vector-store/index.ts

@@ -101,7 +101,9 @@ export type VectorStoreByType = {
 };
 
 export type VectorStoreBaseParams = {
+  // @deprecated: use embedModel instead
   embeddingModel?: BaseEmbedding | undefined;
+  embedModel?: BaseEmbedding | undefined;
 };
 
 export abstract class BaseVectorStore<Client = unknown, T = unknown> {
@@ -117,7 +119,8 @@ export abstract class BaseVectorStore<Client = unknown, T = unknown> {
   ): Promise<VectorStoreQueryResult>;
 
   protected constructor(params?: VectorStoreBaseParams) {
-    this.embedModel = params?.embeddingModel ?? Settings.embedModel;
+    this.embedModel =
+      params?.embedModel ?? params?.embeddingModel ?? Settings.embedModel;
   }
 }
 

packages/llamaindex/src/vector-store/SimpleVectorStore.ts

@@ -272,7 +272,7 @@ export class SimpleVectorStore extends BaseVectorStore {
 
   static async fromPersistPath(
     persistPath: string,
-    embeddingModel?: BaseEmbedding,
+    embedModel?: BaseEmbedding,
   ): Promise<SimpleVectorStore> {
     const dirPath = path.dirname(persistPath);
     if (!(await exists(dirPath))) {
@@ -300,20 +300,20 @@ export class SimpleVectorStore extends BaseVectorStore {
     data.textIdToRefDocId = dataDict.textIdToRefDocId ?? {};
     // @ts-expect-error TS2322
     data.metadataDict = dataDict.metadataDict ?? {};
-    const store = new SimpleVectorStore({ data, embeddingModel });
+    const store = new SimpleVectorStore({ data, embedModel });
     store.persistPath = persistPath;
     return store;
   }
 
   static fromDict(
     saveDict: SimpleVectorStoreData,
-    embeddingModel?: BaseEmbedding,
+    embedModel?: BaseEmbedding,
   ): SimpleVectorStore {
     const data = new SimpleVectorStoreData();
     data.embeddingDict = saveDict.embeddingDict;
     data.textIdToRefDocId = saveDict.textIdToRefDocId;
     data.metadataDict = saveDict.metadataDict;
-    return new SimpleVectorStore({ data, embeddingModel });
+    return new SimpleVectorStore({ data, embedModel });
   }
 
   toDict(): SimpleVectorStoreData {

packages/llamaindex/tests/vector-stores/SimpleVectorStore.test.ts

@@ -59,7 +59,7 @@ describe("SimpleVectorStore", () => {
       }),
     ];
     store = new SimpleVectorStore({
-      embeddingModel: {} as BaseEmbedding, // Mocking the embedModel
+      embedModel: {} as BaseEmbedding, // Mocking the embedModel
       data: {
         embeddingDict: {},
         textIdToRefDocId: {},