add handle llm tool call step

---------
Co-authored-by: Jac-Zac <jacopozac@icloud.com>
Co-authored-by: Thuc Pham <51660321+thucpn@users.noreply.github.com>

.changeset/lovely-ligers-decide.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+bump: use latest ai package version

.changeset/real-zebras-attack.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Dynamically select model for Groq

.changeset/yellow-files-prove.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+feat: enhance style for markdown

.changeset/yellow-jokes-protect.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Add multi agents template for Express

CHANGELOG.md

@@ -1,5 +1,29 @@
 # create-llama
 
+## 0.2.6
+
+### Patch Changes
+
+- adc40cf: fix: vercel ai update crash sending annotations
+
+## 0.2.5
+
+### Patch Changes
+
+- 38a8be8: fix: filter in mongo vector store
+
+## 0.2.4
+
+### Patch Changes
+
+- 917e862: Fix errors in building the frontend
+
+## 0.2.3
+
+### Patch Changes
+
+- b6da3c2: Ensure the generation script always works
+
 ## 0.2.2
 
 ### Patch Changes

helpers/datasources.ts

@@ -36,74 +36,66 @@ export async function writeLoadersConfig(
   dataSources: TemplateDataSource[],
   useLlamaParse?: boolean,
 ) {
-  if (dataSources.length === 0) return; // no datasources, no config needed
-  const loaderConfig = new Document({});
-  // Web loader config
+  const loaderConfig: Record<string, any> = {};
+
+  // Always set file loader config
+  loaderConfig.file = createFileLoaderConfig(useLlamaParse);
+
   if (dataSources.some((ds) => ds.type === "web")) {
-    const webLoaderConfig = new Document({});
-
-    // Create config for browser driver arguments
-    const driverArgNodeValue = webLoaderConfig.createNode([
-      "--no-sandbox",
-      "--disable-dev-shm-usage",
-    ]);
-    driverArgNodeValue.commentBefore =
-      " The arguments to pass to the webdriver. E.g.: add --headless to run in headless mode";
-    webLoaderConfig.set("driver_arguments", driverArgNodeValue);
-
-    // Create config for urls
-    const urlConfigs = dataSources
-      .filter((ds) => ds.type === "web")
-      .map((ds) => {
-        const dsConfig = ds.config as WebSourceConfig;
-        return {
-          base_url: dsConfig.baseUrl,
-          prefix: dsConfig.prefix,
-          depth: dsConfig.depth,
-        };
-      });
-    const urlConfigNode = webLoaderConfig.createNode(urlConfigs);
-    urlConfigNode.commentBefore = ` base_url: The URL to start crawling with
- prefix: Only crawl URLs matching the specified prefix
- depth: The maximum depth for BFS traversal
- You can add more websites by adding more entries (don't forget the - prefix from YAML)`;
-    webLoaderConfig.set("urls", urlConfigNode);
-
-    // Add web config to the loaders config
-    loaderConfig.set("web", webLoaderConfig);
+    loaderConfig.web = createWebLoaderConfig(dataSources);
   }
 
-  // File loader config
-  if (dataSources.some((ds) => ds.type === "file")) {
-    // Add documentation to web loader config
-    const node = loaderConfig.createNode({
-      use_llama_parse: useLlamaParse,
-    });
-    node.commentBefore = ` use_llama_parse: Use LlamaParse if \`true\`. Needs a \`LLAMA_CLOUD_API_KEY\` from https://cloud.llamaindex.ai set as environment variable`;
-    loaderConfig.set("file", node);
-  }
-
-  // DB loader config
   const dbLoaders = dataSources.filter((ds) => ds.type === "db");
   if (dbLoaders.length > 0) {
-    const dbLoaderConfig = new Document({});
-    const configEntries = dbLoaders.map((ds) => {
-      const dsConfig = ds.config as DbSourceConfig;
-      return {
-        uri: dsConfig.uri,
-        queries: [dsConfig.queries],
-      };
-    });
-
-    const node = dbLoaderConfig.createNode(configEntries);
-    node.commentBefore = ` The configuration for the database loader, only supports MySQL and PostgreSQL databases for now.
- uri: The URI for the database. E.g.: mysql+pymysql://user:password@localhost:3306/db or postgresql+psycopg2://user:password@localhost:5432/db
- query: The query to fetch data from the database. E.g.: SELECT * FROM table`;
-    loaderConfig.set("db", node);
+    loaderConfig.db = createDbLoaderConfig(dbLoaders);
   }
 
+  // Create a new Document with the loaderConfig
+  const yamlDoc = new Document(loaderConfig);
+
   // Write loaders config
   const loaderConfigPath = path.join(root, "config", "loaders.yaml");
   await fs.mkdir(path.join(root, "config"), { recursive: true });
-  await fs.writeFile(loaderConfigPath, yaml.stringify(loaderConfig));
+  await fs.writeFile(loaderConfigPath, yaml.stringify(yamlDoc));
+}
+
+function createWebLoaderConfig(dataSources: TemplateDataSource[]): any {
+  const webLoaderConfig: Record<string, any> = {};
+
+  // Create config for browser driver arguments
+  webLoaderConfig.driver_arguments = [
+    "--no-sandbox",
+    "--disable-dev-shm-usage",
+  ];
+
+  // Create config for urls
+  const urlConfigs = dataSources
+    .filter((ds) => ds.type === "web")
+    .map((ds) => {
+      const dsConfig = ds.config as WebSourceConfig;
+      return {
+        base_url: dsConfig.baseUrl,
+        prefix: dsConfig.prefix,
+        depth: dsConfig.depth,
+      };
+    });
+  webLoaderConfig.urls = urlConfigs;
+
+  return webLoaderConfig;
+}
+
+function createFileLoaderConfig(useLlamaParse?: boolean): any {
+  return {
+    use_llama_parse: useLlamaParse,
+  };
+}
+
+function createDbLoaderConfig(dbLoaders: TemplateDataSource[]): any {
+  return dbLoaders.map((ds) => {
+    const dsConfig = ds.config as DbSourceConfig;
+    return {
+      uri: dsConfig.uri,
+      queries: [dsConfig.queries],
+    };
+  });
 }

helpers/index.ts

@@ -96,10 +96,11 @@ async function generateContextData(
   }
 }
 
-const copyContextData = async (
+const prepareContextData = async (
   root: string,
   dataSources: TemplateDataSource[],
 ) => {
+  await makeDir(path.join(root, "data"));
   for (const dataSource of dataSources) {
     const dataSourceConfig = dataSource?.config as FileSourceConfig;
     // Copy local data
@@ -174,25 +175,25 @@ export const installTemplate = async (
       await createBackendEnvFile(props.root, props);
     }
 
-    if (props.dataSources.length > 0) {
+    await prepareContextData(
+      props.root,
+      props.dataSources.filter((ds) => ds.type === "file"),
+    );
+
+    if (
+      props.dataSources.length > 0 &&
+      (props.postInstallAction === "runApp" ||
+        props.postInstallAction === "dependencies")
+    ) {
       console.log("\nGenerating context data...\n");
-      await copyContextData(
-        props.root,
-        props.dataSources.filter((ds) => ds.type === "file"),
+      await generateContextData(
+        props.framework,
+        props.modelConfig,
+        props.packageManager,
+        props.vectorDb,
+        props.llamaCloudKey,
+        props.useLlamaParse,
       );
-      if (
-        props.postInstallAction === "runApp" ||
-        props.postInstallAction === "dependencies"
-      ) {
-        await generateContextData(
-          props.framework,
-          props.modelConfig,
-          props.packageManager,
-          props.vectorDb,
-          props.llamaCloudKey,
-          props.useLlamaParse,
-        );
-      }
     }
 
     // Create outputs directory

helpers/providers/groq.ts

@@ -3,8 +3,55 @@ import prompts from "prompts";
 import { ModelConfigParams } from ".";
 import { questionHandlers, toChoice } from "../../questions";
 
-const MODELS = ["llama3-8b", "llama3-70b", "mixtral-8x7b"];
-const DEFAULT_MODEL = MODELS[0];
+import got from "got";
+import ora from "ora";
+import { red } from "picocolors";
+
+const GROQ_API_URL = "https://api.groq.com/openai/v1";
+
+async function getAvailableModelChoicesGroq(apiKey: string) {
+  if (!apiKey) {
+    throw new Error("Need Groq API key to retrieve model choices");
+  }
+
+  const spinner = ora("Fetching available models from Groq").start();
+  try {
+    const response = await got(`${GROQ_API_URL}/models`, {
+      headers: {
+        Authorization: `Bearer ${apiKey}`,
+      },
+      timeout: 5000,
+      responseType: "json",
+    });
+    const data: any = await response.body;
+    spinner.stop();
+
+    // Filter out the Whisper models
+    return data.data
+      .filter((model: any) => !model.id.toLowerCase().includes("whisper"))
+      .map((el: any) => {
+        return {
+          title: el.id,
+          value: el.id,
+        };
+      });
+  } catch (error: unknown) {
+    spinner.stop();
+    console.log(error);
+    if ((error as any).response?.statusCode === 401) {
+      console.log(
+        red(
+          "Invalid Groq API key provided! Please provide a valid key and try again!",
+        ),
+      );
+    } else {
+      console.log(red("Request failed: " + error));
+    }
+    process.exit(1);
+  }
+}
+
+const DEFAULT_MODEL = "llama3-70b-8192";
 
 // Use huggingface embedding models for now as Groq doesn't support embedding models
 enum HuggingFaceEmbeddingModelType {
@@ -66,12 +113,14 @@ export async function askGroqQuestions({
   // use default model values in CI or if user should not be asked
   const useDefaults = ciInfo.isCI || !askModels;
   if (!useDefaults) {
+    const modelChoices = await getAvailableModelChoicesGroq(config.apiKey!);
+
     const { model } = await prompts(
       {
         type: "select",
         name: "model",
         message: "Which LLM model would you like to use?",
-        choices: MODELS.map(toChoice),
+        choices: modelChoices,
         initial: 0,
       },
       questionHandlers,

helpers/typescript.ts

@@ -33,8 +33,7 @@ export const installTSTemplate = async ({
    * Copy the template files to the target directory.
    */
   console.log("\nInitializing project with template:", template, "\n");
-  const type = template === "multiagent" ? "streaming" : template; // use nextjs streaming template for multiagent
-  const templatePath = path.join(templatesDir, "types", type, framework);
+  const templatePath = path.join(templatesDir, "types", "streaming", framework);
   const copySource = ["**"];
 
   await copy(copySource, root, {
@@ -124,6 +123,30 @@ export const installTSTemplate = async ({
     cwd: path.join(compPath, "vectordbs", "typescript", vectorDb ?? "none"),
   });
 
+  if (template === "multiagent") {
+    const multiagentPath = path.join(compPath, "multiagent", "typescript");
+
+    // copy workflow code for multiagent template
+    await copy("**", path.join(root, relativeEngineDestPath, "workflow"), {
+      parents: true,
+      cwd: path.join(multiagentPath, "workflow"),
+    });
+
+    if (framework === "nextjs") {
+      // patch route.ts file
+      await copy("**", path.join(root, relativeEngineDestPath), {
+        parents: true,
+        cwd: path.join(multiagentPath, "nextjs"),
+      });
+    } else if (framework === "express") {
+      // patch chat.controller.ts file
+      await copy("**", path.join(root, relativeEngineDestPath), {
+        parents: true,
+        cwd: path.join(multiagentPath, "express"),
+      });
+    }
+  }
+
   // copy loader component (TS only supports llama_parse and file for now)
   const loaderFolder = useLlamaParse ? "llama_parse" : "file";
   await copy("**", enginePath, {
@@ -145,6 +168,11 @@ export const installTSTemplate = async ({
     cwd: path.join(compPath, "engines", "typescript", engine),
   });
 
+  // copy settings to engine folder
+  await copy("**", enginePath, {
+    cwd: path.join(compPath, "settings", "typescript"),
+  });
+
   /**
    * Copy the selected UI files to the target directory and reference it.
    */

package.json

@@ -1,6 +1,6 @@
 {
   "name": "create-llama",
-  "version": "0.2.2",
+  "version": "0.2.6",
   "description": "Create LlamaIndex-powered apps with one command",
   "keywords": [
     "rag",

questions.ts

@@ -410,10 +410,7 @@ export const askQuestions = async (
     return; // early return - no further questions needed for llamapack projects
   }
 
-  if (program.template === "multiagent") {
-    // TODO: multi-agents currently only supports FastAPI
-    program.framework = preferences.framework = "fastapi";
-  } else if (program.template === "extractor") {
+  if (program.template === "extractor") {
     // Extractor template only supports FastAPI, empty data sources, and llamacloud
     // So we just use example file for extractor template, this allows user to choose vector database later
     program.dataSources = [EXAMPLE_FILE];
@@ -637,6 +634,7 @@ export const askQuestions = async (
               type: "db",
               config: await prompts(dbPrompts, questionHandlers),
             });
+            break;
           }
           case "llamacloud": {
             program.dataSources.push({

templates/components/multiagent/typescript/express/chat.controller.ts

@@ -0,0 +1,34 @@
+import { Message, StreamData, streamToResponse } from "ai";
+import { Request, Response } from "express";
+import { ChatMessage } from "llamaindex";
+import { createStreamTimeout } from "./llamaindex/streaming/events";
+import { createWorkflow } from "./workflow/factory";
+import { toDataStream } from "./workflow/stream";
+
+export const chat = async (req: Request, res: Response) => {
+  const vercelStreamData = new StreamData();
+  const streamTimeout = createStreamTimeout(vercelStreamData);
+  try {
+    const { messages }: { messages: Message[] } = req.body;
+    const userMessage = messages.pop();
+    if (!messages || !userMessage || userMessage.role !== "user") {
+      return res.status(400).json({
+        error:
+          "messages are required in the request body and the last message must be from the user",
+      });
+    }
+
+    const chatHistory = messages as ChatMessage[];
+    const agent = await createWorkflow(chatHistory, vercelStreamData);
+    agent.run(userMessage.content);
+    const stream = toDataStream(agent.streamEvents(), vercelStreamData);
+    return streamToResponse(stream, res, {}, vercelStreamData);
+  } catch (error) {
+    console.error("[LlamaIndex]", error);
+    return res.status(500).json({
+      detail: (error as Error).message,
+    });
+  } finally {
+    clearTimeout(streamTimeout);
+  }
+};

templates/components/multiagent/typescript/nextjs/route.ts

@@ -0,0 +1,53 @@
+import { initObservability } from "@/app/observability";
+import { Message, StreamData, StreamingTextResponse } from "ai";
+import { ChatMessage } from "llamaindex";
+import { NextRequest, NextResponse } from "next/server";
+import { initSettings } from "./engine/settings";
+import { createStreamTimeout } from "./llamaindex/streaming/events";
+import { createWorkflow } from "./workflow/factory";
+import { toDataStream } from "./workflow/stream";
+
+initObservability();
+initSettings();
+
+export const runtime = "nodejs";
+export const dynamic = "force-dynamic";
+
+export async function POST(request: NextRequest) {
+  // Init Vercel AI StreamData and timeout
+  const vercelStreamData = new StreamData();
+  const streamTimeout = createStreamTimeout(vercelStreamData);
+
+  try {
+    const body = await request.json();
+    const { messages, data }: { messages: Message[]; data?: any } = body;
+    const userMessage = messages.pop();
+    if (!messages || !userMessage || userMessage.role !== "user") {
+      return NextResponse.json(
+        {
+          error:
+            "messages are required in the request body and the last message must be from the user",
+        },
+        { status: 400 },
+      );
+    }
+
+    const chatHistory = messages as ChatMessage[];
+    const agent = await createWorkflow(chatHistory, vercelStreamData);
+    agent.run(userMessage.content);
+    const stream = toDataStream(agent.streamEvents(), vercelStreamData);
+    return new StreamingTextResponse(stream, {}, vercelStreamData);
+  } catch (error) {
+    console.error("[LlamaIndex]", error);
+    return NextResponse.json(
+      {
+        detail: (error as Error).message,
+      },
+      {
+        status: 500,
+      },
+    );
+  } finally {
+    clearTimeout(streamTimeout);
+  }
+}

templates/components/multiagent/typescript/workflow/agents.ts

@@ -0,0 +1,49 @@
+import { ChatMessage, QueryEngineTool } from "llamaindex";
+import { getDataSource } from "../engine";
+import { FunctionCallingAgent } from "./single-agent";
+
+const getQueryEngineTool = async () => {
+  const index = await getDataSource();
+  if (!index) {
+    throw new Error("Index not found. Please create an index first.");
+  }
+
+  const topK = process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined;
+  return new QueryEngineTool({
+    queryEngine: index.asQueryEngine({
+      similarityTopK: topK,
+    }),
+    metadata: {
+      name: "query_index",
+      description: `Use this tool to retrieve information about the text corpus from the index.`,
+    },
+  });
+};
+
+export const createResearcher = async (chatHistory: ChatMessage[]) => {
+  return new FunctionCallingAgent({
+    name: "researcher",
+    tools: [await getQueryEngineTool()],
+    systemPrompt:
+      "You are a researcher agent. You are given a researching task. You must use your tools to complete the research.",
+    chatHistory,
+  });
+};
+
+export const createWriter = (chatHistory: ChatMessage[]) => {
+  return new FunctionCallingAgent({
+    name: "writer",
+    systemPrompt:
+      "You are an expert in writing blog posts. You are given a task to write a blog post. Don't make up any information yourself.",
+    chatHistory,
+  });
+};
+
+export const createReviewer = (chatHistory: ChatMessage[]) => {
+  return new FunctionCallingAgent({
+    name: "reviewer",
+    systemPrompt:
+      "You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. Only if the post is good enough for publishing, then you MUST return 'The post is good.'. In all other cases return your review.",
+    chatHistory,
+  });
+};

templates/components/multiagent/typescript/workflow/factory.ts

@@ -0,0 +1,138 @@
+import {
+  Context,
+  StartEvent,
+  StopEvent,
+  Workflow,
+  WorkflowEvent,
+} from "@llamaindex/core/workflow";
+import { StreamData } from "ai";
+import { ChatMessage, ChatResponseChunk } from "llamaindex";
+import { createResearcher, createReviewer, createWriter } from "./agents";
+import { AgentInput, AgentRunEvent, AgentRunResult } from "./type";
+
+const TIMEOUT = 360 * 1000;
+const MAX_ATTEMPTS = 2;
+
+class ResearchEvent extends WorkflowEvent<{ input: string }> {}
+class WriteEvent extends WorkflowEvent<{
+  input: string;
+  isGood: boolean;
+}> {}
+class ReviewEvent extends WorkflowEvent<{ input: string }> {}
+
+export const createWorkflow = async (
+  chatHistory: ChatMessage[],
+  stream: StreamData,
+) => {
+  const appendStream = (agent: string, text: string) => {
+    stream.appendMessageAnnotation({
+      type: "agent",
+      data: { agent, text },
+    });
+  };
+
+  const runAgent = async (agent: Workflow, input: AgentInput) => {
+    const run = agent.run(new StartEvent({ input }));
+    for await (const event of agent.streamEvents()) {
+      if (event.data instanceof AgentRunEvent) {
+        const { name, msg } = event.data.data;
+        // TODO: better using context.writeEventToStream here instead of directly append to stream
+        // But not sure why it's fail to write to stream from the third event
+        appendStream(name, msg);
+      }
+    }
+    return await run;
+  };
+
+  const start = async (context: Context, ev: StartEvent) => {
+    context.set("task", ev.data.input);
+    return new ResearchEvent({
+      input: `Research for this task: ${ev.data.input}`,
+    });
+  };
+
+  const research = async (context: Context, ev: ResearchEvent) => {
+    const researcher = await createResearcher(chatHistory);
+    const researchRes = await runAgent(researcher, { message: ev.data.input });
+    const researchResult = researchRes.data.result;
+    return new WriteEvent({
+      input: `Write a blog post given this task: ${context.get("task")} using this research content: ${researchResult}`,
+      isGood: false,
+    });
+  };
+
+  const write = async (context: Context, ev: WriteEvent) => {
+    context.set("attempts", context.get("attempts", 0) + 1);
+    const tooManyAttempts = context.get("attempts") > MAX_ATTEMPTS;
+    if (tooManyAttempts) {
+      appendStream(
+        "writer",
+        `Too many attempts (${MAX_ATTEMPTS}) to write the blog post. Proceeding with the current version.`,
+      );
+    }
+
+    if (ev.data.isGood || tooManyAttempts) {
+      const writer = createWriter(chatHistory);
+      const writeRes = (await runAgent(writer, {
+        message: ev.data.input,
+        streaming: true,
+      })) as unknown as StopEvent<AsyncGenerator<ChatResponseChunk>>;
+
+      const result = writeRes.data.result;
+      context.writeEventToStream({
+        data: new AgentRunResult(result),
+      });
+      return new StopEvent({ result }); // stop the workflow
+    }
+
+    const writer = createWriter(chatHistory);
+    const writeRes = await runAgent(writer, { message: ev.data.input });
+    const writeResult = writeRes.data.result;
+    context.set("result", writeResult); // store the last result
+    return new ReviewEvent({ input: writeResult });
+  };
+
+  const review = async (context: Context, ev: ReviewEvent) => {
+    const reviewer = createReviewer(chatHistory);
+    const reviewRes = await reviewer.run(
+      new StartEvent<AgentInput>({ input: { message: ev.data.input } }),
+    );
+    const reviewResult = reviewRes.data.result;
+    const oldContent = context.get("result");
+    const postIsGood = reviewResult.toLowerCase().includes("post is good");
+    appendStream(
+      "reviewer",
+      `The post is ${postIsGood ? "" : "not "}good enough for publishing. Sending back to the writer${
+        postIsGood ? " for publication." : "."
+      }`,
+    );
+    if (postIsGood) {
+      return new WriteEvent({
+        input: `You're blog post is ready for publication. Please respond with just the blog post. Blog post: \`\`\`${oldContent}\`\`\``,
+        isGood: true,
+      });
+    }
+
+    return new WriteEvent({
+      input: `Improve the writing of a given blog post by using a given review.
+            Blog post:
+            \`\`\`
+            ${oldContent}
+            \`\`\`
+
+            Review:
+            \`\`\`
+            ${reviewResult}
+            \`\`\``,
+      isGood: false,
+    });
+  };
+
+  const workflow = new Workflow({ timeout: TIMEOUT, validate: true });
+  workflow.addStep(StartEvent, start, { outputs: ResearchEvent });
+  workflow.addStep(ResearchEvent, research, { outputs: WriteEvent });
+  workflow.addStep(WriteEvent, write, { outputs: [ReviewEvent, StopEvent] });
+  workflow.addStep(ReviewEvent, review, { outputs: WriteEvent });
+
+  return workflow;
+};

templates/components/multiagent/typescript/workflow/single-agent.ts

@@ -0,0 +1,252 @@
+import {
+  Context,
+  StartEvent,
+  StopEvent,
+  Workflow,
+  WorkflowEvent,
+} from "@llamaindex/core/workflow";
+import {
+  BaseToolWithCall,
+  ChatMemoryBuffer,
+  ChatMessage,
+  ChatResponse,
+  ChatResponseChunk,
+  LLM,
+  Settings,
+  ToolCall,
+  ToolCallLLM,
+} from "llamaindex";
+import { AgentInput, AgentRunEvent } from "./type";
+
+class InputEvent extends WorkflowEvent<{
+  input: ChatMessage[];
+}> {}
+
+class ToolCallEvent extends WorkflowEvent<{
+  toolCalls: ToolCall[];
+}> {}
+
+export class FunctionCallingAgent extends Workflow {
+  name: string;
+  llm: LLM;
+  memory: ChatMemoryBuffer;
+  tools: BaseToolWithCall[];
+  systemPrompt?: string;
+  writeEvents: boolean;
+  role?: string;
+  toolCalled: boolean = false;
+
+  constructor(options: {
+    name: string;
+    llm?: LLM;
+    chatHistory?: ChatMessage[];
+    tools?: BaseToolWithCall[];
+    systemPrompt?: string;
+    writeEvents?: boolean;
+    role?: string;
+    verbose?: boolean;
+    timeout?: number;
+  }) {
+    super({
+      verbose: options?.verbose ?? false,
+      timeout: options?.timeout ?? 360,
+    });
+    this.name = options?.name;
+    this.llm = options.llm ?? Settings.llm;
+    this.checkToolCallSupport();
+    this.memory = new ChatMemoryBuffer({
+      llm: this.llm,
+      chatHistory: options.chatHistory,
+    });
+    this.tools = options?.tools ?? [];
+    this.systemPrompt = options.systemPrompt;
+    this.writeEvents = options?.writeEvents ?? true;
+    this.role = options?.role;
+
+    // add steps
+    this.addStep(StartEvent<AgentInput>, this.prepareChatHistory, {
+      outputs: InputEvent,
+    });
+    this.addStep(InputEvent, this.handleLLMInput, {
+      outputs: [ToolCallEvent, StopEvent],
+    });
+    this.addStep(ToolCallEvent, this.handleToolCalls, {
+      outputs: InputEvent,
+    });
+  }
+
+  private get chatHistory() {
+    return this.memory.getAllMessages();
+  }
+
+  private get toolsByName() {
+    return this.tools.reduce((acc: Record<string, BaseToolWithCall>, tool) => {
+      acc[tool.metadata.name] = tool;
+      return acc;
+    }, {});
+  }
+
+  private async prepareChatHistory(
+    ctx: Context,
+    ev: StartEvent<AgentInput>,
+  ): Promise<InputEvent> {
+    this.toolCalled = false;
+    const { message, streaming } = ev.data.input;
+    ctx.set("streaming", streaming);
+    this.writeEvent(`Start to work on: ${message}`, ctx);
+    if (this.systemPrompt) {
+      this.memory.put({ role: "system", content: this.systemPrompt });
+    }
+    this.memory.put({ role: "user", content: message });
+    return new InputEvent({ input: this.chatHistory });
+  }
+
+  private async handleLLMInput(
+    ctx: Context,
+    ev: InputEvent,
+  ): Promise<StopEvent<string | AsyncGenerator> | ToolCallEvent> {
+    const isStreaming = ctx.get("streaming");
+    const llmArgs = { messages: this.chatHistory, tools: this.tools };
+
+    if (isStreaming) {
+      return await this.handleLLMInputStream(ctx, ev);
+    }
+
+    const nonStreamingRes = await this.llm.chat({ ...llmArgs });
+    const toolCalls = this.getToolCallsFromResponse(nonStreamingRes);
+    if (toolCalls.length && !this.toolCalled) {
+      return new ToolCallEvent({ toolCalls });
+    }
+    this.writeEvent("Finished task", ctx);
+    const result = nonStreamingRes.message.content.toString();
+    return new StopEvent({ result });
+  }
+
+  private async handleLLMInputStream(
+    context: Context,
+    ev: InputEvent,
+  ): Promise<StopEvent<AsyncGenerator> | ToolCallEvent> {
+    const { llm, tools, memory } = this;
+    const llmArgs = { messages: this.chatHistory, tools };
+
+    const responseGenerator = async function* () {
+      const responseStream = await llm.chat({ ...llmArgs, stream: true });
+
+      let fullResponse = null;
+      let yieldedIndicator = false;
+      for await (const chunk of responseStream) {
+        const hasToolCalls = chunk.options && "toolCall" in chunk.options;
+        if (!hasToolCalls) {
+          if (!yieldedIndicator) {
+            yield false;
+            yieldedIndicator = true;
+          }
+          yield chunk;
+        } else if (!yieldedIndicator) {
+          yield true;
+          yieldedIndicator = true;
+        }
+
+        fullResponse = chunk;
+      }
+
+      if (fullResponse) {
+        memory.put({
+          role: "system",
+          content: fullResponse.delta,
+        });
+        yield fullResponse;
+      }
+    };
+
+    const generator = responseGenerator();
+    const isToolCall = await generator.next();
+    if (isToolCall.value) {
+      const fullResponse = await generator.next();
+      const toolCalls = this.getToolCallsFromResponse(
+        fullResponse.value as ChatResponseChunk<object>,
+      );
+      return new ToolCallEvent({ toolCalls });
+    }
+
+    this.writeEvent("Finished task", context);
+    return new StopEvent({ result: generator });
+  }
+
+  private async handleToolCalls(
+    ctx: Context,
+    ev: ToolCallEvent,
+  ): Promise<InputEvent> {
+    this.toolCalled = true;
+    const { toolCalls } = ev.data;
+
+    const toolMsgs: ChatMessage[] = [];
+    for (const toolCall of toolCalls) {
+      const tool = this.toolsByName[toolCall.name];
+      const options = {
+        tool_call_id: toolCall.id,
+        name: tool.metadata.name,
+      };
+      if (!tool) {
+        toolMsgs.push({
+          role: "system",
+          content: `Tool ${toolCall.name} does not exist`,
+          options,
+        });
+        continue;
+      }
+
+      try {
+        const toolInput = JSON.parse(toolCall.input.toString());
+        const toolOutput = await tool.call(toolInput);
+        toolMsgs.push({
+          role: "system",
+          content: toolOutput.toString(),
+          options,
+        });
+      } catch (e) {
+        console.error(e);
+        toolMsgs.push({
+          role: "system",
+          content: `Encountered error in tool call: ${e}`,
+          options,
+        });
+      }
+    }
+
+    for (const msg of toolMsgs) {
+      this.memory.put(msg);
+    }
+
+    return new InputEvent({ input: this.memory.getAllMessages() });
+  }
+
+  private writeEvent(msg: string, context: Context) {
+    if (!this.writeEvents) return;
+    context.writeEventToStream({
+      data: new AgentRunEvent({ name: this.name, msg }),
+    });
+  }
+
+  private checkToolCallSupport() {
+    const { supportToolCall } = this.llm as ToolCallLLM;
+    if (!supportToolCall) throw new Error("LLM does not support tool calls");
+  }
+
+  // TODO: in LITS, llm should have a method to get tool calls from response
+  // then we don't need to use toolCalled flag
+  private getToolCallsFromResponse(
+    response: ChatResponse<object> | ChatResponseChunk<object>,
+  ): ToolCall[] {
+    let options;
+    if ("message" in response) {
+      options = response.message.options;
+    } else {
+      options = response.options;
+    }
+    if (options && "toolCall" in options) {
+      return options.toolCall as ToolCall[];
+    }
+    return [];
+  }
+}

templates/components/multiagent/typescript/workflow/stream.ts

@@ -0,0 +1,44 @@
+import { WorkflowEvent } from "@llamaindex/core/workflow";
+import {
+  createCallbacksTransformer,
+  createStreamDataTransformer,
+  StreamData,
+  trimStartOfStreamHelper,
+  type AIStreamCallbacksAndOptions,
+} from "ai";
+import { AgentRunResult } from "./type";
+
+export function toDataStream(
+  generator: AsyncGenerator<WorkflowEvent, void>,
+  data: StreamData,
+  callbacks?: AIStreamCallbacksAndOptions,
+) {
+  return toReadableStream(generator, data)
+    .pipeThrough(createCallbacksTransformer(callbacks))
+    .pipeThrough(createStreamDataTransformer());
+}
+
+function toReadableStream(
+  generator: AsyncGenerator<WorkflowEvent, void>,
+  data: StreamData,
+) {
+  const trimStartOfStream = trimStartOfStreamHelper();
+  return new ReadableStream<string>({
+    start(controller) {
+      controller.enqueue(""); // Kickstart the stream
+    },
+    async pull(controller): Promise<void> {
+      const { value, done } = await generator.next();
+      if (done) return;
+      if (value.data instanceof AgentRunResult) {
+        const finalResultStream = value.data.response;
+        for await (const event of finalResultStream) {
+          const text = trimStartOfStream(event.delta ?? "");
+          if (text) controller.enqueue(text);
+        }
+        controller.close();
+        data.close();
+      }
+    },
+  });
+}

templates/components/multiagent/typescript/workflow/type.ts

@@ -0,0 +1,18 @@
+import { WorkflowEvent } from "@llamaindex/core/workflow";
+import { ChatResponseChunk } from "llamaindex";
+
+export type AgentInput = {
+  message: string;
+  streaming?: boolean;
+};
+
+export class AgentRunEvent extends WorkflowEvent<{
+  name: string;
+  msg: string;
+}> {}
+
+export class AgentRunResult {
+  constructor(
+    public response: AsyncGenerator<ChatResponseChunk, any, unknown>,
+  ) {}
+}

templates/components/services/python/file.py

@@ -3,7 +3,7 @@ import mimetypes
 import os
 from io import BytesIO
 from pathlib import Path
-from typing import Any, List, Tuple
+from typing import List, Optional, Tuple
 
 from app.engine.index import IndexConfig, get_index
 from llama_index.core import VectorStoreIndex
@@ -72,7 +72,12 @@ class PrivateFileService:
         return documents
 
     @staticmethod
-    def process_file(file_name: str, base64_content: str, params: Any) -> List[str]:
+    def process_file(
+        file_name: str, base64_content: str, params: Optional[dict] = None
+    ) -> List[str]:
+        if params is None:
+            params = {}
+
         file_data, extension = PrivateFileService.preprocess_base64_file(base64_content)
 
         # Add the nodes to the index and persist it

templates/components/settings/python/settings.py

@@ -126,13 +126,7 @@ def init_fastembed():
 def init_groq():
     from llama_index.llms.groq import Groq
 
-    model_map: Dict[str, str] = {
-        "llama3-8b": "llama3-8b-8192",
-        "llama3-70b": "llama3-70b-8192",
-        "mixtral-8x7b": "mixtral-8x7b-32768",
-    }
-
-    Settings.llm = Groq(model=model_map[os.getenv("MODEL")])
+    Settings.llm = Groq(model=os.getenv("MODEL"))
     # Groq does not provide embeddings, so we use FastEmbed instead
     init_fastembed()
 

templates/components/settings/typescript/settings.ts

@@ -138,14 +138,8 @@ function initGroq() {
     "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
   };
 
-  const modelMap: Record<string, string> = {
-    "llama3-8b": "llama3-8b-8192",
-    "llama3-70b": "llama3-70b-8192",
-    "mixtral-8x7b": "mixtral-8x7b-32768",
-  };
-
   Settings.llm = new Groq({
-    model: modelMap[process.env.MODEL!],
+    model: process.env.MODEL!,
   });
 
   Settings.embedModel = new HuggingFaceEmbedding({

templates/components/vectordbs/typescript/mongo/generate.ts

@@ -1,14 +1,11 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  MongoDBAtlasVectorSearch,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { storageContextFromDefaults, VectorStoreIndex } from "llamaindex";
+import { MongoDBAtlasVectorSearch } from "llamaindex/storage/vectorStore/MongoDBAtlasVectorStore";
 import { MongoClient } from "mongodb";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
-import { checkRequiredEnvVars } from "./shared";
+import { checkRequiredEnvVars, POPULATED_METADATA_FIELDS } from "./shared";
 
 dotenv.config();
 
@@ -30,6 +27,12 @@ async function loadAndIndex() {
     dbName: databaseName,
     collectionName: vectorCollectionName, // this is where your embeddings will be stored
     indexName: indexName, // this is the name of the index you will need to create
+    indexedMetadataFields: POPULATED_METADATA_FIELDS,
+    embeddingDefinition: {
+      dimensions: process.env.EMBEDDING_DIM
+        ? parseInt(process.env.EMBEDDING_DIM)
+        : 1536,
+    },
   });
 
   // now create an index from all the Documents and store them in Atlas

templates/components/vectordbs/typescript/mongo/index.ts

@@ -1,16 +1,23 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
-import { MongoDBAtlasVectorSearch, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { MongoDBAtlasVectorSearch } from "llamaindex/storage/vectorStore/MongoDBAtlasVectorStore";
 import { MongoClient } from "mongodb";
-import { checkRequiredEnvVars } from "./shared";
+import { checkRequiredEnvVars, POPULATED_METADATA_FIELDS } from "./shared";
 
 export async function getDataSource(params?: any) {
   checkRequiredEnvVars();
-  const client = new MongoClient(process.env.MONGO_URI!);
+  const client = new MongoClient(process.env.MONGODB_URI!);
   const store = new MongoDBAtlasVectorSearch({
     mongodbClient: client,
     dbName: process.env.MONGODB_DATABASE!,
     collectionName: process.env.MONGODB_VECTORS!,
     indexName: process.env.MONGODB_VECTOR_INDEX,
+    indexedMetadataFields: POPULATED_METADATA_FIELDS,
+    embeddingDefinition: {
+      dimensions: process.env.EMBEDDING_DIM
+        ? parseInt(process.env.EMBEDDING_DIM)
+        : 1536,
+    },
   });
 
   return await VectorStoreIndex.fromVectorStore(store);

templates/components/vectordbs/typescript/mongo/shared.ts

@@ -5,6 +5,8 @@ const REQUIRED_ENV_VARS = [
   "MONGODB_VECTOR_INDEX",
 ];
 
+export const POPULATED_METADATA_FIELDS = ["private", "doc_id"]; // for filtering in MongoDB VectorSearchIndex
+
 export function checkRequiredEnvVars() {
   const missingEnvVars = REQUIRED_ENV_VARS.filter((envVar) => {
     return !process.env[envVar];

templates/types/multiagent/fastapi/pyproject.toml

@@ -12,7 +12,8 @@ generate = "app.engine.generate:generate_datasource"
 [tool.poetry.dependencies]
 python = "^3.11"
 llama-index-agent-openai = ">=0.3.0,<0.4.0"
-llama-index = "^0.11.4"
+llama-index = "0.11.9"
+llama-index-core = "0.11.9"
 fastapi = "^0.112.2"
 python-dotenv = "^1.0.0"
 uvicorn = { extras = ["standard"], version = "^0.23.2" }

templates/types/streaming/express/package.json

@@ -15,12 +15,12 @@
     "dev": "concurrently \"tsup index.ts --format esm --dts --watch\" \"nodemon --watch dist/index.js\""
   },
   "dependencies": {
-    "ai": "^3.0.21",
+    "ai": "3.3.38",
     "cors": "^2.8.5",
     "dotenv": "^16.3.1",
     "duck-duck-scrape": "^2.2.5",
     "express": "^4.18.2",
-    "llamaindex": "0.5.20",
+    "llamaindex": "0.6.2",
     "pdf2json": "3.0.5",
     "ajv": "^8.12.0",
     "@e2b/code-interpreter": "^0.0.5",

templates/types/streaming/nextjs/app/api/chat/engine/settings.ts

@@ -1,185 +0,0 @@
-import {
-  ALL_AVAILABLE_MISTRAL_MODELS,
-  Anthropic,
-  GEMINI_EMBEDDING_MODEL,
-  GEMINI_MODEL,
-  Gemini,
-  GeminiEmbedding,
-  Groq,
-  MistralAI,
-  MistralAIEmbedding,
-  MistralAIEmbeddingModelType,
-  OpenAI,
-  OpenAIEmbedding,
-  Settings,
-} from "llamaindex";
-import { HuggingFaceEmbedding } from "llamaindex/embeddings/HuggingFaceEmbedding";
-import { OllamaEmbedding } from "llamaindex/embeddings/OllamaEmbedding";
-import { ALL_AVAILABLE_ANTHROPIC_MODELS } from "llamaindex/llm/anthropic";
-import { Ollama } from "llamaindex/llm/ollama";
-
-const CHUNK_SIZE = 512;
-const CHUNK_OVERLAP = 20;
-
-export const initSettings = async () => {
-  // HINT: you can delete the initialization code for unused model providers
-  console.log(`Using '${process.env.MODEL_PROVIDER}' model provider`);
-
-  if (!process.env.MODEL || !process.env.EMBEDDING_MODEL) {
-    throw new Error("'MODEL' and 'EMBEDDING_MODEL' env variables must be set.");
-  }
-
-  switch (process.env.MODEL_PROVIDER) {
-    case "ollama":
-      initOllama();
-      break;
-    case "groq":
-      initGroq();
-      break;
-    case "anthropic":
-      initAnthropic();
-      break;
-    case "gemini":
-      initGemini();
-      break;
-    case "mistral":
-      initMistralAI();
-      break;
-    case "azure-openai":
-      initAzureOpenAI();
-      break;
-    default:
-      initOpenAI();
-      break;
-  }
-  Settings.chunkSize = CHUNK_SIZE;
-  Settings.chunkOverlap = CHUNK_OVERLAP;
-};
-
-function initOpenAI() {
-  Settings.llm = new OpenAI({
-    model: process.env.MODEL ?? "gpt-4o-mini",
-    maxTokens: process.env.LLM_MAX_TOKENS
-      ? Number(process.env.LLM_MAX_TOKENS)
-      : undefined,
-  });
-  Settings.embedModel = new OpenAIEmbedding({
-    model: process.env.EMBEDDING_MODEL,
-    dimensions: process.env.EMBEDDING_DIM
-      ? parseInt(process.env.EMBEDDING_DIM)
-      : undefined,
-  });
-}
-
-function initAzureOpenAI() {
-  // Map Azure OpenAI model names to OpenAI model names (only for TS)
-  const AZURE_OPENAI_MODEL_MAP: Record<string, string> = {
-    "gpt-35-turbo": "gpt-3.5-turbo",
-    "gpt-35-turbo-16k": "gpt-3.5-turbo-16k",
-    "gpt-4o": "gpt-4o",
-    "gpt-4": "gpt-4",
-    "gpt-4-32k": "gpt-4-32k",
-    "gpt-4-turbo": "gpt-4-turbo",
-    "gpt-4-turbo-2024-04-09": "gpt-4-turbo",
-    "gpt-4-vision-preview": "gpt-4-vision-preview",
-    "gpt-4-1106-preview": "gpt-4-1106-preview",
-    "gpt-4o-2024-05-13": "gpt-4o-2024-05-13",
-  };
-
-  const azureConfig = {
-    apiKey: process.env.AZURE_OPENAI_KEY,
-    endpoint: process.env.AZURE_OPENAI_ENDPOINT,
-    apiVersion:
-      process.env.AZURE_OPENAI_API_VERSION || process.env.OPENAI_API_VERSION,
-  };
-
-  Settings.llm = new OpenAI({
-    model:
-      AZURE_OPENAI_MODEL_MAP[process.env.MODEL ?? "gpt-35-turbo"] ??
-      "gpt-3.5-turbo",
-    maxTokens: process.env.LLM_MAX_TOKENS
-      ? Number(process.env.LLM_MAX_TOKENS)
-      : undefined,
-    azure: {
-      ...azureConfig,
-      deployment: process.env.AZURE_OPENAI_LLM_DEPLOYMENT,
-    },
-  });
-
-  Settings.embedModel = new OpenAIEmbedding({
-    model: process.env.EMBEDDING_MODEL,
-    dimensions: process.env.EMBEDDING_DIM
-      ? parseInt(process.env.EMBEDDING_DIM)
-      : undefined,
-    azure: {
-      ...azureConfig,
-      deployment: process.env.AZURE_OPENAI_EMBEDDING_DEPLOYMENT,
-    },
-  });
-}
-
-function initOllama() {
-  const config = {
-    host: process.env.OLLAMA_BASE_URL ?? "http://127.0.0.1:11434",
-  };
-  Settings.llm = new Ollama({
-    model: process.env.MODEL ?? "",
-    config,
-  });
-  Settings.embedModel = new OllamaEmbedding({
-    model: process.env.EMBEDDING_MODEL ?? "",
-    config,
-  });
-}
-
-function initGroq() {
-  const embedModelMap: Record<string, string> = {
-    "all-MiniLM-L6-v2": "Xenova/all-MiniLM-L6-v2",
-    "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
-  };
-
-  const modelMap: Record<string, string> = {
-    "llama3-8b": "llama3-8b-8192",
-    "llama3-70b": "llama3-70b-8192",
-    "mixtral-8x7b": "mixtral-8x7b-32768",
-  };
-
-  Settings.llm = new Groq({
-    model: modelMap[process.env.MODEL!],
-  });
-
-  Settings.embedModel = new HuggingFaceEmbedding({
-    modelType: embedModelMap[process.env.EMBEDDING_MODEL!],
-  });
-}
-
-function initAnthropic() {
-  const embedModelMap: Record<string, string> = {
-    "all-MiniLM-L6-v2": "Xenova/all-MiniLM-L6-v2",
-    "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
-  };
-  Settings.llm = new Anthropic({
-    model: process.env.MODEL as keyof typeof ALL_AVAILABLE_ANTHROPIC_MODELS,
-  });
-  Settings.embedModel = new HuggingFaceEmbedding({
-    modelType: embedModelMap[process.env.EMBEDDING_MODEL!],
-  });
-}
-
-function initGemini() {
-  Settings.llm = new Gemini({
-    model: process.env.MODEL as GEMINI_MODEL,
-  });
-  Settings.embedModel = new GeminiEmbedding({
-    model: process.env.EMBEDDING_MODEL as GEMINI_EMBEDDING_MODEL,
-  });
-}
-
-function initMistralAI() {
-  Settings.llm = new MistralAI({
-    model: process.env.MODEL as keyof typeof ALL_AVAILABLE_MISTRAL_MODELS,
-  });
-  Settings.embedModel = new MistralAIEmbedding({
-    model: process.env.EMBEDDING_MODEL as MistralAIEmbeddingModelType,
-  });
-}

templates/types/streaming/nextjs/app/components/chat-section.tsx

@@ -29,6 +29,7 @@ export default function ChatSection() {
       const message = JSON.parse(error.message);
       alert(message.detail);
     },
+    sendExtraMessageFields: true,
   });
 
   return (

templates/types/streaming/nextjs/app/components/ui/chat/chat-message/codeblock.tsx

@@ -68,7 +68,7 @@ const CodeBlock: FC<Props> = memo(({ language, value }) => {
       3,
       true,
     )}${fileExtension}`;
-    const fileName = window.prompt("Enter file name" || "", suggestedFileName);
+    const fileName = window.prompt("Enter file name", suggestedFileName);
 
     if (!fileName) {
       // User pressed cancel on prompt.

templates/types/streaming/nextjs/app/components/ui/chat/chat-message/markdown.tsx

@@ -133,7 +133,11 @@ export default function Markdown({
               return <></>;
             }
           }
-          return <a href={href}>{children}</a>;
+          return (
+            <a href={href} target="_blank">
+              {children}
+            </a>
+          );
         },
       }}
     >

templates/types/streaming/nextjs/app/markdown.css

@@ -2,11 +2,15 @@
 .custom-markdown ul {
   list-style-type: disc;
   margin-left: 20px;
+  margin-top: 20px;
+  margin-bottom: 20px;
 }
 
 .custom-markdown ol {
   list-style-type: decimal;
   margin-left: 20px;
+  margin-top: 20px;
+  margin-bottom: 20px;
 }
 
 .custom-markdown li {
@@ -21,3 +25,55 @@
 .custom-markdown ol ol {
   margin-left: 20px;
 }
+
+.custom-markdown img {
+  border-radius: 8px;
+  box-shadow: 0 0 10px rgba(0, 0, 0, 0.1);
+  margin: 10px 0;
+}
+
+.custom-markdown a {
+  text-decoration: underline;
+  color: #007bff;
+}
+
+.custom-markdown h1,
+h2,
+h3,
+h4,
+h5,
+h6 {
+  font-weight: bold;
+  margin-bottom: 20px;
+  margin-top: 20px;
+}
+
+.custom-markdown h6 {
+  font-size: 16px;
+}
+
+.custom-markdown h5 {
+  font-size: 18px;
+}
+
+.custom-markdown h4 {
+  font-size: 20px;
+}
+
+.custom-markdown h3 {
+  font-size: 22px;
+}
+
+.custom-markdown h2 {
+  font-size: 24px;
+}
+
+.custom-markdown h1 {
+  font-size: 26px;
+}
+
+.custom-markdown hr {
+  border: 0;
+  border-top: 1px solid #e1e4e8;
+  margin: 20px 0;
+}

templates/types/streaming/nextjs/package.json

@@ -17,7 +17,7 @@
     "@radix-ui/react-hover-card": "^1.0.7",
     "@radix-ui/react-select": "^2.1.1",
     "@radix-ui/react-slot": "^1.0.2",
-    "ai": "^3.0.21",
+    "ai": "3.3.38",
     "ajv": "^8.12.0",
     "class-variance-authority": "^0.7.0",
     "clsx": "^2.1.1",
@@ -25,7 +25,7 @@
     "duck-duck-scrape": "^2.2.5",
     "formdata-node": "^6.0.3",
     "got": "^14.4.1",
-    "llamaindex": "0.5.20",
+    "llamaindex": "0.6.2",
     "lucide-react": "^0.294.0",
     "next": "^14.2.4",
     "react": "^18.2.0",