Release 0.3.27 (#492)

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

CHANGELOG.md

@@ -1,5 +1,102 @@
 # create-llama
 
+## 0.3.27
+
+### Patch Changes
+
+- b4e41aa: Add deep research over own documents use case (Python)
+
+## 0.3.26
+
+### Patch Changes
+
+- f73d46b: Fix missing copy of the multiagent code
+
+## 0.3.25
+
+### Patch Changes
+
+- 5450096: bump: react 19 stable
+
+## 0.3.24
+
+### Patch Changes
+
+- a84743c: Change --agents paramameter to --use-case
+- a84743c: Add LlamaCloud support for Reflex templates
+- a7a6592: Fix the npm issue on the full-stack Python template
+- fc5e56e: bump: code interpreter v1
+
+## 0.3.23
+
+### Patch Changes
+
+- 9077cae: Add contract review use case (Python)
+
+## 0.3.22
+
+### Patch Changes
+
+- 25667d4: Make OpenAPI spec usable by custom GPTs
+
+## 0.3.21
+
+### Patch Changes
+
+- 95227a7: Add query endpoint
+
+## 0.3.20
+
+### Patch Changes
+
+- 27d2499: Bump the LlamaCloud library and fix breaking changes (Python).
+
+## 0.3.19
+
+### Patch Changes
+
+- f9a057d: Add support multimodal indexes (e.g. from LlamaCloud)
+- aedd73d: bump: chat-ui
+
+## 0.3.18
+
+### Patch Changes
+
+- fe90a7e: chore: bump ai v4
+- 02b2473: Show streaming errors in Python, optimize system prompts for tool usage and set the weather tool as default for the Agentic RAG use case
+- 63e961e: Use auto_routed retriever mode for LlamaCloudIndex
+
+## 0.3.17
+
+### Patch Changes
+
+- 28c8808: Add fly.io deployment
+- 0a7dfcf: Generate NEXT_PUBLIC_CHAT_API for NextJS backend to specify alternative backend
+
+## 0.3.16
+
+### Patch Changes
+
+- 8b371d8: Set pydantic version to <2.10 to avoid incompatibility with llama-index.
+- 30fe269: Deactive duckduckgo tool for TS
+- 30fe269: Replace DuckDuckGo by Wikipedia tool for agentic template
+
+## 0.3.15
+
+### Patch Changes
+
+- fc5b266: Improve DX for Python template (use one deployment instead of two)
+- f8f97d2: Add support for python 3.13
+
+## 0.3.14
+
+### Patch Changes
+
+- 00f0b3a: fix: dont include user message in chat history
+- 4663dec: chore: bump react19 rc
+- 44b34fb: chore: update eslint 9, nextjs 15, react 19
+- 6925676: feat: use latest chat UI
+
 ## 0.3.13
 
 ### Patch Changes

create-app.ts

@@ -7,17 +7,16 @@ import { getOnline } from "./helpers/is-online";
 import { isWriteable } from "./helpers/is-writeable";
 import { makeDir } from "./helpers/make-dir";
 
-import fs from "fs";
 import terminalLink from "terminal-link";
 import type { InstallTemplateArgs, TemplateObservability } from "./helpers";
 import { installTemplate } from "./helpers";
-import { writeDevcontainer } from "./helpers/devcontainer";
 import { templatesDir } from "./helpers/dir";
 import { toolsRequireConfig } from "./helpers/tools";
+import { configVSCode } from "./helpers/vscode";
 
 export type InstallAppArgs = Omit<
   InstallTemplateArgs,
-  "appName" | "root" | "isOnline" | "customApiPath"
+  "appName" | "root" | "isOnline" | "port"
 > & {
   appPath: string;
   frontend: boolean;
@@ -35,13 +34,12 @@ export async function createApp({
   communityProjectConfig,
   llamapack,
   vectorDb,
-  externalPort,
   postInstallAction,
   dataSources,
   tools,
   useLlamaParse,
   observability,
-  agents,
+  useCase,
 }: InstallAppArgs): Promise<void> {
   const root = path.resolve(appPath);
 
@@ -81,40 +79,30 @@ export async function createApp({
     communityProjectConfig,
     llamapack,
     vectorDb,
-    externalPort,
     postInstallAction,
     dataSources,
     tools,
     useLlamaParse,
     observability,
-    agents,
+    useCase,
   };
 
-  if (frontend) {
-    // install backend
-    const backendRoot = path.join(root, "backend");
-    await makeDir(backendRoot);
-    await installTemplate({ ...args, root: backendRoot, backend: true });
+  // Install backend
+  await installTemplate({ ...args, backend: true });
+
+  if (frontend && framework === "fastapi") {
     // install frontend
-    const frontendRoot = path.join(root, "frontend");
+    const frontendRoot = path.join(root, ".frontend");
     await makeDir(frontendRoot);
     await installTemplate({
       ...args,
       root: frontendRoot,
       framework: "nextjs",
-      customApiPath: `http://localhost:${externalPort ?? 8000}/api/chat`,
       backend: false,
     });
-    // copy readme for fullstack
-    await fs.promises.copyFile(
-      path.join(templatesDir, "README-fullstack.md"),
-      path.join(root, "README.md"),
-    );
-  } else {
-    await installTemplate({ ...args, backend: true });
   }
 
-  await writeDevcontainer(root, templatesDir, framework, frontend);
+  await configVSCode(root, templatesDir, framework);
 
   process.chdir(root);
   if (tryGitInit(root)) {

e2e/python/resolve_dependencies.spec.ts

@@ -63,7 +63,6 @@ if (
             vectorDb,
             tools: "none",
             port: 3000,
-            externalPort: 8000,
             postInstallAction: "none",
             templateUI: undefined,
             appType: "--no-frontend",
@@ -101,7 +100,6 @@ if (
             vectorDb: "none",
             tools: tool,
             port: 3000,
-            externalPort: 8000,
             postInstallAction: "none",
             templateUI: undefined,
             appType: "--no-frontend",
@@ -135,7 +133,6 @@ if (
             vectorDb: "none",
             tools: "none",
             port: 3000,
-            externalPort: 8000,
             postInstallAction: "none",
             templateUI: undefined,
             appType: "--no-frontend",
@@ -169,7 +166,6 @@ if (
             vectorDb: "none",
             tools: "none",
             port: 3000,
-            externalPort: 8000,
             postInstallAction: "none",
             templateUI: undefined,
             appType: "--no-frontend",

e2e/shared/extractor_template.spec.ts

@@ -1,63 +0,0 @@
-/* eslint-disable turbo/no-undeclared-env-vars */
-import { expect, test } from "@playwright/test";
-import { ChildProcess } from "child_process";
-import fs from "fs";
-import path from "path";
-import { TemplateFramework } from "../../helpers";
-import { createTestDir, runCreateLlama } from "../utils";
-
-const templateFramework: TemplateFramework = process.env.FRAMEWORK
-  ? (process.env.FRAMEWORK as TemplateFramework)
-  : "fastapi";
-const dataSource: string = process.env.DATASOURCE
-  ? process.env.DATASOURCE
-  : "--example-file";
-
-// The extractor template currently only works with FastAPI and files (and not on Windows)
-if (
-  process.platform !== "win32" &&
-  templateFramework === "fastapi" &&
-  dataSource === "--example-file"
-) {
-  test.describe("Test extractor template", async () => {
-    let frontendPort: number;
-    let backendPort: number;
-    let name: string;
-    let appProcess: ChildProcess;
-    let cwd: string;
-
-    // Create extractor app
-    test.beforeAll(async () => {
-      cwd = await createTestDir();
-      frontendPort = Math.floor(Math.random() * 10000) + 10000;
-      backendPort = frontendPort + 1;
-      const result = await runCreateLlama({
-        cwd,
-        templateType: "extractor",
-        templateFramework: "fastapi",
-        dataSource: "--example-file",
-        vectorDb: "none",
-        port: frontendPort,
-        externalPort: backendPort,
-        postInstallAction: "runApp",
-      });
-      name = result.projectName;
-      appProcess = result.appProcess;
-    });
-
-    test.afterAll(async () => {
-      appProcess.kill();
-    });
-
-    test("App folder should exist", async () => {
-      const dirExists = fs.existsSync(path.join(cwd, name));
-      expect(dirExists).toBeTruthy();
-    });
-    test("Frontend should have a title", async ({ page }) => {
-      await page.goto(`http://localhost:${frontendPort}`);
-      await expect(page.getByText("Built by LlamaIndex")).toBeVisible({
-        timeout: 2000 * 60,
-      });
-    });
-  });
-}

e2e/shared/multiagent_template.spec.ts

@@ -16,18 +16,17 @@ const templateFramework: TemplateFramework = process.env.FRAMEWORK
 const dataSource: string = "--example-file";
 const templateUI: TemplateUI = "shadcn";
 const templatePostInstallAction: TemplatePostInstallAction = "runApp";
-const appType: AppType = templateFramework === "nextjs" ? "" : "--frontend";
+const appType: AppType = templateFramework === "fastapi" ? "--frontend" : "";
 const userMessage = "Write a blog post about physical standards for letters";
-const templateAgents = ["financial_report", "blog", "form_filling"];
+const templateUseCases = ["financial_report", "blog", "form_filling"];
 
-for (const agents of templateAgents) {
-  test.describe(`Test multiagent template ${agents} ${templateFramework} ${dataSource} ${templateUI} ${appType} ${templatePostInstallAction}`, async () => {
+for (const useCase of templateUseCases) {
+  test.describe(`Test multiagent template ${useCase} ${templateFramework} ${dataSource} ${templateUI} ${appType} ${templatePostInstallAction}`, async () => {
     test.skip(
       process.platform !== "linux" || process.env.DATASOURCE === "--no-files",
       "The multiagent template currently only works with files. We also only run on Linux to speed up tests.",
     );
     let port: number;
-    let externalPort: number;
     let cwd: string;
     let name: string;
     let appProcess: ChildProcess;
@@ -36,7 +35,6 @@ for (const agents of templateAgents) {
 
     test.beforeAll(async () => {
       port = Math.floor(Math.random() * 10000) + 10000;
-      externalPort = port + 1;
       cwd = await createTestDir();
       const result = await runCreateLlama({
         cwd,
@@ -45,11 +43,10 @@ for (const agents of templateAgents) {
         dataSource,
         vectorDb,
         port,
-        externalPort,
         postInstallAction: templatePostInstallAction,
         templateUI,
         appType,
-        agents,
+        useCase,
       });
       name = result.projectName;
       appProcess = result.appProcess;
@@ -61,6 +58,10 @@ for (const agents of templateAgents) {
     });
 
     test("Frontend should have a title", async ({ page }) => {
+      test.skip(
+        templatePostInstallAction !== "runApp" ||
+          templateFramework === "express",
+      );
       await page.goto(`http://localhost:${port}`);
       await expect(page.getByText("Built by LlamaIndex")).toBeVisible();
     });
@@ -69,7 +70,10 @@ for (const agents of templateAgents) {
       page,
     }) => {
       test.skip(
-        agents === "financial_report" || agents === "form_filling",
+        templatePostInstallAction !== "runApp" ||
+          useCase === "financial_report" ||
+          useCase === "form_filling" ||
+          templateFramework === "express",
         "Skip chat tests for financial report and form filling.",
       );
       await page.goto(`http://localhost:${port}`);

e2e/shared/reflex_template.spec.ts

@@ -0,0 +1,64 @@
+/* eslint-disable turbo/no-undeclared-env-vars */
+import { expect, test } from "@playwright/test";
+import { ChildProcess } from "child_process";
+import fs from "fs";
+import path from "path";
+import { TemplateFramework, TemplateUseCase } from "../../helpers";
+import { createTestDir, runCreateLlama } from "../utils";
+
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "fastapi";
+const dataSource: string = process.env.DATASOURCE
+  ? process.env.DATASOURCE
+  : "--example-file";
+const templateUseCases: TemplateUseCase[] = ["extractor", "contract_review"];
+
+// The reflex template currently only works with FastAPI and files (and not on Windows)
+if (
+  process.platform !== "win32" &&
+  templateFramework === "fastapi" &&
+  dataSource === "--example-file"
+) {
+  for (const useCase of templateUseCases) {
+    test.describe(`Test reflex template ${useCase} ${templateFramework} ${dataSource}`, async () => {
+      let appPort: number;
+      let name: string;
+      let appProcess: ChildProcess;
+      let cwd: string;
+
+      // Create reflex app
+      test.beforeAll(async () => {
+        cwd = await createTestDir();
+        appPort = Math.floor(Math.random() * 10000) + 10000;
+        const result = await runCreateLlama({
+          cwd,
+          templateType: "reflex",
+          templateFramework: "fastapi",
+          dataSource: "--example-file",
+          vectorDb: "none",
+          port: appPort,
+          postInstallAction: "runApp",
+          useCase,
+        });
+        name = result.projectName;
+        appProcess = result.appProcess;
+      });
+
+      test.afterAll(async () => {
+        appProcess.kill();
+      });
+
+      test("App folder should exist", async () => {
+        const dirExists = fs.existsSync(path.join(cwd, name));
+        expect(dirExists).toBeTruthy();
+      });
+      test("Frontend should have a title", async ({ page }) => {
+        await page.goto(`http://localhost:${appPort}`);
+        await expect(page.getByText("Built by LlamaIndex")).toBeVisible({
+          timeout: 2000 * 60,
+        });
+      });
+    });
+  }
+}

e2e/shared/streaming_template.spec.ts

@@ -22,7 +22,7 @@ const templatePostInstallAction: TemplatePostInstallAction = "runApp";
 const llamaCloudProjectName = "create-llama";
 const llamaCloudIndexName = "e2e-test";
 
-const appType: AppType = templateFramework === "nextjs" ? "" : "--frontend";
+const appType: AppType = templateFramework === "fastapi" ? "--frontend" : "";
 const userMessage =
   dataSource !== "--no-files" ? "Physical standard for letters" : "Hello";
 
@@ -35,7 +35,6 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
   }
 
   let port: number;
-  let externalPort: number;
   let cwd: string;
   let name: string;
   let appProcess: ChildProcess;
@@ -44,7 +43,6 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
 
   test.beforeAll(async () => {
     port = Math.floor(Math.random() * 10000) + 10000;
-    externalPort = port + 1;
     cwd = await createTestDir();
     const result = await runCreateLlama({
       cwd,
@@ -53,7 +51,6 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
       dataSource,
       vectorDb,
       port,
-      externalPort,
       postInstallAction: templatePostInstallAction,
       templateUI,
       appType,
@@ -68,8 +65,11 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
     const dirExists = fs.existsSync(path.join(cwd, name));
     expect(dirExists).toBeTruthy();
   });
+
   test("Frontend should have a title", async ({ page }) => {
-    test.skip(templatePostInstallAction !== "runApp");
+    test.skip(
+      templatePostInstallAction !== "runApp" || templateFramework === "express",
+    );
     await page.goto(`http://localhost:${port}`);
     await expect(page.getByText("Built by LlamaIndex")).toBeVisible();
   });
@@ -77,7 +77,9 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
   test("Frontend should be able to submit a message and receive a response", async ({
     page,
   }) => {
-    test.skip(templatePostInstallAction !== "runApp");
+    test.skip(
+      templatePostInstallAction !== "runApp" || templateFramework === "express",
+    );
     await page.goto(`http://localhost:${port}`);
     await page.fill("form textarea", userMessage);
     const [response] = await Promise.all([
@@ -102,7 +104,7 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
     test.skip(templatePostInstallAction !== "runApp");
     test.skip(templateFramework === "nextjs");
     const response = await request.post(
-      `http://localhost:${externalPort}/api/chat/request`,
+      `http://localhost:${port}/api/chat/request`,
       {
         data: {
           messages: [

e2e/typescript/resolve_dependencies.spec.ts

@@ -56,7 +56,6 @@ test.describe("Test resolve TS dependencies", () => {
       dataSource: dataSource,
       vectorDb: vectorDb,
       port: 3000,
-      externalPort: 8000,
       postInstallAction: "none",
       templateUI: undefined,
       appType: templateFramework === "nextjs" ? "" : "--no-frontend",

e2e/utils.ts

@@ -25,7 +25,6 @@ export type RunCreateLlamaOptions = {
   dataSource: string;
   vectorDb: TemplateVectorDB;
   port: number;
-  externalPort: number;
   postInstallAction: TemplatePostInstallAction;
   templateUI?: TemplateUI;
   appType?: AppType;
@@ -34,7 +33,7 @@ export type RunCreateLlamaOptions = {
   tools?: string;
   useLlamaParse?: boolean;
   observability?: string;
-  agents?: string;
+  useCase?: string;
 };
 
 export async function runCreateLlama({
@@ -44,7 +43,6 @@ export async function runCreateLlama({
   dataSource,
   vectorDb,
   port,
-  externalPort,
   postInstallAction,
   templateUI,
   appType,
@@ -53,7 +51,7 @@ export async function runCreateLlama({
   tools,
   useLlamaParse,
   observability,
-  agents,
+  useCase,
 }: RunCreateLlamaOptions): Promise<CreateLlamaResult> {
   if (!process.env.OPENAI_API_KEY || !process.env.LLAMA_CLOUD_API_KEY) {
     throw new Error(
@@ -90,21 +88,15 @@ export async function runCreateLlama({
     ...dataSourceArgs,
     "--vector-db",
     vectorDb,
-    "--open-ai-key",
-    process.env.OPENAI_API_KEY,
-    "--use-pnpm",
+    "--use-npm",
     "--port",
     port,
-    "--external-port",
-    externalPort,
     "--post-install-action",
     postInstallAction,
     "--tools",
     tools ?? "none",
     "--observability",
     "none",
-    "--llama-cloud-key",
-    process.env.LLAMA_CLOUD_API_KEY,
   ];
 
   if (templateUI) {
@@ -121,8 +113,8 @@ export async function runCreateLlama({
   if (observability) {
     commandArgs.push("--observability", observability);
   }
-  if (templateType === "multiagent" && agents) {
-    commandArgs.push("--agents", agents);
+  if ((templateType === "multiagent" || templateType === "reflex") && useCase) {
+    commandArgs.push("--use-case", useCase);
   }
 
   const command = commandArgs.join(" ");
@@ -146,12 +138,7 @@ export async function runCreateLlama({
 
   // Wait for app to start
   if (postInstallAction === "runApp") {
-    await checkAppHasStarted(
-      appType === "--frontend",
-      templateFramework,
-      port,
-      externalPort,
-    );
+    await waitPorts([port]);
   } else if (postInstallAction === "dependencies") {
     await waitForProcess(appProcess, 1000 * 60); // wait 1 min for dependencies to be resolved
   } else {
@@ -171,19 +158,6 @@ export async function createTestDir() {
   return cwd;
 }
 
-// eslint-disable-next-line max-params
-async function checkAppHasStarted(
-  frontend: boolean,
-  framework: TemplateFramework,
-  port: number,
-  externalPort: number,
-) {
-  const portsToWait = frontend
-    ? [port, externalPort]
-    : [framework === "nextjs" ? port : externalPort];
-  await waitPorts(portsToWait);
-}
-
 async function waitPorts(ports: number[]): Promise<void> {
   const waitForPort = async (port: number): Promise<void> => {
     await waitPort({

helpers/copy.ts

@@ -61,6 +61,9 @@ export const assetRelocator = (name: string) => {
     case "README-template.md": {
       return "README.md";
     }
+    case "vscode_settings.json": {
+      return "settings.json";
+    }
     default: {
       return name;
     }

helpers/datasources.ts

@@ -18,6 +18,7 @@ export const EXAMPLE_10K_SEC_FILES: TemplateDataSource[] = [
       url: new URL(
         "https://s2.q4cdn.com/470004039/files/doc_earnings/2023/q4/filing/_10-K-Q4-2023-As-Filed.pdf",
       ),
+      filename: "apple_10k_report.pdf",
     },
   },
   {
@@ -26,10 +27,31 @@ export const EXAMPLE_10K_SEC_FILES: TemplateDataSource[] = [
       url: new URL(
         "https://ir.tesla.com/_flysystem/s3/sec/000162828024002390/tsla-20231231-gen.pdf",
       ),
+      filename: "tesla_10k_report.pdf",
     },
   },
 ];
 
+export const EXAMPLE_GDPR: TemplateDataSource = {
+  type: "file",
+  config: {
+    url: new URL(
+      "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32016R0679",
+    ),
+    filename: "gdpr.pdf",
+  },
+};
+
+export const AI_REPORTS: TemplateDataSource = {
+  type: "file",
+  config: {
+    url: new URL(
+      "https://www.europarl.europa.eu/RegData/etudes/ATAG/2024/760392/EPRS_ATA(2024)760392_EN.pdf",
+    ),
+    filename: "EPRS_ATA_2024_760392_EN.pdf",
+  },
+};
+
 export function getDataSources(
   files?: string,
   exampleFile?: boolean,

helpers/env-variables.ts

@@ -13,6 +13,12 @@ import {
 
 import { TSYSTEMS_LLMHUB_API_URL } from "./providers/llmhub";
 
+const DEFAULT_SYSTEM_PROMPT =
+  "You are a helpful assistant who helps users with their questions.";
+
+const DATA_SOURCES_PROMPT =
+  "You have access to a knowledge base including the facts that you should start with to find the answer for the user question. Use the query engine tool to retrieve the facts from the knowledge base.";
+
 export type EnvVar = {
   name?: string;
   description?: string;
@@ -407,6 +413,13 @@ const getFrameworkEnvs = (
       ],
     );
   }
+  if (framework === "nextjs") {
+    result.push({
+      name: "NEXT_PUBLIC_CHAT_API",
+      description:
+        "The API for the chat endpoint. Set when using a custom backend (e.g. Express). Use full URL like http://localhost:8000/api/chat",
+    });
+  }
   return result;
 };
 
@@ -442,9 +455,6 @@ const getSystemPromptEnv = (
   dataSources?: TemplateDataSource[],
   template?: TemplateType,
 ): EnvVar[] => {
-  const defaultSystemPrompt =
-    "You are a helpful assistant who helps users with their questions.";
-
   const systemPromptEnv: EnvVar[] = [];
   // build tool system prompt by merging all tool system prompts
   // multiagent template doesn't need system prompt
@@ -459,9 +469,12 @@ const getSystemPromptEnv = (
       }
     });
 
-    const systemPrompt = toolSystemPrompt
-      ? `\"${toolSystemPrompt}\"`
-      : defaultSystemPrompt;
+    const systemPrompt =
+      '"' +
+      DEFAULT_SYSTEM_PROMPT +
+      (dataSources?.length ? `\n${DATA_SOURCES_PROMPT}` : "") +
+      (toolSystemPrompt ? `\n${toolSystemPrompt}` : "") +
+      '"';
 
     systemPromptEnv.push({
       name: "SYSTEM_PROMPT",
@@ -512,7 +525,7 @@ Here is the conversation history
 ---------------------
 {conversation}
 ---------------------
-Given the conversation history, please give me 3 questions that you might ask next!
+Given the conversation history, please give me 3 questions that user might ask next!
 Your answer should be wrapped in three sticks which follows the following format:
 \`\`\`
 <question 1>
@@ -553,7 +566,7 @@ export const createBackendEnvFile = async (
     | "framework"
     | "dataSources"
     | "template"
-    | "externalPort"
+    | "port"
     | "tools"
     | "observability"
   >,
@@ -570,7 +583,7 @@ export const createBackendEnvFile = async (
     ...getModelEnvs(opts.modelConfig),
     ...getEngineEnvs(),
     ...getVectorDBEnvs(opts.vectorDb, opts.framework),
-    ...getFrameworkEnvs(opts.framework, opts.externalPort),
+    ...getFrameworkEnvs(opts.framework, opts.port),
     ...getToolEnvs(opts.tools),
     ...getTemplateEnvs(opts.template),
     ...getObservabilityEnvs(opts.observability),
@@ -585,18 +598,10 @@ export const createBackendEnvFile = async (
 export const createFrontendEnvFile = async (
   root: string,
   opts: {
-    customApiPath?: string;
     vectorDb?: TemplateVectorDB;
   },
 ) => {
   const defaultFrontendEnvs = [
-    {
-      name: "NEXT_PUBLIC_CHAT_API",
-      description: "The backend API for chat endpoint.",
-      value: opts.customApiPath
-        ? opts.customApiPath
-        : "http://localhost:8000/api/chat",
-    },
     {
       name: "NEXT_PUBLIC_USE_LLAMACLOUD",
       description: "Let's the user change indexes in LlamaCloud projects",

helpers/index.ts

@@ -118,7 +118,8 @@ const prepareContextData = async (
       const destPath = path.join(
         root,
         "data",
-        path.basename(dataSourceConfig.url.toString()),
+        dataSourceConfig.filename ??
+          path.basename(dataSourceConfig.url.toString()),
       );
       await downloadFile(dataSourceConfig.url.toString(), destPath);
     } else {
@@ -192,7 +193,7 @@ export const installTemplate = async (
     if (
       props.template === "streaming" ||
       props.template === "multiagent" ||
-      props.template === "extractor"
+      props.template === "reflex"
     ) {
       await createBackendEnvFile(props.root, props);
     }
@@ -225,7 +226,6 @@ export const installTemplate = async (
   } else {
     // this is a frontend for a full-stack app, create .env file with model information
     await createFrontendEnvFile(props.root, {
-      customApiPath: props.customApiPath,
       vectorDb: props.vectorDb,
     });
   }

helpers/providers/openai.ts

@@ -3,6 +3,7 @@ import ora from "ora";
 import { red } from "picocolors";
 import prompts from "prompts";
 import { ModelConfigParams, ModelConfigQuestionsParams } from ".";
+import { isCI } from "../../questions";
 import { questionHandlers } from "../../questions/utils";
 
 const OPENAI_API_URL = "https://api.openai.com/v1";
@@ -30,7 +31,7 @@ export async function askOpenAIQuestions({
     },
   };
 
-  if (!config.apiKey) {
+  if (!config.apiKey && !isCI) {
     const { key } = await prompts(
       {
         type: "text",

helpers/python.ts

@@ -20,6 +20,7 @@ interface Dependency {
   name: string;
   version?: string;
   extras?: string[];
+  constraints?: Record<string, string>;
 }
 
 const getAdditionalDependencies = (
@@ -36,28 +37,31 @@ const getAdditionalDependencies = (
     case "mongo": {
       dependencies.push({
         name: "llama-index-vector-stores-mongodb",
-        version: "^0.3.1",
+        version: "^0.6.0",
       });
       break;
     }
     case "pg": {
       dependencies.push({
         name: "llama-index-vector-stores-postgres",
-        version: "^0.2.5",
+        version: "^0.3.2",
       });
       break;
     }
     case "pinecone": {
       dependencies.push({
         name: "llama-index-vector-stores-pinecone",
-        version: "^0.2.1",
+        version: "^0.4.1",
+        constraints: {
+          python: ">=3.11,<3.13",
+        },
       });
       break;
     }
     case "milvus": {
       dependencies.push({
         name: "llama-index-vector-stores-milvus",
-        version: "^0.2.0",
+        version: "^0.3.0",
       });
       dependencies.push({
         name: "pymilvus",
@@ -68,35 +72,38 @@ const getAdditionalDependencies = (
     case "astra": {
       dependencies.push({
         name: "llama-index-vector-stores-astra-db",
-        version: "^0.2.0",
+        version: "^0.4.0",
       });
       break;
     }
     case "qdrant": {
       dependencies.push({
         name: "llama-index-vector-stores-qdrant",
-        version: "^0.3.0",
+        version: "^0.4.0",
+        constraints: {
+          python: ">=3.11,<3.13",
+        },
       });
       break;
     }
     case "chroma": {
       dependencies.push({
         name: "llama-index-vector-stores-chroma",
-        version: "^0.2.0",
+        version: "^0.4.0",
       });
       break;
     }
     case "weaviate": {
       dependencies.push({
         name: "llama-index-vector-stores-weaviate",
-        version: "^1.1.1",
+        version: "^1.2.3",
       });
       break;
     }
     case "llamacloud":
       dependencies.push({
         name: "llama-index-indices-managed-llama-cloud",
-        version: "^0.3.1",
+        version: "^0.6.3",
       });
       break;
   }
@@ -115,13 +122,13 @@ const getAdditionalDependencies = (
         case "web":
           dependencies.push({
             name: "llama-index-readers-web",
-            version: "^0.2.2",
+            version: "^0.3.0",
           });
           break;
         case "db":
           dependencies.push({
             name: "llama-index-readers-database",
-            version: "^0.2.0",
+            version: "^0.3.0",
           });
           dependencies.push({
             name: "pymysql",
@@ -160,15 +167,15 @@ const getAdditionalDependencies = (
       if (templateType !== "multiagent") {
         dependencies.push({
           name: "llama-index-llms-openai",
-          version: "^0.2.0",
+          version: "^0.3.2",
         });
         dependencies.push({
           name: "llama-index-embeddings-openai",
-          version: "^0.2.3",
+          version: "^0.3.1",
         });
         dependencies.push({
           name: "llama-index-agent-openai",
-          version: "^0.3.0",
+          version: "^0.4.0",
         });
       }
       break;
@@ -279,14 +286,19 @@ const mergePoetryDependencies = (
     value.version = dependency.version ?? value.version;
     value.extras = dependency.extras ?? value.extras;
 
+    // Merge constraints if they exist
+    if (dependency.constraints) {
+      value = { ...value, ...dependency.constraints };
+    }
+
     if (value.version === undefined) {
       throw new Error(
         `Dependency "${dependency.name}" is missing attribute "version"!`,
       );
     }
 
-    // Serialize separately only if extras are provided
-    if (value.extras && value.extras.length > 0) {
+    // Serialize as object if there are any additional properties
+    if (Object.keys(value).length > 1) {
       existingDependencies[dependency.name] = value;
     } else {
       // Otherwise, serialize just the version string
@@ -368,33 +380,37 @@ export const installPythonDependencies = (
 };
 
 export const installPythonTemplate = async ({
+  appName,
   root,
   template,
   framework,
   vectorDb,
+  postInstallAction,
+  modelConfig,
   dataSources,
   tools,
-  postInstallAction,
+  useLlamaParse,
+  useCase,
   observability,
-  modelConfig,
-  agents,
 }: Pick<
   InstallTemplateArgs,
+  | "appName"
   | "root"
-  | "framework"
   | "template"
+  | "framework"
   | "vectorDb"
+  | "postInstallAction"
+  | "modelConfig"
   | "dataSources"
   | "tools"
-  | "postInstallAction"
+  | "useLlamaParse"
+  | "useCase"
   | "observability"
-  | "modelConfig"
-  | "agents"
 >) => {
   console.log("\nInitializing Python project with template:", template, "\n");
   let templatePath;
-  if (template === "extractor") {
-    templatePath = path.join(templatesDir, "types", "extractor", framework);
+  if (template === "reflex") {
+    templatePath = path.join(templatesDir, "types", "reflex");
   } else {
     templatePath = path.join(templatesDir, "types", "streaming", framework);
   }
@@ -460,37 +476,39 @@ export const installPythonTemplate = async ({
       cwd: path.join(compPath, "engines", "python", engine),
     });
 
-    // Copy agent code
-    if (template === "multiagent") {
-      if (agents) {
-        await copy("**", path.join(root), {
-          parents: true,
-          cwd: path.join(compPath, "agents", "python", agents),
-          rename: assetRelocator,
-        });
-      } else {
-        console.log(
-          red(
-            "There is no agent selected for multi-agent template. Please pick an agent to use via --agents flag.",
-          ),
-        );
-        process.exit(1);
-      }
-    }
-
     // Copy router code
     await copyRouterCode(root, tools ?? []);
   }
 
+  // Copy multiagents overrides
   if (template === "multiagent") {
-    // Copy multi-agent code
     await copy("**", path.join(root), {
-      parents: true,
       cwd: path.join(compPath, "multiagent", "python"),
-      rename: assetRelocator,
     });
   }
 
+  if (template === "multiagent" || template === "reflex") {
+    if (useCase) {
+      const sourcePath =
+        template === "multiagent"
+          ? path.join(compPath, "agents", "python", useCase)
+          : path.join(compPath, "reflex", useCase);
+
+      await copy("**", path.join(root), {
+        parents: true,
+        cwd: sourcePath,
+        rename: assetRelocator,
+      });
+    } else {
+      console.log(
+        red(
+          `There is no use case selected for ${template} template. Please pick a use case to use via --use-case flag.`,
+        ),
+      );
+      process.exit(1);
+    }
+  }
+
   console.log("Adding additional dependencies");
 
   const addOnDependencies = getAdditionalDependencies(
@@ -512,7 +530,10 @@ export const installPythonTemplate = async ({
     if (observability === "llamatrace") {
       addOnDependencies.push({
         name: "llama-index-callbacks-arize-phoenix",
-        version: "^0.2.1",
+        version: "^0.3.0",
+        constraints: {
+          python: ">=3.11,<3.13",
+        },
       });
     }
 
@@ -533,9 +554,4 @@ export const installPythonTemplate = async ({
   if (postInstallAction === "runApp" || postInstallAction === "dependencies") {
     installPythonDependencies();
   }
-
-  // Copy deployment files for python
-  await copy("**", root, {
-    cwd: path.join(compPath, "deployments", "python"),
-  });
 };

helpers/run-app.ts

@@ -1,40 +1,39 @@
-import { ChildProcess, SpawnOptions, spawn } from "child_process";
-import path from "path";
-import { TemplateFramework } from "./types";
+import { SpawnOptions, spawn } from "child_process";
+import { TemplateFramework, TemplateType } from "./types";
 
 const createProcess = (
   command: string,
   args: string[],
   options: SpawnOptions,
-) => {
-  return spawn(command, args, {
-    ...options,
-    shell: true,
-  })
-    .on("exit", function (code) {
-      if (code !== 0) {
-        console.log(`Child process exited with code=${code}`);
-        process.exit(1);
-      }
+): Promise<void> => {
+  return new Promise((resolve, reject) => {
+    spawn(command, args, {
+      ...options,
+      shell: true,
     })
-    .on("error", function (err) {
-      console.log("Error when running chill process: ", err);
-      process.exit(1);
-    });
+      .on("exit", function (code) {
+        if (code !== 0) {
+          console.log(`Child process exited with code=${code}`);
+          reject(code);
+        } else {
+          resolve();
+        }
+      })
+      .on("error", function (err) {
+        console.log("Error when running child process: ", err);
+        reject(err);
+      });
+  });
 };
 
-export function runReflexApp(
-  appPath: string,
-  frontendPort?: number,
-  backendPort?: number,
-) {
-  const commandArgs = ["run", "reflex", "run"];
-  if (frontendPort) {
-    commandArgs.push("--frontend-port", frontendPort.toString());
-  }
-  if (backendPort) {
-    commandArgs.push("--backend-port", backendPort.toString());
-  }
+export function runReflexApp(appPath: string, port: number) {
+  const commandArgs = [
+    "run",
+    "reflex",
+    "run",
+    "--frontend-port",
+    port.toString(),
+  ];
   return createProcess("poetry", commandArgs, {
     stdio: "inherit",
     cwd: appPath,
@@ -42,11 +41,10 @@ export function runReflexApp(
 }
 
 export function runFastAPIApp(appPath: string, port: number) {
-  const commandArgs = ["run", "uvicorn", "main:app", "--port=" + port];
-
-  return createProcess("poetry", commandArgs, {
+  return createProcess("poetry", ["run", "dev"], {
     stdio: "inherit",
     cwd: appPath,
+    env: { ...process.env, APP_PORT: `${port}` },
   });
 }
 
@@ -60,40 +58,24 @@ export function runTSApp(appPath: string, port: number) {
 
 export async function runApp(
   appPath: string,
-  template: string,
-  frontend: boolean,
+  template: TemplateType,
   framework: TemplateFramework,
   port?: number,
-  externalPort?: number,
-): Promise<any> {
-  const processes: ChildProcess[] = [];
+): Promise<void> {
+  try {
+    // Start the app
+    const defaultPort =
+      framework === "nextjs" || template === "reflex" ? 3000 : 8000;
 
-  // Callback to kill all sub processes if the main process is killed
-  process.on("exit", () => {
-    console.log("Killing app processes...");
-    processes.forEach((p) => p.kill());
-  });
-
-  // Default sub app paths
-  const backendPath = path.join(appPath, "backend");
-  const frontendPath = path.join(appPath, "frontend");
-
-  if (template === "extractor") {
-    processes.push(runReflexApp(appPath, port, externalPort));
+    const appRunner =
+      template === "reflex"
+        ? runReflexApp
+        : framework === "fastapi"
+          ? runFastAPIApp
+          : runTSApp;
+    await appRunner(appPath, port || defaultPort);
+  } catch (error) {
+    console.error("Failed to run app:", error);
+    throw error;
   }
-  if (template === "streaming" || template === "multiagent") {
-    if (framework === "fastapi" || framework === "express") {
-      const backendRunner = framework === "fastapi" ? runFastAPIApp : runTSApp;
-      if (frontend) {
-        processes.push(backendRunner(backendPath, externalPort || 8000));
-        processes.push(runTSApp(frontendPath, port || 3000));
-      } else {
-        processes.push(backendRunner(appPath, externalPort || 8000));
-      }
-    } else if (framework === "nextjs") {
-      processes.push(runTSApp(appPath, port || 3000));
-    }
-  }
-
-  return Promise.all(processes);
 }

helpers/tools.ts

@@ -41,7 +41,7 @@ export const supportedTools: Tool[] = [
     dependencies: [
       {
         name: "llama-index-tools-google",
-        version: "^0.2.0",
+        version: "^0.3.0",
       },
     ],
     supportedFrameworks: ["fastapi"],
@@ -62,17 +62,16 @@ export const supportedTools: Tool[] = [
     dependencies: [
       {
         name: "duckduckgo-search",
-        version: "6.1.7",
+        version: "^6.3.5",
       },
     ],
-    supportedFrameworks: ["fastapi", "nextjs", "express"],
+    supportedFrameworks: ["fastapi"], // TODO: Re-enable this tool once the duck-duck-scrape TypeScript library works again
     type: ToolType.LOCAL,
     envVars: [
       {
         name: TOOL_SYSTEM_PROMPT_ENV_VAR,
         description: "System prompt for DuckDuckGo search tool.",
-        value: `You are a DuckDuckGo search agent. 
-You can use the duckduckgo search tool to get information from the web to answer user questions.
+        value: `You have access to the duckduckgo search tool. Use it to get information from the web to answer user questions.
 For better results, you can specify the region parameter to get results from a specific region but it's optional.`,
       },
     ],
@@ -83,18 +82,11 @@ For better results, you can specify the region parameter to get results from a s
     dependencies: [
       {
         name: "llama-index-tools-wikipedia",
-        version: "^0.2.0",
+        version: "^0.3.0",
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
     type: ToolType.LLAMAHUB,
-    envVars: [
-      {
-        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
-        description: "System prompt for wiki tool.",
-        value: `You are a Wikipedia agent. You help users to get information from Wikipedia.`,
-      },
-    ],
   },
   {
     display: "Weather",
@@ -102,13 +94,6 @@ For better results, you can specify the region parameter to get results from a s
     dependencies: [],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
     type: ToolType.LOCAL,
-    envVars: [
-      {
-        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
-        description: "System prompt for weather tool.",
-        value: `You are a weather forecast agent. You help users to get the weather forecast for a given location.`,
-      },
-    ],
   },
   {
     display: "Document generator",
@@ -139,7 +124,7 @@ For better results, you can specify the region parameter to get results from a s
     dependencies: [
       {
         name: "e2b_code_interpreter",
-        version: "0.0.11b38",
+        version: "1.0.3",
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
@@ -170,7 +155,7 @@ For better results, you can specify the region parameter to get results from a s
     dependencies: [
       {
         name: "e2b_code_interpreter",
-        version: "0.0.11b38",
+        version: "1.0.3",
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
@@ -211,14 +196,6 @@ For better results, you can specify the region parameter to get results from a s
     },
     supportedFrameworks: ["fastapi", "express", "nextjs"],
     type: ToolType.LOCAL,
-    envVars: [
-      {
-        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
-        description: "System prompt for openapi action tool.",
-        value:
-          "You are an OpenAPI action agent. You help users to make requests to the provided OpenAPI schema.",
-      },
-    ],
   },
   {
     display: "Image Generator",
@@ -231,11 +208,6 @@ For better results, you can specify the region parameter to get results from a s
         description:
           "STABILITY_API_KEY key is required to run image generator. Get it here: https://platform.stability.ai/account/keys",
       },
-      {
-        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
-        description: "System prompt for image generator tool.",
-        value: `You are an image generator agent. You help users to generate images using the Stability API.`,
-      },
     ],
   },
   {

helpers/types.ts

@@ -20,11 +20,11 @@ export type ModelConfig = {
   isConfigured(): boolean;
 };
 export type TemplateType =
-  | "extractor"
   | "streaming"
   | "community"
   | "llamapack"
-  | "multiagent";
+  | "multiagent"
+  | "reflex";
 export type TemplateFramework = "nextjs" | "express" | "fastapi";
 export type TemplateUI = "html" | "shadcn";
 export type TemplateVectorDB =
@@ -49,14 +49,22 @@ export type TemplateDataSource = {
 };
 export type TemplateDataSourceType = "file" | "web" | "db";
 export type TemplateObservability = "none" | "traceloop" | "llamatrace";
-export type TemplateAgents = "financial_report" | "blog" | "form_filling";
+export type TemplateUseCase =
+  | "financial_report"
+  | "blog"
+  | "deep_research"
+  | "form_filling"
+  | "extractor"
+  | "contract_review";
 // Config for both file and folder
 export type FileSourceConfig =
   | {
       path: string;
+      filename?: string;
     }
   | {
       url: URL;
+      filename?: string;
     };
 export type WebSourceConfig = {
   baseUrl?: string;
@@ -89,16 +97,15 @@ export interface InstallTemplateArgs {
   framework: TemplateFramework;
   ui: TemplateUI;
   dataSources: TemplateDataSource[];
-  customApiPath?: string;
   modelConfig: ModelConfig;
   llamaCloudKey?: string;
   useLlamaParse?: boolean;
   communityProjectConfig?: CommunityProjectConfig;
   llamapack?: string;
   vectorDb?: TemplateVectorDB;
-  externalPort?: number;
+  port?: number;
   postInstallAction?: TemplatePostInstallAction;
   tools?: Tool[];
   observability?: TemplateObservability;
-  agents?: TemplateAgents;
+  useCase?: TemplateUseCase;
 }

helpers/typescript.ts

@@ -26,7 +26,7 @@ export const installTSTemplate = async ({
   tools,
   dataSources,
   useLlamaParse,
-  agents,
+  useCase,
 }: InstallTemplateArgs & { backend: boolean }) => {
   console.log(bold(`Using ${packageManager}.`));
 
@@ -58,11 +58,9 @@ export const installTSTemplate = async ({
       console.log("\nUsing static site generation\n");
     } else {
       if (vectorDb === "milvus") {
-        nextConfigJson.experimental.serverComponentsExternalPackages =
-          nextConfigJson.experimental.serverComponentsExternalPackages ?? [];
-        nextConfigJson.experimental.serverComponentsExternalPackages.push(
-          "@zilliz/milvus2-sdk-node",
-        );
+        nextConfigJson.serverExternalPackages =
+          nextConfigJson.serverExternalPackages ?? [];
+        nextConfigJson.serverExternalPackages.push("@zilliz/milvus2-sdk-node");
       }
     }
     await fs.writeFile(
@@ -133,16 +131,16 @@ export const installTSTemplate = async ({
       cwd: path.join(multiagentPath, "workflow"),
     });
 
-    // Copy agents use case code for multiagent template
-    if (agents) {
-      console.log("\nCopying agent:", agents, "\n");
-      const useCasePath = path.join(compPath, "agents", "typescript", agents);
-      const agentsCodePath = path.join(useCasePath, "workflow");
+    // Copy use case code for multiagent template
+    if (useCase) {
+      console.log("\nCopying use case:", useCase, "\n");
+      const useCasePath = path.join(compPath, "agents", "typescript", useCase);
+      const useCaseCodePath = path.join(useCasePath, "workflow");
 
-      // Copy agent codes
+      // Copy use case codes
       await copy("**", path.join(root, relativeEngineDestPath, "workflow"), {
         parents: true,
-        cwd: agentsCodePath,
+        cwd: useCaseCodePath,
         rename: assetRelocator,
       });
 
@@ -155,7 +153,7 @@ export const installTSTemplate = async ({
     } else {
       console.log(
         red(
-          "There is no agent selected for multi-agent template. Please pick an agent to use via --agents flag.",
+          `There is no use case selected for ${template} template. Please pick a use case to use via --use-case flag.`,
         ),
       );
       process.exit(1);
@@ -243,14 +241,12 @@ export const installTSTemplate = async ({
     vectorDb,
   });
 
-  if (postInstallAction === "runApp" || postInstallAction === "dependencies") {
+  if (
+    backend &&
+    (postInstallAction === "runApp" || postInstallAction === "dependencies")
+  ) {
     await installTSDependencies(packageJson, packageManager, isOnline);
   }
-
-  // Copy deployment files for typescript
-  await copy("**", root, {
-    cwd: path.join(compPath, "deployments", "typescript"),
-  });
 };
 
 async function updatePackageJson({

helpers/vscode.ts

@@ -1,40 +1,26 @@
 import fs from "fs";
 import path from "path";
+import { assetRelocator, copy } from "./copy";
 import { TemplateFramework } from "./types";
 
 function renderDevcontainerContent(
   templatesDir: string,
   framework: TemplateFramework,
-  frontend: boolean,
 ) {
   const devcontainerJson: any = JSON.parse(
     fs.readFileSync(path.join(templatesDir, "devcontainer.json"), "utf8"),
   );
 
   // Modify postCreateCommand
-  if (frontend) {
-    devcontainerJson.postCreateCommand =
-      framework === "fastapi"
-        ? "cd backend && poetry install && cd ../frontend && npm install"
-        : "cd backend && npm install && cd ../frontend && npm install";
-  } else {
-    devcontainerJson.postCreateCommand =
-      framework === "fastapi" ? "poetry install" : "npm install";
-  }
+  devcontainerJson.postCreateCommand =
+    framework === "fastapi" ? "poetry install" : "npm install";
 
   // Modify containerEnv
   if (framework === "fastapi") {
-    if (frontend) {
-      devcontainerJson.containerEnv = {
-        ...devcontainerJson.containerEnv,
-        PYTHONPATH: "${PYTHONPATH}:${workspaceFolder}/backend",
-      };
-    } else {
-      devcontainerJson.containerEnv = {
-        ...devcontainerJson.containerEnv,
-        PYTHONPATH: "${PYTHONPATH}:${workspaceFolder}",
-      };
-    }
+    devcontainerJson.containerEnv = {
+      ...devcontainerJson.containerEnv,
+      PYTHONPATH: "${PYTHONPATH}:${workspaceFolder}",
+    };
   }
 
   return JSON.stringify(devcontainerJson, null, 2);
@@ -44,7 +30,6 @@ export const writeDevcontainer = async (
   root: string,
   templatesDir: string,
   framework: TemplateFramework,
-  frontend: boolean,
 ) => {
   const devcontainerDir = path.join(root, ".devcontainer");
   if (fs.existsSync(devcontainerDir)) {
@@ -54,7 +39,6 @@ export const writeDevcontainer = async (
   const devcontainerContent = renderDevcontainerContent(
     templatesDir,
     framework,
-    frontend,
   );
   fs.mkdirSync(devcontainerDir);
   await fs.promises.writeFile(
@@ -62,3 +46,25 @@ export const writeDevcontainer = async (
     devcontainerContent,
   );
 };
+
+export const copyVSCodeSettings = async (
+  root: string,
+  templatesDir: string,
+) => {
+  const vscodeDir = path.join(root, ".vscode");
+  await copy("vscode_settings.json", vscodeDir, {
+    cwd: templatesDir,
+    rename: assetRelocator,
+  });
+};
+
+export const configVSCode = async (
+  root: string,
+  templatesDir: string,
+  framework: TemplateFramework,
+) => {
+  await writeDevcontainer(root, templatesDir, framework);
+  if (framework === "fastapi") {
+    await copyVSCodeSettings(root, templatesDir);
+  }
+};

index.ts

@@ -134,13 +134,6 @@ const program = new Command(packageJson.name)
     `
 
   Select UI port.
-`,
-  )
-  .option(
-    "--external-port <external>",
-    `
-
-  Select external port.
 `,
   )
   .option(
@@ -209,10 +202,10 @@ const program = new Command(packageJson.name)
     false,
   )
   .option(
-    "--agents <agents>",
+    "--use-case <useCase>",
     `
 
-  Select which agents to use for the multi-agent template (e.g: financial_report, blog).
+  Select which use case to use for the multi-agent template (e.g: financial_report, blog).
 `,
   )
   .allowUnknownOption()
@@ -222,7 +215,7 @@ const options = program.opts();
 
 if (
   process.argv.includes("--no-llama-parse") ||
-  options.template === "extractor"
+  options.template === "reflex"
 ) {
   options.useLlamaParse = false;
 }
@@ -333,7 +326,6 @@ async function run(): Promise<void> {
     ...answers,
     appPath: resolvedProjectPath,
     packageManager,
-    externalPort: options.externalPort,
   });
 
   if (answers.postInstallAction === "VSCode") {
@@ -362,14 +354,7 @@ Please check ${cyan(
     }
   } else if (answers.postInstallAction === "runApp") {
     console.log(`Running app in ${root}...`);
-    await runApp(
-      root,
-      answers.template,
-      answers.frontend,
-      answers.framework,
-      options.port,
-      options.externalPort,
-    );
+    await runApp(root, answers.template, answers.framework, options.port);
   }
 }
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "create-llama",
-  "version": "0.3.13",
+  "version": "0.3.27",
   "description": "Create LlamaIndex-powered apps with one command",
   "keywords": [
     "rag",
@@ -43,7 +43,7 @@
     "@types/cross-spawn": "6.0.0",
     "@types/fs-extra": "11.0.4",
     "@types/node": "^20.11.7",
-    "@types/prompts": "2.0.1",
+    "@types/prompts": "2.4.2",
     "@types/tar": "6.1.5",
     "@types/validate-npm-package-name": "3.0.0",
     "async-retry": "1.3.1",

pnpm-lock.yaml

@@ -24,8 +24,8 @@ importers:
         specifier: ^20.11.7
         version: 20.12.10
       '@types/prompts':
-        specifier: 2.0.1
-        version: 2.0.1
+        specifier: 2.4.2
+        version: 2.4.2
       '@types/tar':
         specifier: 6.1.5
         version: 6.1.5
@@ -298,8 +298,8 @@ packages:
   '@types/normalize-package-data@2.4.4':
     resolution: {integrity: sha512-37i+OaWTh9qeK4LSHPsyRC7NahnGotNuZvjLSgcPzblpHB3rrCJxAOgI5gCdKm7coonsaX1Of0ILiTcnZjbfxA==}
 
-  '@types/prompts@2.0.1':
-    resolution: {integrity: sha512-AhtMcmETelF8wFDV1ucbChKhLgsc+ytXZXkNz/nnTAMSDeqsjALknEFxi7ZtLgS/G8bV2rp90LhDW5SGACimIQ==}
+  '@types/prompts@2.4.2':
+    resolution: {integrity: sha512-TwNx7qsjvRIUv/BCx583tqF5IINEVjCNqg9ofKHRlSoUHE62WBHrem4B1HGXcIrG511v29d1kJ9a/t2Esz7MIg==}
 
   '@types/responselike@1.0.3':
     resolution: {integrity: sha512-H/+L+UkTV33uf49PH5pCAUBVPNj2nDBXTN+qS1dOwyyg24l3CcicicCA7ca+HMvJBZcFgl5r8e+RR6elsb4Lyw==}
@@ -2208,7 +2208,10 @@ snapshots:
 
   '@types/normalize-package-data@2.4.4': {}
 
-  '@types/prompts@2.0.1': {}
+  '@types/prompts@2.4.2':
+    dependencies:
+      '@types/node': 20.12.10
+      kleur: 3.0.3
 
   '@types/responselike@1.0.3':
     dependencies:

questions/datasources.ts

@@ -49,7 +49,7 @@ export const getDataSourceChoices = (
     );
   }
 
-  if (framework === "fastapi" && template !== "extractor") {
+  if (framework === "fastapi" && template !== "reflex") {
     choices.push({
       title: "Use website content (requires Chrome)",
       value: "web",

questions/index.ts

@@ -4,10 +4,12 @@ import { askProQuestions } from "./questions";
 import { askSimpleQuestions } from "./simple";
 import { QuestionArgs, QuestionResults } from "./types";
 
+export const isCI = ciInfo.isCI || process.env.PLAYWRIGHT_TEST === "1";
+
 export const askQuestions = async (
   args: QuestionArgs,
 ): Promise<QuestionResults> => {
-  if (ciInfo.isCI || process.env.PLAYWRIGHT_TEST === "1") {
+  if (isCI) {
     return await getCIQuestionResults(args);
   } else if (args.pro) {
     // TODO: refactor pro questions to return a result object

questions/questions.ts

@@ -1,7 +1,8 @@
-import { blue, green } from "picocolors";
+import { blue } from "picocolors";
 import prompts from "prompts";
+import { isCI } from ".";
 import { COMMUNITY_OWNER, COMMUNITY_REPO } from "../helpers/constant";
-import { EXAMPLE_FILE } from "../helpers/datasources";
+import { EXAMPLE_FILE, EXAMPLE_GDPR } from "../helpers/datasources";
 import { getAvailableLlamapackOptions } from "../helpers/llama-pack";
 import { askModelConfig } from "../helpers/providers";
 import { getProjectOptions } from "../helpers/repo";
@@ -32,7 +33,7 @@ export const askProQuestions = async (program: QuestionArgs) => {
             title: "Multi-agent app (using workflows)",
             value: "multiagent",
           },
-          { title: "Structured Extractor", value: "extractor" },
+          { title: "Fullstack python template with Reflex", value: "reflex" },
           {
             title: `Community template from ${styledRepo}`,
             value: "community",
@@ -94,11 +95,29 @@ export const askProQuestions = async (program: QuestionArgs) => {
     return; // early return - no further questions needed for llamapack projects
   }
 
-  if (program.template === "extractor") {
-    // Extractor template only supports FastAPI, empty data sources, and llamacloud
+  if (program.template === "reflex") {
+    // Reflex 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];
     program.framework = "fastapi";
+    // Ask for which Reflex use case to use
+    const { useCase } = await prompts(
+      {
+        type: "select",
+        name: "useCase",
+        message: "Which use case would you like to build?",
+        choices: [
+          { title: "Structured Extractor", value: "extractor" },
+          {
+            title: "Contract review (using Workflow)",
+            value: "contract_review",
+          },
+        ],
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    program.useCase = useCase;
   }
 
   if (!program.framework) {
@@ -122,24 +141,17 @@ export const askProQuestions = async (program: QuestionArgs) => {
   }
 
   if (
-    (program.framework === "express" || program.framework === "fastapi") &&
+    program.framework === "fastapi" &&
     (program.template === "streaming" || program.template === "multiagent")
   ) {
     // if a backend-only framework is selected, ask whether we should create a frontend
     if (program.frontend === undefined) {
       const styledNextJS = blue("NextJS");
-      const styledBackend = green(
-        program.framework === "express"
-          ? "Express "
-          : program.framework === "fastapi"
-            ? "FastAPI (Python) "
-            : "",
-      );
       const { frontend } = await prompts({
         onState: onPromptState,
         type: "toggle",
         name: "frontend",
-        message: `Would you like to generate a ${styledNextJS} frontend for your ${styledBackend}backend?`,
+        message: `Would you like to generate a ${styledNextJS} frontend for your FastAPI backend?`,
         initial: false,
         active: "Yes",
         inactive: "No",
@@ -177,32 +189,50 @@ export const askProQuestions = async (program: QuestionArgs) => {
     program.observability = observability;
   }
 
-  // Ask agents
-  if (program.template === "multiagent" && !program.agents) {
-    const { agents } = await prompts(
+  if (
+    (program.template === "reflex" || program.template === "multiagent") &&
+    !program.useCase
+  ) {
+    const choices =
+      program.template === "reflex"
+        ? [
+            { title: "Structured Extractor", value: "extractor" },
+            {
+              title: "Contract review (using Workflow)",
+              value: "contract_review",
+            },
+          ]
+        : [
+            {
+              title: "Financial report (generate a financial report)",
+              value: "financial_report",
+            },
+            {
+              title: "Form filling (fill missing value in a CSV file)",
+              value: "form_filling",
+            },
+            { title: "Blog writer (Write a blog post)", value: "blog" },
+          ];
+
+    const { useCase } = await prompts(
       {
         type: "select",
-        name: "agents",
-        message: "Which agents would you like to use?",
-        choices: [
-          {
-            title: "Financial report (generate a financial report)",
-            value: "financial_report",
-          },
-          {
-            title: "Form filling (fill missing value in a CSV file)",
-            value: "form_filling",
-          },
-          {
-            title: "Blog writer (Write a blog post)",
-            value: "blog_writer",
-          },
-        ],
+        name: "useCase",
+        message: "Which use case would you like to use?",
+        choices,
         initial: 0,
       },
       questionHandlers,
     );
-    program.agents = agents;
+    program.useCase = useCase;
+  }
+
+  // Configure framework and data sources for Reflex template
+  if (program.template === "reflex") {
+    program.framework = "fastapi";
+
+    program.dataSources =
+      program.useCase === "extractor" ? [EXAMPLE_FILE] : [EXAMPLE_GDPR];
   }
 
   if (!program.modelConfig) {
@@ -228,8 +258,8 @@ export const askProQuestions = async (program: QuestionArgs) => {
     program.vectorDb = vectorDb;
   }
 
-  if (program.vectorDb === "llamacloud") {
-    // When using a LlamaCloud index, don't ask for data sources just copy an example file
+  if (program.vectorDb === "llamacloud" && program.dataSources.length === 0) {
+    // When using a LlamaCloud index and no data sources are provided, just copy an example file
     program.dataSources = [EXAMPLE_FILE];
   }
 
@@ -360,11 +390,8 @@ export const askProQuestions = async (program: QuestionArgs) => {
     // default to use LlamaParse if using LlamaCloud
     program.useLlamaParse = true;
   } else {
-    // Extractor template doesn't support LlamaParse and LlamaCloud right now (cannot use asyncio loop in Reflex)
-    if (
-      program.useLlamaParse === undefined &&
-      program.template !== "extractor"
-    ) {
+    // Reflex template doesn't support LlamaParse right now (cannot use asyncio loop in Reflex)
+    if (program.useLlamaParse === undefined && program.template !== "reflex") {
       // if already set useLlamaParse, don't ask again
       if (program.dataSources.some((ds) => ds.type === "file")) {
         const { useLlamaParse } = await prompts(
@@ -386,7 +413,7 @@ export const askProQuestions = async (program: QuestionArgs) => {
 
   // Ask for LlamaCloud API key when using a LlamaCloud index or LlamaParse
   if (isUsingLlamaCloud || program.useLlamaParse) {
-    if (!program.llamaCloudKey) {
+    if (!program.llamaCloudKey && !isCI) {
       // if already set, don't ask again
       // Ask for LlamaCloud API key
       const { llamaCloudKey } = await prompts(

questions/simple.ts

@@ -1,5 +1,10 @@
 import prompts from "prompts";
-import { EXAMPLE_10K_SEC_FILES, EXAMPLE_FILE } from "../helpers/datasources";
+import {
+  AI_REPORTS,
+  EXAMPLE_10K_SEC_FILES,
+  EXAMPLE_FILE,
+  EXAMPLE_GDPR,
+} from "../helpers/datasources";
 import { askModelConfig } from "../helpers/providers";
 import { getTools } from "../helpers/tools";
 import { ModelConfig, TemplateFramework } from "../helpers/types";
@@ -12,7 +17,9 @@ type AppType =
   | "financial_report_agent"
   | "form_filling"
   | "extractor"
-  | "data_scientist";
+  | "contract_review"
+  | "data_scientist"
+  | "deep_research";
 
 type SimpleAnswers = {
   appType: AppType;
@@ -29,19 +36,56 @@ export const askSimpleQuestions = async (
       type: "select",
       name: "appType",
       message: "What app do you want to build?",
+      hint: "🤖: Agent, 🔀: Workflow",
       choices: [
-        { title: "Agentic RAG", value: "rag" },
-        { title: "Data Scientist", value: "data_scientist" },
         {
-          title: "Financial Report Generator (using Workflows)",
+          title: "🤖 Agentic RAG",
+          value: "rag",
+          description:
+            "Chatbot that answers questions based on provided documents.",
+        },
+        {
+          title: "🤖 Data Scientist",
+          value: "data_scientist",
+          description:
+            "Agent that analyzes data and generates visualizations by using a code interpreter.",
+        },
+        {
+          title: "🤖 Code Artifact Agent",
+          value: "code_artifact",
+          description:
+            "Agent that writes code, runs it in a sandbox, and shows the output in the chat UI.",
+        },
+        {
+          title: "🤖 Information Extractor",
+          value: "extractor",
+          description:
+            "Extracts information from documents and returns it as a structured JSON object.",
+        },
+        {
+          title: "🔀 Financial Report Generator",
           value: "financial_report_agent",
+          description:
+            "Generates a financial report by analyzing the provided 10-K SEC data. Uses a code interpreter to create charts or to conduct further analysis.",
         },
         {
-          title: "Form Filler (using Workflows)",
+          title: "🔀 Financial 10k SEC Form Filler",
           value: "form_filling",
+          description:
+            "Extracts information from 10k SEC data and uses it to fill out a CSV form.",
+        },
+        {
+          title: "🔀 Contract Reviewer",
+          value: "contract_review",
+          description:
+            "Extracts and reviews contracts to ensure compliance with GDPR regulations",
+        },
+        {
+          title: "🔀 Deep Researcher",
+          value: "deep_research",
+          description:
+            "Researches and analyzes provided documents from multiple perspectives, generating a comprehensive report with citations to support key findings and insights.",
         },
-        { title: "Code Artifact Agent", value: "code_artifact" },
-        { title: "Information Extractor", value: "extractor" },
       ],
     },
     questionHandlers,
@@ -51,7 +95,11 @@ export const askSimpleQuestions = async (
   let llamaCloudKey = args.llamaCloudKey;
   let useLlamaCloud = false;
 
-  if (appType !== "extractor") {
+  if (
+    appType !== "extractor" &&
+    appType !== "contract_review" &&
+    appType !== "deep_research"
+  ) {
     const { language: newLanguage } = await prompts(
       {
         type: "select",
@@ -65,34 +113,34 @@ export const askSimpleQuestions = async (
       questionHandlers,
     );
     language = newLanguage;
+  }
 
-    const { useLlamaCloud: newUseLlamaCloud } = await prompts(
+  const { useLlamaCloud: newUseLlamaCloud } = await prompts(
+    {
+      type: "toggle",
+      name: "useLlamaCloud",
+      message: "Do you want to use LlamaCloud services?",
+      initial: false,
+      active: "Yes",
+      inactive: "No",
+      hint: "see https://www.llamaindex.ai/enterprise for more info",
+    },
+    questionHandlers,
+  );
+  useLlamaCloud = newUseLlamaCloud;
+
+  if (useLlamaCloud && !llamaCloudKey) {
+    // Ask for LlamaCloud API key, if not set
+    const { llamaCloudKey: newLlamaCloudKey } = await prompts(
       {
-        type: "toggle",
-        name: "useLlamaCloud",
-        message: "Do you want to use LlamaCloud services?",
-        initial: false,
-        active: "Yes",
-        inactive: "No",
-        hint: "see https://www.llamaindex.ai/enterprise for more info",
+        type: "text",
+        name: "llamaCloudKey",
+        message:
+          "Please provide your LlamaCloud API key (leave blank to skip):",
       },
       questionHandlers,
     );
-    useLlamaCloud = newUseLlamaCloud;
-
-    if (useLlamaCloud && !llamaCloudKey) {
-      // Ask for LlamaCloud API key, if not set
-      const { llamaCloudKey: newLlamaCloudKey } = await prompts(
-        {
-          type: "text",
-          name: "llamaCloudKey",
-          message:
-            "Please provide your LlamaCloud API key (leave blank to skip):",
-        },
-        questionHandlers,
-      );
-      llamaCloudKey = newLlamaCloudKey || process.env.LLAMA_CLOUD_API_KEY;
-    }
+    llamaCloudKey = newLlamaCloudKey || process.env.LLAMA_CLOUD_API_KEY;
   }
 
   const results = await convertAnswers(args, {
@@ -124,14 +172,14 @@ const convertAnswers = async (
     AppType,
     Pick<
       QuestionResults,
-      "template" | "tools" | "frontend" | "dataSources" | "agents"
+      "template" | "tools" | "frontend" | "dataSources" | "useCase"
     > & {
       modelConfig?: ModelConfig;
     }
   > = {
     rag: {
       template: "streaming",
-      tools: getTools(["duckduckgo"]),
+      tools: getTools(["weather"]),
       frontend: true,
       dataSources: [EXAMPLE_FILE],
     },
@@ -151,7 +199,7 @@ const convertAnswers = async (
     },
     financial_report_agent: {
       template: "multiagent",
-      agents: "financial_report",
+      useCase: "financial_report",
       tools: getTools(["document_generator", "interpreter"]),
       dataSources: EXAMPLE_10K_SEC_FILES,
       frontend: true,
@@ -159,18 +207,33 @@ const convertAnswers = async (
     },
     form_filling: {
       template: "multiagent",
-      agents: "form_filling",
+      useCase: "form_filling",
       tools: getTools(["form_filling"]),
       dataSources: EXAMPLE_10K_SEC_FILES,
       frontend: true,
       modelConfig: MODEL_GPT4o,
     },
     extractor: {
-      template: "extractor",
+      template: "reflex",
+      useCase: "extractor",
       tools: [],
       frontend: false,
       dataSources: [EXAMPLE_FILE],
     },
+    contract_review: {
+      template: "reflex",
+      useCase: "contract_review",
+      tools: [],
+      frontend: false,
+      dataSources: [EXAMPLE_GDPR],
+    },
+    deep_research: {
+      template: "multiagent",
+      useCase: "deep_research",
+      tools: [],
+      frontend: true,
+      dataSources: [AI_REPORTS],
+    },
   };
   const results = lookup[answers.appType];
   return {

questions/types.ts

@@ -2,7 +2,7 @@ import { InstallAppArgs } from "../create-app";
 
 export type QuestionResults = Omit<
   InstallAppArgs,
-  "appPath" | "packageManager" | "externalPort"
+  "appPath" | "packageManager"
 >;
 
 export type PureQuestionArgs = {

templates/.gitignore

@@ -1,3 +0,0 @@
-__pycache__
-poetry.lock
-storage

templates/README-fullstack.md

@@ -1,18 +0,0 @@
-This is a [LlamaIndex](https://www.llamaindex.ai/) project bootstrapped with [`create-llama`](https://github.com/run-llama/LlamaIndexTS/tree/main/packages/create-llama).
-
-## Getting Started
-
-First, startup the backend as described in the [backend README](./backend/README.md).
-
-Second, run the development server of the frontend as described in the [frontend README](./frontend/README.md).
-
-Open [http://localhost:3000](http://localhost:3000) with your browser to see the result.
-
-## Learn More
-
-To learn more about LlamaIndex, take a look at the following resources:
-
-- [LlamaIndex Documentation](https://docs.llamaindex.ai) - learn about LlamaIndex (Python features).
-- [LlamaIndexTS Documentation](https://ts.llamaindex.ai) - learn about LlamaIndex (Typescript features).
-
-You can check out [the LlamaIndexTS GitHub repository](https://github.com/run-llama/LlamaIndexTS) - your feedback and contributions are welcome!

templates/components/agents/python/blog/README-template.md

@@ -32,7 +32,7 @@ poetry run generate
 Third, run the development server:
 
 ```shell
-poetry run python main.py
+poetry run dev
 ```
 
 Per default, the example is using the explicit workflow. You can change the example by setting the `EXAMPLE_TYPE` environment variable to `choreography` or `orchestrator`.
@@ -47,14 +47,18 @@ curl --location 'localhost:8000/api/chat' \
 
 You can start editing the API by modifying `app/api/routers/chat.py` or `app/examples/workflow.py`. The API auto-updates as you save the files.
 
-Open [http://localhost:8000/docs](http://localhost:8000/docs) with your browser to see the Swagger UI of the API.
+Open [http://localhost:8000](http://localhost:8000) with your browser to start the app.
 
-The API allows CORS for all origins to simplify development. You can change this behavior by setting the `ENVIRONMENT` environment variable to `prod`:
+To start the app optimized for **production**, run:
 
 ```
-ENVIRONMENT=prod poetry run python main.py
+poetry run prod
 ```
 
+## Deployments
+
+For production deployments, check the [DEPLOY.md](DEPLOY.md) file.
+
 ## Learn More
 
 To learn more about LlamaIndex, take a look at the following resources:

templates/components/agents/python/blog/app/agents/researcher.py

@@ -1,4 +1,3 @@
-import os
 from textwrap import dedent
 from typing import List
 
@@ -6,42 +5,24 @@ from app.engine.index import IndexConfig, get_index
 from app.engine.tools import ToolFactory
 from app.workflows.single import FunctionCallingAgent
 from llama_index.core.chat_engine.types import ChatMessage
-from llama_index.core.tools import QueryEngineTool, ToolMetadata
+from app.engine.tools.query_engine import get_query_engine_tool
 
 
-def _create_query_engine_tool(params=None) -> QueryEngineTool:
-    """
-    Provide an agent worker that can be used to query the index.
-    """
-    # Add query tool if index exists
-    index_config = IndexConfig(**(params or {}))
-    index = get_index(index_config)
-    if index is None:
-        return None
-    top_k = int(os.getenv("TOP_K", 0))
-    query_engine = index.as_query_engine(
-        **({"similarity_top_k": top_k} if top_k != 0 else {})
-    )
-    return QueryEngineTool(
-        query_engine=query_engine,
-        metadata=ToolMetadata(
-            name="query_index",
-            description="""
-                Use this tool to retrieve information about the text corpus from the index.
-            """,
-        ),
-    )
-
-
-def _get_research_tools(**kwargs) -> QueryEngineTool:
+def _get_research_tools(**kwargs):
     """
     Researcher take responsibility for retrieving information.
     Try init wikipedia or duckduckgo tool if available.
     """
     tools = []
-    query_engine_tool = _create_query_engine_tool(**kwargs)
-    if query_engine_tool is not None:
-        tools.append(query_engine_tool)
+    # Create query engine tool
+    index_config = IndexConfig(**kwargs)
+    index = get_index(index_config)
+    if index is not None:
+        query_engine_tool = get_query_engine_tool(index=index)
+        if query_engine_tool is not None:
+            tools.append(query_engine_tool)
+
+    # Create duckduckgo tool
     researcher_tool_names = [
         "duckduckgo_search",
         "duckduckgo_image_search",

templates/components/agents/python/blog/app/workflows/planner.py

@@ -317,7 +317,7 @@ class Planner:
         # gather completed sub-tasks and response pairs
         completed_outputs_str = ""
         for sub_task_name, task_output in completed_sub_task.items():
-            task_str = f"{sub_task_name}:\n" f"\t{task_output!s}\n"
+            task_str = f"{sub_task_name}:\n\t{task_output!s}\n"
             completed_outputs_str += task_str
 
         # get a string for the remaining sub-tasks

templates/components/agents/python/deep_research/README-template.md

@@ -0,0 +1,47 @@
+This is a [LlamaIndex](https://www.llamaindex.ai/) multi-agents project using [Workflows](https://docs.llamaindex.ai/en/stable/understanding/workflows/).
+
+## Getting Started
+
+First, setup the environment with poetry:
+
+> **_Note:_** This step is not needed if you are using the dev-container.
+
+```shell
+poetry install
+```
+
+Then check the parameters that have been pre-configured in the `.env` file in this directory. (E.g. you might need to configure an `OPENAI_API_KEY` if you're using OpenAI as model provider).
+Second, generate the embeddings of the documents in the `./data` directory:
+
+```shell
+poetry run generate
+```
+
+Third, run the development server:
+
+```shell
+poetry run dev
+```
+
+## Use Case: Deep Research over own documents
+
+The workflow performs deep research by retrieving and analyzing documents from the [data](./data) directory from multiple perspectives. The project includes a sample PDF about AI investment in 2024 to help you get started. You can also add your own documents by placing them in the data directory and running the generate script again to index them.
+
+After starting the server, go to [http://localhost:8000](http://localhost:8000) and send a message to the agent to write a blog post.
+E.g: "AI investment in 2024"
+
+To update the workflow, you can edit the [deep_research.py](./app/workflows/deep_research.py) file.
+
+By default, the workflow retrieves 10 results from your documents. To customize the amount of information covered in the answer, you can adjust the `TOP_K` environment variable in the `.env` file. A higher value will retrieve more results from your documents, potentially providing more comprehensive answers.
+
+## Deployments
+
+For production deployments, check the [DEPLOY.md](DEPLOY.md) file.
+
+## Learn More
+
+To learn more about LlamaIndex, take a look at the following resources:
+
+- [LlamaIndex Documentation](https://docs.llamaindex.ai) - learn about LlamaIndex.
+- [Workflows Introduction](https://docs.llamaindex.ai/en/stable/understanding/workflows/) - learn about LlamaIndex workflows.
+  You can check out [the LlamaIndex GitHub repository](https://github.com/run-llama/llama_index) - your feedback and contributions are welcome!

templates/components/agents/python/deep_research/app/workflows/__init__.py

@@ -0,0 +1,3 @@
+from .deep_research import create_workflow
+
+__all__ = ["create_workflow"]

templates/components/agents/python/deep_research/app/workflows/agents.py

@@ -0,0 +1,158 @@
+from typing import List, Literal, Optional
+
+from llama_index.core.base.llms.types import (
+    CompletionResponse,
+    CompletionResponseAsyncGen,
+)
+from llama_index.core.memory.simple_composable_memory import SimpleComposableMemory
+from llama_index.core.prompts import PromptTemplate
+from llama_index.core.schema import MetadataMode, Node, NodeWithScore
+from llama_index.core.settings import Settings
+from pydantic import BaseModel, Field
+
+
+class AnalysisDecision(BaseModel):
+    decision: Literal["research", "write", "cancel"] = Field(
+        description="Whether to continue research, write a report, or cancel the research after several retries"
+    )
+    research_questions: Optional[List[str]] = Field(
+        description="Questions to research if continuing research. Maximum 3 questions. Set to null or empty if writing a report.",
+        default_factory=list,
+    )
+    cancel_reason: Optional[str] = Field(
+        description="The reason for cancellation if the decision is to cancel research.",
+        default=None,
+    )
+
+
+async def plan_research(
+    memory: SimpleComposableMemory,
+    context_nodes: List[Node],
+    user_request: str,
+) -> AnalysisDecision:
+    analyze_prompt = PromptTemplate(
+        """
+      You are a professor who is guiding a researcher to research a specific request/problem.
+      Your task is to decide on a research plan for the researcher.
+      The possible actions are:
+      + Provide a list of questions for the researcher to investigate, with the purpose of clarifying the request.
+      + Write a report if the researcher has already gathered enough research on the topic and can resolve the initial request.
+      + Cancel the research if most of the answers from researchers indicate there is insufficient information to research the request. Do not attempt more than 3 research iterations or too many questions.
+      The workflow should be:
+      + Always begin by providing some initial questions for the researcher to investigate.
+      + Analyze the provided answers against the initial topic/request. If the answers are insufficient to resolve the initial request, provide additional questions for the researcher to investigate.
+      + If the answers are sufficient to resolve the initial request, instruct the researcher to write a report.
+      <User request>
+      {user_request}
+      </User request>
+
+      <Collected information>
+      {context_str}
+      </Collected information>
+
+      <Conversation context>
+      {conversation_context}
+      </Conversation context>
+      """
+    )
+    conversation_context = "\n".join(
+        [f"{message.role}: {message.content}" for message in memory.get_all()]
+    )
+    context_str = "\n".join(
+        [node.get_content(metadata_mode=MetadataMode.LLM) for node in context_nodes]
+    )
+    res = await Settings.llm.astructured_predict(
+        output_cls=AnalysisDecision,
+        prompt=analyze_prompt,
+        user_request=user_request,
+        context_str=context_str,
+        conversation_context=conversation_context,
+    )
+    return res
+
+
+async def research(
+    question: str,
+    context_nodes: List[NodeWithScore],
+) -> str:
+    prompt = """
+    You are a researcher who is in the process of answering the question.
+    The purpose is to answer the question based on the collected information, without using prior knowledge or making up any new information.
+    Always add citations to the sentence/point/paragraph using the id of the provided content.
+    The citation should follow this format: [citation:id]() where id is the id of the content.
+    
+    E.g:
+    If we have a context like this:
+    <Citation id='abc-xyz'>
+    Baby llama is called cria
+    </Citation id='abc-xyz'>
+
+    And your answer uses the content, then the citation should be:
+    - Baby llama is called cria [citation:abc-xyz]()
+
+    Here is the provided context for the question:
+    <Collected information>
+    {context_str}
+    </Collected information>`
+
+    No prior knowledge, just use the provided context to answer the question: {question}
+    """
+    context_str = "\n".join(
+        [_get_text_node_content_for_citation(node) for node in context_nodes]
+    )
+    res = await Settings.llm.acomplete(
+        prompt=prompt.format(question=question, context_str=context_str),
+    )
+    return res.text
+
+
+async def write_report(
+    memory: SimpleComposableMemory,
+    user_request: str,
+    stream: bool = False,
+) -> CompletionResponse | CompletionResponseAsyncGen:
+    report_prompt = """
+    You are a researcher writing a report based on a user request and the research context.
+    You have researched various perspectives related to the user request.
+    The report should provide a comprehensive outline covering all important points from the researched perspectives.
+    Create a well-structured outline for the research report that covers all the answers.
+
+    # IMPORTANT when writing in markdown format:
+    + Use tables or figures where appropriate to enhance presentation.
+    + Preserve all citation syntax (the `[citation:id]()` parts in the provided context). Keep these citations in the final report - no separate reference section is needed.
+    + Do not add links, a table of contents, or a references section to the report.
+
+    <User request>
+    {user_request}
+    </User request>
+
+    <Research context>
+    {research_context}
+    </Research context>
+
+    Now, write a report addressing the user request based on the research provided following the format and guidelines above.
+    """
+    research_context = "\n".join(
+        [f"{message.role}: {message.content}" for message in memory.get_all()]
+    )
+
+    llm_complete_func = (
+        Settings.llm.astream_complete if stream else Settings.llm.acomplete
+    )
+
+    res = await llm_complete_func(
+        prompt=report_prompt.format(
+            user_request=user_request,
+            research_context=research_context,
+        ),
+    )
+    return res
+
+
+def _get_text_node_content_for_citation(node: NodeWithScore) -> str:
+    """
+    Construct node content for LLM with citation flag.
+    """
+    node_id = node.node.node_id
+    content = f"<Citation id='{node_id}'>\n{node.get_content(metadata_mode=MetadataMode.LLM)}</Citation id='{node_id}'>"
+    return content

templates/components/agents/python/deep_research/app/workflows/deep_research.py

@@ -0,0 +1,309 @@
+import logging
+import os
+import uuid
+from typing import Any, Dict, List, Optional
+
+from llama_index.core.indices.base import BaseIndex
+from llama_index.core.memory import ChatMemoryBuffer
+from llama_index.core.memory.simple_composable_memory import SimpleComposableMemory
+from llama_index.core.schema import Node
+from llama_index.core.types import ChatMessage, MessageRole
+from llama_index.core.workflow import (
+    Context,
+    StartEvent,
+    StopEvent,
+    Workflow,
+    step,
+)
+
+from app.engine.index import IndexConfig, get_index
+from app.workflows.agents import plan_research, research, write_report
+from app.workflows.events import SourceNodesEvent
+from app.workflows.models import (
+    CollectAnswersEvent,
+    DataEvent,
+    PlanResearchEvent,
+    ReportEvent,
+    ResearchEvent,
+)
+
+logger = logging.getLogger("uvicorn")
+logger.setLevel(logging.INFO)
+
+
+def create_workflow(
+    chat_history: Optional[List[ChatMessage]] = None,
+    params: Optional[Dict[str, Any]] = None,
+    **kwargs,
+) -> Workflow:
+    index_config = IndexConfig(**params)
+    index = get_index(index_config)
+    if index is None:
+        raise ValueError(
+            "Index is not found. Try run generation script to create the index first."
+        )
+
+    return DeepResearchWorkflow(
+        index=index,
+        chat_history=chat_history,
+        timeout=120.0,
+    )
+
+
+class DeepResearchWorkflow(Workflow):
+    """
+    A workflow to research and analyze documents from multiple perspectives and write a comprehensive report.
+
+    Requirements:
+    - An indexed documents containing the knowledge base related to the topic
+
+    Steps:
+    1. Retrieve information from the knowledge base
+    2. Analyze the retrieved information and provide questions for answering
+    3. Answer the questions
+    4. Write the report based on the research results
+    """
+
+    memory: SimpleComposableMemory
+    context_nodes: List[Node]
+    index: BaseIndex
+    user_request: str
+    stream: bool = True
+
+    def __init__(
+        self,
+        index: BaseIndex,
+        chat_history: Optional[List[ChatMessage]] = None,
+        stream: bool = True,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.index = index
+        self.context_nodes = []
+        self.stream = stream
+        self.chat_history = chat_history
+        self.memory = SimpleComposableMemory.from_defaults(
+            primary_memory=ChatMemoryBuffer.from_defaults(
+                chat_history=chat_history,
+            ),
+        )
+
+    @step
+    def retrieve(self, ctx: Context, ev: StartEvent) -> PlanResearchEvent:
+        """
+        Initiate the workflow: memory, tools, agent
+        """
+        self.user_request = ev.get("input")
+        self.memory.put_messages(
+            messages=[
+                ChatMessage(
+                    role=MessageRole.USER,
+                    content=self.user_request,
+                )
+            ]
+        )
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "retrieve",
+                    "state": "inprogress",
+                },
+            )
+        )
+        retriever = self.index.as_retriever(
+            similarity_top_k=int(os.getenv("TOP_K", 10)),
+        )
+        nodes = retriever.retrieve(self.user_request)
+        self.context_nodes.extend(nodes)
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "retrieve",
+                    "state": "done",
+                },
+            )
+        )
+        # Send source nodes to the stream
+        # Use SourceNodesEvent to display source nodes in the UI.
+        ctx.write_event_to_stream(
+            SourceNodesEvent(
+                nodes=nodes,
+            )
+        )
+        return PlanResearchEvent(
+            context_nodes=self.context_nodes,
+        )
+
+    @step
+    async def analyze(
+        self, ctx: Context, ev: PlanResearchEvent
+    ) -> ResearchEvent | ReportEvent | StopEvent:
+        """
+        Analyze the retrieved information
+        """
+        logger.info("Analyzing the retrieved information")
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "analyze",
+                    "state": "inprogress",
+                },
+            )
+        )
+        res = await plan_research(
+            memory=self.memory,
+            context_nodes=self.context_nodes,
+            user_request=self.user_request,
+        )
+        if res.decision == "cancel":
+            ctx.write_event_to_stream(
+                DataEvent(
+                    type="deep_research_event",
+                    data={
+                        "event": "analyze",
+                        "state": "done",
+                    },
+                )
+            )
+            return StopEvent(
+                result=res.cancel_reason,
+            )
+        elif res.decision == "write":
+            self.memory.put(
+                message=ChatMessage(
+                    role=MessageRole.ASSISTANT,
+                    content="No more idea to analyze. We should report the answers.",
+                )
+            )
+            ctx.send_event(ReportEvent())
+        else:
+            await ctx.set("n_questions", len(res.research_questions))
+            self.memory.put(
+                message=ChatMessage(
+                    role=MessageRole.ASSISTANT,
+                    content="We need to find answers to the following questions:\n"
+                    + "\n".join(res.research_questions),
+                )
+            )
+            for question in res.research_questions:
+                question_id = str(uuid.uuid4())
+                ctx.write_event_to_stream(
+                    DataEvent(
+                        type="deep_research_event",
+                        data={
+                            "event": "answer",
+                            "state": "pending",
+                            "id": question_id,
+                            "question": question,
+                            "answer": None,
+                        },
+                    )
+                )
+                ctx.send_event(
+                    ResearchEvent(
+                        question_id=question_id,
+                        question=question,
+                        context_nodes=self.context_nodes,
+                    )
+                )
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "analyze",
+                    "state": "done",
+                },
+            )
+        )
+        return None
+
+    @step(num_workers=2)
+    async def answer(self, ctx: Context, ev: ResearchEvent) -> CollectAnswersEvent:
+        """
+        Answer the question
+        """
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "answer",
+                    "state": "inprogress",
+                    "id": ev.question_id,
+                    "question": ev.question,
+                },
+            )
+        )
+        try:
+            answer = await research(
+                context_nodes=ev.context_nodes,
+                question=ev.question,
+            )
+        except Exception as e:
+            logger.error(f"Error answering question {ev.question}: {e}")
+            answer = f"Got error when answering the question: {ev.question}"
+        ctx.write_event_to_stream(
+            DataEvent(
+                type="deep_research_event",
+                data={
+                    "event": "answer",
+                    "state": "done",
+                    "id": ev.question_id,
+                    "question": ev.question,
+                    "answer": answer,
+                },
+            )
+        )
+
+        return CollectAnswersEvent(
+            question_id=ev.question_id,
+            question=ev.question,
+            answer=answer,
+        )
+
+    @step
+    async def collect_answers(
+        self, ctx: Context, ev: CollectAnswersEvent
+    ) -> PlanResearchEvent:
+        """
+        Collect answers to all questions
+        """
+        num_questions = await ctx.get("n_questions")
+        results = ctx.collect_events(
+            ev,
+            expected=[CollectAnswersEvent] * num_questions,
+        )
+        if results is None:
+            return None
+        for result in results:
+            self.memory.put(
+                message=ChatMessage(
+                    role=MessageRole.ASSISTANT,
+                    content=f"<Question>{result.question}</Question>\n<Answer>{result.answer}</Answer>",
+                )
+            )
+        await ctx.set("n_questions", 0)
+        self.memory.put(
+            message=ChatMessage(
+                role=MessageRole.ASSISTANT,
+                content="Researched all the questions. Now, i need to analyze if it's ready to write a report or need to research more.",
+            )
+        )
+        return PlanResearchEvent()
+
+    @step
+    async def report(self, ctx: Context, ev: ReportEvent) -> StopEvent:
+        """
+        Report the answers
+        """
+        logger.info("Writing the report")
+        res = await write_report(
+            memory=self.memory,
+            user_request=self.user_request,
+            stream=self.stream,
+        )
+        return StopEvent(
+            result=res,
+        )

templates/components/agents/python/deep_research/app/workflows/models.py

@@ -0,0 +1,43 @@
+from typing import List, Literal, Optional
+
+from llama_index.core.schema import NodeWithScore
+from llama_index.core.workflow import Event
+from pydantic import BaseModel
+
+
+# Workflow events
+class PlanResearchEvent(Event):
+    pass
+
+
+class ResearchEvent(Event):
+    question_id: str
+    question: str
+    context_nodes: List[NodeWithScore]
+
+
+class CollectAnswersEvent(Event):
+    question_id: str
+    question: str
+    answer: str
+
+
+class ReportEvent(Event):
+    pass
+
+
+# Events that are streamed to the frontend and rendered there
+class DeepResearchEventData(BaseModel):
+    event: Literal["retrieve", "analyze", "answer"]
+    state: Literal["pending", "inprogress", "done", "error"]
+    id: Optional[str] = None
+    question: Optional[str] = None
+    answer: Optional[str] = None
+
+
+class DataEvent(Event):
+    type: Literal["deep_research_event"]
+    data: DeepResearchEventData
+
+    def to_response(self):
+        return self.model_dump()

templates/components/agents/python/financial_report/README-template.md

@@ -21,7 +21,7 @@ poetry run generate
 Third, run the development server:
 
 ```shell
-poetry run python main.py
+poetry run dev
 ```
 
 The example provides one streaming API endpoint `/api/chat`.
@@ -35,14 +35,18 @@ curl --location 'localhost:8000/api/chat' \
 
 You can start editing the API by modifying `app/api/routers/chat.py` or `app/workflows/financial_report.py`. The API auto-updates as you save the files.
 
-Open [http://localhost:8000/docs](http://localhost:8000/docs) with your browser to see the Swagger UI of the API.
+Open [http://localhost:8000](http://localhost:8000) with your browser to start the app.
 
-The API allows CORS for all origins to simplify development. You can change this behavior by setting the `ENVIRONMENT` environment variable to `prod`:
+To start the app optimized for **production**, run:
 
 ```
-ENVIRONMENT=prod poetry run python main.py
+poetry run prod
 ```
 
+## Deployments
+
+For production deployments, check the [DEPLOY.md](DEPLOY.md) file.
+
 ## Learn More
 
 To learn more about LlamaIndex, take a look at the following resources:

templates/components/agents/python/financial_report/app/workflows/financial_report.py

@@ -1,8 +1,8 @@
-import os
 from typing import Any, Dict, List, Optional
 
 from app.engine.index import IndexConfig, get_index
 from app.engine.tools import ToolFactory
+from app.engine.tools.query_engine import get_query_engine_tool
 from app.workflows.events import AgentRunEvent
 from app.workflows.tools import (
     call_tools,
@@ -10,7 +10,6 @@ from app.workflows.tools import (
 )
 from llama_index.core import Settings
 from llama_index.core.base.llms.types import ChatMessage, MessageRole
-from llama_index.core.indices.vector_store import VectorStoreIndex
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.tools import FunctionTool, QueryEngineTool, ToolSelection
@@ -27,16 +26,16 @@ from llama_index.core.workflow import (
 def create_workflow(
     chat_history: Optional[List[ChatMessage]] = None,
     params: Optional[Dict[str, Any]] = None,
-    filters: Optional[List[Any]] = None,
+    **kwargs,
 ) -> Workflow:
+    # Create query engine tool
     index_config = IndexConfig(**params)
-    index: VectorStoreIndex = get_index(config=index_config)
+    index = get_index(index_config)
     if index is None:
-        query_engine_tool = None
-    else:
-        top_k = int(os.getenv("TOP_K", 10))
-        query_engine = index.as_query_engine(similarity_top_k=top_k, filters=filters)
-        query_engine_tool = QueryEngineTool.from_defaults(query_engine=query_engine)
+        raise ValueError(
+            "Index is not found. Try run generation script to create the index first."
+        )
+    query_engine_tool = get_query_engine_tool(index=index)
 
     configured_tools: Dict[str, FunctionTool] = ToolFactory.from_env(map_result=True)  # type: ignore
     code_interpreter_tool = configured_tools.get("interpret")
@@ -109,13 +108,13 @@ class FinancialReportWorkflow(Workflow):
         self.query_engine_tool = query_engine_tool
         self.code_interpreter_tool = code_interpreter_tool
         self.document_generator_tool = document_generator_tool
-        assert (
-            query_engine_tool is not None
-        ), "Query engine tool is not found. Try run generation script or upload a document file first."
+        assert query_engine_tool is not None, (
+            "Query engine tool is not found. Try run generation script or upload a document file first."
+        )
         assert code_interpreter_tool is not None, "Code interpreter tool is required"
-        assert (
-            document_generator_tool is not None
-        ), "Document generator tool is required"
+        assert document_generator_tool is not None, (
+            "Document generator tool is required"
+        )
         self.tools = [
             self.query_engine_tool,
             self.code_interpreter_tool,

templates/components/agents/python/form_filling/README-template.md

@@ -16,7 +16,7 @@ Make sure you have the `OPENAI_API_KEY` set.
 Second, run the development server:
 
 ```shell
-poetry run python main.py
+poetry run dev
 ```
 
 ## Use Case: Filling Financial CSV Template
@@ -41,14 +41,18 @@ curl --location 'localhost:8000/api/chat' \
 
 You can start editing the API by modifying `app/api/routers/chat.py` or `app/workflows/form_filling.py`. The API auto-updates as you save the files.
 
-Open [http://localhost:8000/docs](http://localhost:8000/docs) with your browser to see the Swagger UI of the API.
+Open [http://localhost:8000](http://localhost:8000) with your browser to start the app.
 
-The API allows CORS for all origins to simplify development. You can change this behavior by setting the `ENVIRONMENT` environment variable to `prod`:
+To start the app optimized for **production**, run:
 
 ```
-ENVIRONMENT=prod poetry run python main.py
+poetry run prod
 ```
 
+## Deployments
+
+For production deployments, check the [DEPLOY.md](DEPLOY.md) file.
+
 ## Learn More
 
 To learn more about LlamaIndex, take a look at the following resources:

templates/components/agents/python/form_filling/app/workflows/form_filling.py

@@ -1,16 +1,7 @@
-import os
 from typing import Any, Dict, List, Optional
 
-from app.engine.index import IndexConfig, get_index
-from app.engine.tools import ToolFactory
-from app.workflows.events import AgentRunEvent
-from app.workflows.tools import (
-    call_tools,
-    chat_with_tools,
-)
 from llama_index.core import Settings
 from llama_index.core.base.llms.types import ChatMessage, MessageRole
-from llama_index.core.indices.vector_store import VectorStoreIndex
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.tools import FunctionTool, QueryEngineTool, ToolSelection
@@ -23,24 +14,28 @@ from llama_index.core.workflow import (
     step,
 )
 
+from app.engine.index import IndexConfig, get_index
+from app.engine.tools import ToolFactory
+from app.engine.tools.query_engine import get_query_engine_tool
+from app.workflows.events import AgentRunEvent
+from app.workflows.tools import (
+    call_tools,
+    chat_with_tools,
+)
+
 
 def create_workflow(
     chat_history: Optional[List[ChatMessage]] = None,
     params: Optional[Dict[str, Any]] = None,
-    filters: Optional[List[Any]] = None,
+    **kwargs,
 ) -> Workflow:
-    if params is None:
-        params = {}
-    if filters is None:
-        filters = []
+    # Create query engine tool
     index_config = IndexConfig(**params)
-    index: VectorStoreIndex = get_index(config=index_config)
+    index = get_index(index_config)
     if index is None:
         query_engine_tool = None
     else:
-        top_k = int(os.getenv("TOP_K", 10))
-        query_engine = index.as_query_engine(similarity_top_k=top_k, filters=filters)
-        query_engine_tool = QueryEngineTool.from_defaults(query_engine=query_engine)
+        query_engine_tool = get_query_engine_tool(index=index)
 
     configured_tools = ToolFactory.from_env(map_result=True)
     extractor_tool = configured_tools.get("extract_questions")  # type: ignore

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

@@ -1,15 +1,15 @@
 import { ChatMessage } from "llamaindex";
 import { getTool } from "../engine/tools";
 import { FunctionCallingAgent } from "./single-agent";
-import { getQueryEngineTools } from "./tools";
+import { getQueryEngineTool } from "./tools";
 
 export const createResearcher = async (chatHistory: ChatMessage[]) => {
-  const queryEngineTools = await getQueryEngineTools();
+  const queryEngineTool = await getQueryEngineTool();
   const tools = [
     await getTool("wikipedia_tool"),
     await getTool("duckduckgo_search"),
     await getTool("image_generator"),
-    ...(queryEngineTools ? queryEngineTools : []),
+    queryEngineTool,
   ].filter((tool) => tool !== undefined);
 
   return new FunctionCallingAgent({

templates/components/agents/typescript/financial_report/workflow/factory.ts

@@ -1,7 +1,7 @@
 import { ChatMessage, ToolCallLLM } from "llamaindex";
 import { getTool } from "../engine/tools";
 import { FinancialReportWorkflow } from "./fin-report";
-import { getQueryEngineTools } from "./tools";
+import { getQueryEngineTool } from "./tools";
 
 const TIMEOUT = 360 * 1000;
 
@@ -9,11 +9,19 @@ export async function createWorkflow(options: {
   chatHistory: ChatMessage[];
   llm?: ToolCallLLM;
 }) {
+  const queryEngineTool = await getQueryEngineTool();
+  const codeInterpreterTool = await getTool("interpreter");
+  const documentGeneratorTool = await getTool("document_generator");
+
+  if (!queryEngineTool || !codeInterpreterTool || !documentGeneratorTool) {
+    throw new Error("One or more required tools are not defined");
+  }
+
   return new FinancialReportWorkflow({
     chatHistory: options.chatHistory,
-    queryEngineTools: (await getQueryEngineTools()) || [],
-    codeInterpreterTool: (await getTool("interpreter"))!,
-    documentGeneratorTool: (await getTool("document_generator"))!,
+    queryEngineTool,
+    codeInterpreterTool,
+    documentGeneratorTool,
     llm: options.llm,
     timeout: TIMEOUT,
   });

templates/components/agents/typescript/financial_report/workflow/fin-report.ts

@@ -45,7 +45,7 @@ export class FinancialReportWorkflow extends Workflow<
 > {
   llm: ToolCallLLM;
   memory: ChatMemoryBuffer;
-  queryEngineTools: BaseToolWithCall[];
+  queryEngineTool: BaseToolWithCall;
   codeInterpreterTool: BaseToolWithCall;
   documentGeneratorTool: BaseToolWithCall;
   systemPrompt?: string;
@@ -53,7 +53,7 @@ export class FinancialReportWorkflow extends Workflow<
   constructor(options: {
     llm?: ToolCallLLM;
     chatHistory: ChatMessage[];
-    queryEngineTools: BaseToolWithCall[];
+    queryEngineTool: BaseToolWithCall;
     codeInterpreterTool: BaseToolWithCall;
     documentGeneratorTool: BaseToolWithCall;
     systemPrompt?: string;
@@ -70,7 +70,7 @@ export class FinancialReportWorkflow extends Workflow<
       throw new Error("LLM is not a ToolCallLLM");
     }
     this.systemPrompt = options.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;
-    this.queryEngineTools = options.queryEngineTools;
+    this.queryEngineTool = options.queryEngineTool;
     this.codeInterpreterTool = options.codeInterpreterTool;
 
     this.documentGeneratorTool = options.documentGeneratorTool;
@@ -153,10 +153,11 @@ export class FinancialReportWorkflow extends Workflow<
   > => {
     const chatHistory = ev.data.input;
 
-    const tools = [this.codeInterpreterTool, this.documentGeneratorTool];
-    if (this.queryEngineTools) {
-      tools.push(...this.queryEngineTools);
-    }
+    const tools = [
+      this.codeInterpreterTool,
+      this.documentGeneratorTool,
+      this.queryEngineTool,
+    ];
 
     const toolCallResponse = await chatWithTools(this.llm, tools, chatHistory);
 
@@ -189,10 +190,7 @@ export class FinancialReportWorkflow extends Workflow<
           toolCalls: toolCallResponse.toolCalls,
         });
       default:
-        if (
-          this.queryEngineTools &&
-          this.queryEngineTools.some((tool) => tool.metadata.name === toolName)
-        ) {
+        if (this.queryEngineTool.metadata.name === toolName) {
           return new ResearchEvent({
             toolCalls: toolCallResponse.toolCalls,
           });
@@ -216,7 +214,7 @@ export class FinancialReportWorkflow extends Workflow<
     const { toolCalls } = ev.data;
 
     const toolMsgs = await callTools({
-      tools: this.queryEngineTools,
+      tools: [this.queryEngineTool],
       toolCalls,
       ctx,
       agentName: "Researcher",

templates/components/agents/typescript/form_filling/workflow/factory.ts

@@ -1,7 +1,7 @@
 import { ChatMessage, ToolCallLLM } from "llamaindex";
 import { getTool } from "../engine/tools";
 import { FormFillingWorkflow } from "./form-filling";
-import { getQueryEngineTools } from "./tools";
+import { getQueryEngineTool } from "./tools";
 
 const TIMEOUT = 360 * 1000;
 
@@ -9,11 +9,18 @@ export async function createWorkflow(options: {
   chatHistory: ChatMessage[];
   llm?: ToolCallLLM;
 }) {
+  const extractorTool = await getTool("extract_missing_cells");
+  const fillMissingCellsTool = await getTool("fill_missing_cells");
+
+  if (!extractorTool || !fillMissingCellsTool) {
+    throw new Error("One or more required tools are not defined");
+  }
+
   return new FormFillingWorkflow({
     chatHistory: options.chatHistory,
-    queryEngineTools: (await getQueryEngineTools()) || [],
-    extractorTool: (await getTool("extract_missing_cells"))!,
-    fillMissingCellsTool: (await getTool("fill_missing_cells"))!,
+    queryEngineTool: (await getQueryEngineTool()) || undefined,
+    extractorTool,
+    fillMissingCellsTool,
     llm: options.llm,
     timeout: TIMEOUT,
   });

templates/components/agents/typescript/form_filling/workflow/form-filling.ts

@@ -48,7 +48,7 @@ export class FormFillingWorkflow extends Workflow<
   llm: ToolCallLLM;
   memory: ChatMemoryBuffer;
   extractorTool: BaseToolWithCall;
-  queryEngineTools?: BaseToolWithCall[];
+  queryEngineTool?: BaseToolWithCall;
   fillMissingCellsTool: BaseToolWithCall;
   systemPrompt?: string;
 
@@ -56,7 +56,7 @@ export class FormFillingWorkflow extends Workflow<
     llm?: ToolCallLLM;
     chatHistory: ChatMessage[];
     extractorTool: BaseToolWithCall;
-    queryEngineTools?: BaseToolWithCall[];
+    queryEngineTool?: BaseToolWithCall;
     fillMissingCellsTool: BaseToolWithCall;
     systemPrompt?: string;
     verbose?: boolean;
@@ -73,7 +73,7 @@ export class FormFillingWorkflow extends Workflow<
     }
     this.systemPrompt = options.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;
     this.extractorTool = options.extractorTool;
-    this.queryEngineTools = options.queryEngineTools;
+    this.queryEngineTool = options.queryEngineTool;
     this.fillMissingCellsTool = options.fillMissingCellsTool;
 
     this.memory = new ChatMemoryBuffer({
@@ -156,8 +156,8 @@ export class FormFillingWorkflow extends Workflow<
     const chatHistory = ev.data.input;
 
     const tools = [this.extractorTool, this.fillMissingCellsTool];
-    if (this.queryEngineTools) {
-      tools.push(...this.queryEngineTools);
+    if (this.queryEngineTool) {
+      tools.push(this.queryEngineTool);
     }
 
     const toolCallResponse = await chatWithTools(this.llm, tools, chatHistory);
@@ -192,8 +192,8 @@ export class FormFillingWorkflow extends Workflow<
         });
       default:
         if (
-          this.queryEngineTools &&
-          this.queryEngineTools.some((tool) => tool.metadata.name === toolName)
+          this.queryEngineTool &&
+          this.queryEngineTool.metadata.name === toolName
         ) {
           return new FindAnswersEvent({
             toolCalls: toolCallResponse.toolCalls,
@@ -232,7 +232,7 @@ export class FormFillingWorkflow extends Workflow<
     ev: FindAnswersEvent,
   ): Promise<InputEvent> => {
     const { toolCalls } = ev.data;
-    if (!this.queryEngineTools) {
+    if (!this.queryEngineTool) {
       throw new Error("Query engine tool is not available");
     }
     ctx.sendEvent(
@@ -243,7 +243,7 @@ export class FormFillingWorkflow extends Workflow<
       }),
     );
     const toolMsgs = await callTools({
-      tools: this.queryEngineTools,
+      tools: [this.queryEngineTool],
       toolCalls,
       ctx,
       agentName: "Researcher",

templates/components/engines/python/agent/engine.py

@@ -1,18 +1,18 @@
 import os
 from typing import List
 
-from app.engine.index import IndexConfig, get_index
-from app.engine.tools import ToolFactory
 from llama_index.core.agent import AgentRunner
 from llama_index.core.callbacks import CallbackManager
 from llama_index.core.settings import Settings
 from llama_index.core.tools import BaseTool
-from llama_index.core.tools.query_engine import QueryEngineTool
+
+from app.engine.index import IndexConfig, get_index
+from app.engine.tools import ToolFactory
+from app.engine.tools.query_engine import get_query_engine_tool
 
 
-def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
+def get_chat_engine(params=None, event_handlers=None, **kwargs):
     system_prompt = os.getenv("SYSTEM_PROMPT")
-    top_k = int(os.getenv("TOP_K", 0))
     tools: List[BaseTool] = []
     callback_manager = CallbackManager(handlers=event_handlers or [])
 
@@ -20,10 +20,7 @@ def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
     index_config = IndexConfig(callback_manager=callback_manager, **(params or {}))
     index = get_index(index_config)
     if index is not None:
-        query_engine = index.as_query_engine(
-            filters=filters, **({"similarity_top_k": top_k} if top_k != 0 else {})
-        )
-        query_engine_tool = QueryEngineTool.from_defaults(query_engine=query_engine)
+        query_engine_tool = get_query_engine_tool(index, **kwargs)
         tools.append(query_engine_tool)
 
     # Add additional tools

templates/components/engines/python/agent/tools/interpreter.py

@@ -5,7 +5,7 @@ import uuid
 from typing import List, Optional
 
 from app.services.file import DocumentFile, FileService
-from e2b_code_interpreter import CodeInterpreter
+from e2b_code_interpreter import Sandbox
 from e2b_code_interpreter.models import Logs
 from llama_index.core.tools import FunctionTool
 from pydantic import BaseModel
@@ -61,7 +61,7 @@ class E2BCodeInterpreter:
         Lazily initialize the interpreter.
         """
         logger.info(f"Initializing interpreter with {len(sandbox_files)} files")
-        self.interpreter = CodeInterpreter(api_key=self.api_key)
+        self.interpreter = Sandbox(api_key=self.api_key)
         if len(sandbox_files) > 0:
             for file_path in sandbox_files:
                 file_name = os.path.basename(file_path)
@@ -159,11 +159,11 @@ class E2BCodeInterpreter:
         if self.interpreter is None:
             self._init_interpreter(sandbox_files)
 
-        if self.interpreter and self.interpreter.notebook:
+        if self.interpreter:
             logger.info(
-                f"\n{'='*50}\n> Running following AI-generated code:\n{code}\n{'='*50}"
+                f"\n{'=' * 50}\n> Running following AI-generated code:\n{code}\n{'=' * 50}"
             )
-            exec = self.interpreter.notebook.exec_cell(code)
+            exec = self.interpreter.run_code(code)
 
             if exec.error:
                 error_message = f"The code failed to execute successfully. Error: {exec.error}. Try to fix the code and run again."

templates/components/engines/python/agent/tools/query_engine.py

@@ -0,0 +1,187 @@
+import os
+from typing import Any, Dict, List, Optional, Sequence
+
+from llama_index.core import get_response_synthesizer
+from llama_index.core.base.base_query_engine import BaseQueryEngine
+from llama_index.core.base.response.schema import RESPONSE_TYPE, Response
+from llama_index.core.multi_modal_llms import MultiModalLLM
+from llama_index.core.prompts.base import BasePromptTemplate
+from llama_index.core.prompts.default_prompt_selectors import (
+    DEFAULT_TEXT_QA_PROMPT_SEL,
+)
+from llama_index.core.query_engine.multi_modal import _get_image_and_text_nodes
+from llama_index.core.response_synthesizers.base import BaseSynthesizer, QueryTextType
+from llama_index.core.schema import (
+    ImageNode,
+    NodeWithScore,
+)
+from llama_index.core.tools.query_engine import QueryEngineTool
+from llama_index.core.types import RESPONSE_TEXT_TYPE
+
+from app.settings import get_multi_modal_llm
+
+
+def create_query_engine(index, **kwargs) -> BaseQueryEngine:
+    """
+    Create a query engine for the given index.
+
+    Args:
+        index: The index to create a query engine for.
+        params (optional): Additional parameters for the query engine, e.g: similarity_top_k
+    """
+
+    top_k = int(os.getenv("TOP_K", 0))
+    if top_k != 0 and kwargs.get("filters") is None:
+        kwargs["similarity_top_k"] = top_k
+    multimodal_llm = get_multi_modal_llm()
+    if multimodal_llm:
+        kwargs["response_synthesizer"] = MultiModalSynthesizer(
+            multimodal_model=multimodal_llm,
+        )
+
+    # If index is index is LlamaCloudIndex
+    # use auto_routed mode for better query results
+    if index.__class__.__name__ == "LlamaCloudIndex":
+        if kwargs.get("retrieval_mode") is None:
+            kwargs["retrieval_mode"] = "auto_routed"
+        if multimodal_llm:
+            kwargs["retrieve_image_nodes"] = True
+    return index.as_query_engine(**kwargs)
+
+
+def get_query_engine_tool(
+    index,
+    name: Optional[str] = None,
+    description: Optional[str] = None,
+    **kwargs,
+) -> QueryEngineTool:
+    """
+    Get a query engine tool for the given index.
+
+    Args:
+        index: The index to create a query engine for.
+        name (optional): The name of the tool.
+        description (optional): The description of the tool.
+    """
+    if name is None:
+        name = "query_index"
+    if description is None:
+        description = (
+            "Use this tool to retrieve information about the text corpus from an index."
+        )
+    query_engine = create_query_engine(index, **kwargs)
+    return QueryEngineTool.from_defaults(
+        query_engine=query_engine,
+        name=name,
+        description=description,
+    )
+
+
+class MultiModalSynthesizer(BaseSynthesizer):
+    """
+    A synthesizer that summarizes text nodes and uses a multi-modal LLM to generate a response.
+    """
+
+    def __init__(
+        self,
+        multimodal_model: MultiModalLLM,
+        response_synthesizer: Optional[BaseSynthesizer] = None,
+        text_qa_template: Optional[BasePromptTemplate] = None,
+        *args,
+        **kwargs,
+    ):
+        super().__init__(*args, **kwargs)
+        self._multi_modal_llm = multimodal_model
+        self._response_synthesizer = response_synthesizer or get_response_synthesizer()
+        self._text_qa_template = text_qa_template or DEFAULT_TEXT_QA_PROMPT_SEL
+
+    def _get_prompts(self, **kwargs) -> Dict[str, Any]:
+        return {
+            "text_qa_template": self._text_qa_template,
+        }
+
+    def _update_prompts(self, prompts: Dict[str, Any]) -> None:
+        if "text_qa_template" in prompts:
+            self._text_qa_template = prompts["text_qa_template"]
+
+    async def aget_response(
+        self,
+        *args,
+        **response_kwargs: Any,
+    ) -> RESPONSE_TEXT_TYPE:
+        return await self._response_synthesizer.aget_response(*args, **response_kwargs)
+
+    def get_response(self, *args, **kwargs) -> RESPONSE_TEXT_TYPE:
+        return self._response_synthesizer.get_response(*args, **kwargs)
+
+    async def asynthesize(
+        self,
+        query: QueryTextType,
+        nodes: List[NodeWithScore],
+        additional_source_nodes: Optional[Sequence[NodeWithScore]] = None,
+        **response_kwargs: Any,
+    ) -> RESPONSE_TYPE:
+        image_nodes, text_nodes = _get_image_and_text_nodes(nodes)
+
+        if len(image_nodes) == 0:
+            return await self._response_synthesizer.asynthesize(query, text_nodes)
+
+        # Summarize the text nodes to avoid exceeding the token limit
+        text_response = str(
+            await self._response_synthesizer.asynthesize(query, text_nodes)
+        )
+
+        fmt_prompt = self._text_qa_template.format(
+            context_str=text_response,
+            query_str=query.query_str,  # type: ignore
+        )
+
+        llm_response = await self._multi_modal_llm.acomplete(
+            prompt=fmt_prompt,
+            image_documents=[
+                image_node.node
+                for image_node in image_nodes
+                if isinstance(image_node.node, ImageNode)
+            ],
+        )
+
+        return Response(
+            response=str(llm_response),
+            source_nodes=nodes,
+            metadata={"text_nodes": text_nodes, "image_nodes": image_nodes},
+        )
+
+    def synthesize(
+        self,
+        query: QueryTextType,
+        nodes: List[NodeWithScore],
+        additional_source_nodes: Optional[Sequence[NodeWithScore]] = None,
+        **response_kwargs: Any,
+    ) -> RESPONSE_TYPE:
+        image_nodes, text_nodes = _get_image_and_text_nodes(nodes)
+
+        if len(image_nodes) == 0:
+            return self._response_synthesizer.synthesize(query, text_nodes)
+
+        # Summarize the text nodes to avoid exceeding the token limit
+        text_response = str(self._response_synthesizer.synthesize(query, text_nodes))
+
+        fmt_prompt = self._text_qa_template.format(
+            context_str=text_response,
+            query_str=query.query_str,  # type: ignore
+        )
+
+        llm_response = self._multi_modal_llm.complete(
+            prompt=fmt_prompt,
+            image_documents=[
+                image_node.node
+                for image_node in image_nodes
+                if isinstance(image_node.node, ImageNode)
+            ],
+        )
+
+        return Response(
+            response=str(llm_response),
+            source_nodes=nodes,
+            metadata={"text_nodes": text_nodes, "image_nodes": image_nodes},
+        )

templates/components/engines/python/chat/engine.py

@@ -9,7 +9,7 @@ from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.settings import Settings
 
 
-def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
+def get_chat_engine(params=None, event_handlers=None, **kwargs):
     system_prompt = os.getenv("SYSTEM_PROMPT")
     citation_prompt = os.getenv("SYSTEM_CITATION_PROMPT", None)
     top_k = int(os.getenv("TOP_K", 0))
@@ -33,10 +33,9 @@ def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
                 "StorageContext is empty - call 'poetry run generate' to generate the storage first"
             ),
         )
-
-    retriever = index.as_retriever(
-        filters=filters, **({"similarity_top_k": top_k} if top_k != 0 else {})
-    )
+    if top_k != 0 and kwargs.get("similarity_top_k") is None:
+        kwargs["similarity_top_k"] = top_k
+    retriever = index.as_retriever(**kwargs)
 
     return CondensePlusContextChatEngine(
         llm=llm,

templates/components/engines/typescript/agent/chat.ts

@@ -1,14 +1,9 @@
-import {
-  BaseChatEngine,
-  BaseToolWithCall,
-  LLMAgent,
-  QueryEngineTool,
-} from "llamaindex";
+import { BaseChatEngine, BaseToolWithCall, LLMAgent } from "llamaindex";
 import fs from "node:fs/promises";
 import path from "node:path";
 import { getDataSource } from "./index";
-import { generateFilters } from "./queryFilter";
 import { createTools } from "./tools";
+import { createQueryEngineTool } from "./tools/query-engine";
 
 export async function createChatEngine(documentIds?: string[], params?: any) {
   const tools: BaseToolWithCall[] = [];
@@ -17,17 +12,7 @@ export async function createChatEngine(documentIds?: string[], params?: any) {
   // Delete this code if you don't have a data source
   const index = await getDataSource(params);
   if (index) {
-    tools.push(
-      new QueryEngineTool({
-        queryEngine: index.asQueryEngine({
-          preFilters: generateFilters(documentIds || []),
-        }),
-        metadata: {
-          name: "data_query_engine",
-          description: `A query engine for documents from your data source.`,
-        },
-      }),
-    );
+    tools.push(createQueryEngineTool(index, { documentIds }));
   }
 
   const configFile = path.join("config", "tools.json");

templates/components/engines/typescript/agent/tools/code-generator.ts

@@ -103,6 +103,7 @@ export class CodeGeneratorTool implements BaseTool<CodeGeneratorParameter> {
       const artifact = await this.generateArtifact(
         input.requirement,
         input.oldCode,
+        input.sandboxFiles, // help the generated code use exact files
       );
       if (input.sandboxFiles) {
         artifact.files = input.sandboxFiles;
@@ -117,10 +118,12 @@ export class CodeGeneratorTool implements BaseTool<CodeGeneratorParameter> {
   async generateArtifact(
     query: string,
     oldCode?: string,
+    attachments?: string[],
   ): Promise<CodeArtifact> {
     const userMessage = `
     ${query}
     ${oldCode ? `The existing code is: \n\`\`\`${oldCode}\`\`\`` : ""}
+    ${attachments ? `The attachments are: \n${attachments.join("\n")}` : ""}
     `;
     const messages: ChatMessage[] = [
       { role: "system", content: CODE_GENERATION_PROMPT },

templates/components/engines/typescript/agent/tools/interpreter.ts

@@ -1,4 +1,4 @@
-import { CodeInterpreter, Logs, Result } from "@e2b/code-interpreter";
+import { Logs, Result, Sandbox } from "@e2b/code-interpreter";
 import type { JSONSchemaType } from "ajv";
 import fs from "fs";
 import { BaseTool, ToolMetadata } from "llamaindex";
@@ -82,7 +82,7 @@ export class InterpreterTool implements BaseTool<InterpreterParameter> {
   private apiKey?: string;
   private fileServerURLPrefix?: string;
   metadata: ToolMetadata<JSONSchemaType<InterpreterParameter>>;
-  codeInterpreter?: CodeInterpreter;
+  codeInterpreter?: Sandbox;
 
   constructor(params?: InterpreterToolParams) {
     this.metadata = params?.metadata || DEFAULT_META_DATA;
@@ -104,24 +104,27 @@ export class InterpreterTool implements BaseTool<InterpreterParameter> {
 
   public async initInterpreter(input: InterpreterParameter) {
     if (!this.codeInterpreter) {
-      this.codeInterpreter = await CodeInterpreter.create({
+      this.codeInterpreter = await Sandbox.create({
         apiKey: this.apiKey,
       });
-    }
-    // upload files to sandbox
-    if (input.sandboxFiles) {
-      console.log(`Uploading ${input.sandboxFiles.length} files to sandbox`);
-      try {
-        for (const filePath of input.sandboxFiles) {
-          const fileName = path.basename(filePath);
-          const localFilePath = path.join(this.uploadedFilesDir, fileName);
-          const content = fs.readFileSync(localFilePath);
-          await this.codeInterpreter?.files.write(filePath, content);
+      // upload files to sandbox when it's initialized
+      if (input.sandboxFiles) {
+        console.log(`Uploading ${input.sandboxFiles.length} files to sandbox`);
+        try {
+          for (const filePath of input.sandboxFiles) {
+            const fileName = path.basename(filePath);
+            const localFilePath = path.join(this.uploadedFilesDir, fileName);
+            const content = fs.readFileSync(localFilePath);
+
+            const arrayBuffer = new Uint8Array(content).buffer;
+            await this.codeInterpreter?.files.write(filePath, arrayBuffer);
+          }
+        } catch (error) {
+          console.error("Got error when uploading files to sandbox", error);
         }
-      } catch (error) {
-        console.error("Got error when uploading files to sandbox", error);
       }
     }
+
     return this.codeInterpreter;
   }
 
@@ -148,7 +151,7 @@ export class InterpreterTool implements BaseTool<InterpreterParameter> {
       `\n${"=".repeat(50)}\n> Running following AI-generated code:\n${input.code}\n${"=".repeat(50)}`,
     );
     const interpreter = await this.initInterpreter(input);
-    const exec = await interpreter.notebook.execCell(input.code);
+    const exec = await interpreter.runCode(input.code);
     if (exec.error) console.error("[Code Interpreter error]", exec.error);
     const extraResult = await this.getExtraResult(exec.results[0]);
     const result: InterpreterToolOutput = {
@@ -167,7 +170,7 @@ export class InterpreterTool implements BaseTool<InterpreterParameter> {
   }
 
   async close() {
-    await this.codeInterpreter?.close();
+    await this.codeInterpreter?.kill();
   }
 
   private async getExtraResult(

templates/components/engines/typescript/agent/tools/query-engine.ts

@@ -0,0 +1,57 @@
+import {
+  BaseQueryEngine,
+  CloudRetrieveParams,
+  LlamaCloudIndex,
+  MetadataFilters,
+  QueryEngineTool,
+  VectorStoreIndex,
+} from "llamaindex";
+import { generateFilters } from "../queryFilter";
+
+interface QueryEngineParams {
+  documentIds?: string[];
+  topK?: number;
+}
+
+export function createQueryEngineTool(
+  index: VectorStoreIndex | LlamaCloudIndex,
+  params?: QueryEngineParams,
+  name?: string,
+  description?: string,
+): QueryEngineTool {
+  return new QueryEngineTool({
+    queryEngine: createQueryEngine(index, params),
+    metadata: {
+      name: name || "query_engine",
+      description:
+        description ||
+        `Use this tool to retrieve information about the text corpus from an index.`,
+    },
+  });
+}
+
+function createQueryEngine(
+  index: VectorStoreIndex | LlamaCloudIndex,
+  params?: QueryEngineParams,
+): BaseQueryEngine {
+  const baseQueryParams = {
+    similarityTopK:
+      params?.topK ??
+      (process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined),
+  };
+
+  if (index instanceof LlamaCloudIndex) {
+    return index.asQueryEngine({
+      ...baseQueryParams,
+      retrieval_mode: "auto_routed",
+      preFilters: generateFilters(
+        params?.documentIds || [],
+      ) as CloudRetrieveParams["filters"],
+    });
+  }
+
+  return index.asQueryEngine({
+    ...baseQueryParams,
+    preFilters: generateFilters(params?.documentIds || []) as MetadataFilters,
+  });
+}

templates/components/multiagent/python/app/api/routers/chat.py

@@ -1,12 +1,13 @@
 import logging
 
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Request, status
+
 from app.api.routers.models import (
     ChatData,
 )
 from app.api.routers.vercel_response import VercelStreamResponse
 from app.engine.query_filter import generate_filters
 from app.workflows import create_workflow
-from fastapi import APIRouter, BackgroundTasks, HTTPException, Request, status
 
 chat_router = r = APIRouter()
 
@@ -28,13 +29,16 @@ async def chat(
         params = data.data or {}
 
         workflow = create_workflow(
-            chat_history=messages, params=params, filters=filters
+            chat_history=messages,
+            params=params,
+            filters=filters,
         )
 
         event_handler = workflow.run(input=last_message_content, streaming=True)
         return VercelStreamResponse(
             request=request,
             chat_data=data,
+            background_tasks=background_tasks,
             event_handler=event_handler,
             events=workflow.stream_events(),
         )

templates/components/multiagent/python/app/api/routers/vercel_response.py

@@ -4,10 +4,12 @@ import logging
 from typing import AsyncGenerator, Awaitable, List
 
 from aiostream import stream
+from fastapi import BackgroundTasks, Request
+from fastapi.responses import StreamingResponse
+from llama_index.core.schema import NodeWithScore
+
 from app.api.routers.models import ChatData, Message
 from app.api.services.suggestion import NextQuestionSuggestion
-from fastapi import Request
-from fastapi.responses import StreamingResponse
 
 logger = logging.getLogger("uvicorn")
 
@@ -19,10 +21,19 @@ class VercelStreamResponse(StreamingResponse):
 
     TEXT_PREFIX = "0:"
     DATA_PREFIX = "8:"
+    ERROR_PREFIX = "3:"
 
-    def __init__(self, request: Request, chat_data: ChatData, *args, **kwargs):
+    def __init__(
+        self,
+        request: Request,
+        chat_data: ChatData,
+        background_tasks: BackgroundTasks,
+        *args,
+        **kwargs,
+    ):
         self.request = request
         self.chat_data = chat_data
+        self.background_tasks = background_tasks
         content = self.content_generator(*args, **kwargs)
         super().__init__(content=content)
 
@@ -41,13 +52,16 @@ class VercelStreamResponse(StreamingResponse):
 
                     yield output
         except asyncio.CancelledError:
-            logger.info("Stopping workflow")
-            await event_handler.cancel_run()
+            logger.warning("Workflow has been cancelled!")
         except Exception as e:
             logger.error(
                 f"Unexpected error in content_generator: {str(e)}", exc_info=True
             )
+            yield self.convert_error(
+                "An unexpected error occurred while processing your request, preventing the creation of a final answer. Please try again."
+            )
         finally:
+            await event_handler.cancel_run()
             logger.info("The stream has been stopped!")
 
     def _create_stream(
@@ -74,6 +88,9 @@ class VercelStreamResponse(StreamingResponse):
                         for token in content:
                             final_response += str(token)
                             yield self.convert_text(token)
+                else:
+                    final_response += str(result)
+                    yield self.convert_text(result)
 
             # Generate next questions if next question prompt is configured
             question_data = await self._generate_next_questions(
@@ -82,8 +99,6 @@ class VercelStreamResponse(StreamingResponse):
             if question_data:
                 yield self.convert_data(question_data)
 
-            # TODO: stream sources
-
         # Yield the events from the event handler
         async def _event_generator():
             async for event in events:
@@ -92,10 +107,30 @@ class VercelStreamResponse(StreamingResponse):
                     logger.debug(event_response)
                 if event_response is not None:
                     yield self.convert_data(event_response)
+                if event_response.get("type") == "sources":
+                    self._process_response_nodes(event.nodes, self.background_tasks)
 
         combine = stream.merge(_chat_response_generator(), _event_generator())
         return combine
 
+    @staticmethod
+    def _process_response_nodes(
+        source_nodes: List[NodeWithScore],
+        background_tasks: BackgroundTasks,
+    ):
+        try:
+            # Start background tasks to download documents from LlamaCloud if needed
+            from app.engine.service import LLamaCloudFileService  # type: ignore
+
+            LLamaCloudFileService.download_files_from_nodes(
+                source_nodes, background_tasks
+            )
+        except ImportError:
+            logger.debug(
+                "LlamaCloud is not configured. Skipping post processing of nodes"
+            )
+            pass
+
     @classmethod
     def convert_text(cls, token: str):
         # Escape newlines and double quotes to avoid breaking the stream
@@ -107,6 +142,11 @@ class VercelStreamResponse(StreamingResponse):
         data_str = json.dumps(data)
         return f"{cls.DATA_PREFIX}[{data_str}]\n"
 
+    @classmethod
+    def convert_error(cls, error: str):
+        error_str = json.dumps(error)
+        return f"{cls.ERROR_PREFIX}{error_str}\n"
+
     @staticmethod
     async def _generate_next_questions(chat_history: List[Message], response: str):
         questions = await NextQuestionSuggestion.suggest_next_questions(

templates/components/multiagent/python/app/workflows/events.py

@@ -1,8 +1,11 @@
 from enum import Enum
-from typing import Optional
+from typing import List, Optional
 
+from llama_index.core.schema import NodeWithScore
 from llama_index.core.workflow import Event
 
+from app.api.routers.models import SourceNodes
+
 
 class AgentRunEventType(Enum):
     TEXT = "text"
@@ -25,3 +28,18 @@ class AgentRunEvent(Event):
                 "data": self.data,
             },
         }
+
+
+class SourceNodesEvent(Event):
+    nodes: List[NodeWithScore]
+
+    def to_response(self):
+        return {
+            "type": "sources",
+            "data": {
+                "nodes": [
+                    SourceNodes.from_source_node(node).model_dump()
+                    for node in self.nodes
+                ]
+            },
+        }

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

@@ -1,4 +1,4 @@
-import { Message, streamToResponse } from "ai";
+import { LlamaIndexAdapter, Message } from "ai";
 import { Request, Response } from "express";
 import {
   convertToChatHistory,
@@ -28,7 +28,20 @@ export const chat = async (req: Request, res: Response) => {
     const { stream, dataStream } =
       await createStreamFromWorkflowContext(context);
 
-    return streamToResponse(stream, res, {}, dataStream);
+    const streamResponse = LlamaIndexAdapter.toDataStreamResponse(stream, {
+      data: dataStream,
+    });
+    if (streamResponse.body) {
+      const reader = streamResponse.body.getReader();
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) {
+          res.end();
+          return;
+        }
+        res.write(value);
+      }
+    }
   } catch (error) {
     console.error("[LlamaIndex]", error);
     return res.status(500).json({

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

@@ -1,5 +1,5 @@
 import { initObservability } from "@/app/observability";
-import { StreamingTextResponse, type Message } from "ai";
+import { LlamaIndexAdapter, type Message } from "ai";
 import { NextRequest, NextResponse } from "next/server";
 import { initSettings } from "./engine/settings";
 import {
@@ -41,9 +41,9 @@ export async function POST(request: NextRequest) {
     });
     const { stream, dataStream } =
       await createStreamFromWorkflowContext(context);
-
-    // Return the two streams in one response
-    return new StreamingTextResponse(stream, {}, dataStream);
+    return LlamaIndexAdapter.toDataStreamResponse(stream, {
+      data: dataStream,
+    });
   } catch (error) {
     console.error("[LlamaIndex]", error);
     return NextResponse.json(

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

@@ -3,20 +3,15 @@ import {
   WorkflowContext,
   WorkflowEvent,
 } from "@llamaindex/workflow";
-import {
-  StreamData,
-  createStreamDataTransformer,
-  trimStartOfStreamHelper,
-} from "ai";
-import { ChatResponseChunk } from "llamaindex";
+import { StreamData } from "ai";
+import { ChatResponseChunk, EngineResponse } from "llamaindex";
+import { ReadableStream } from "stream/web";
 import { AgentRunEvent } from "./type";
 
 export async function createStreamFromWorkflowContext<Input, Output, Context>(
   context: WorkflowContext<Input, Output, Context>,
-): Promise<{ stream: ReadableStream<string>; dataStream: StreamData }> {
-  const trimStartOfStream = trimStartOfStreamHelper();
+): Promise<{ stream: ReadableStream<EngineResponse>; dataStream: StreamData }> {
   const dataStream = new StreamData();
-  const encoder = new TextEncoder();
   let generator: AsyncGenerator<ChatResponseChunk> | undefined;
 
   const closeStreams = (controller: ReadableStreamDefaultController) => {
@@ -24,10 +19,10 @@ export async function createStreamFromWorkflowContext<Input, Output, Context>(
     dataStream.close();
   };
 
-  const mainStream = new ReadableStream({
+  const stream = new ReadableStream<EngineResponse>({
     async start(controller) {
       // Kickstart the stream by sending an empty string
-      controller.enqueue(encoder.encode(""));
+      controller.enqueue({ delta: "" } as EngineResponse);
     },
     async pull(controller) {
       while (!generator) {
@@ -46,17 +41,14 @@ export async function createStreamFromWorkflowContext<Input, Output, Context>(
         closeStreams(controller);
         return;
       }
-      const text = trimStartOfStream(chunk.delta ?? "");
-      if (text) {
-        controller.enqueue(encoder.encode(text));
+      const delta = chunk.delta ?? "";
+      if (delta) {
+        controller.enqueue({ delta } as EngineResponse);
       }
     },
   });
 
-  return {
-    stream: mainStream.pipeThrough(createStreamDataTransformer()),
-    dataStream,
-  };
+  return { stream, dataStream };
 }
 
 function handleEvent(

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

@@ -5,7 +5,6 @@ import {
   ChatMessage,
   ChatResponse,
   ChatResponseChunk,
-  LlamaCloudIndex,
   PartialToolCall,
   QueryEngineTool,
   ToolCall,
@@ -14,58 +13,17 @@ import {
 } from "llamaindex";
 import crypto from "node:crypto";
 import { getDataSource } from "../engine";
+import { createQueryEngineTool } from "../engine/tools/query-engine";
 import { AgentRunEvent } from "./type";
 
-export const getQueryEngineTools = async (): Promise<
-  QueryEngineTool[] | null
-> => {
-  const topK = process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined;
-
+export const getQueryEngineTool = async (): Promise<QueryEngineTool | null> => {
   const index = await getDataSource();
+
   if (!index) {
     return null;
   }
-  // index is LlamaCloudIndex use two query engine tools
-  if (index instanceof LlamaCloudIndex) {
-    return [
-      new QueryEngineTool({
-        queryEngine: index.asQueryEngine({
-          similarityTopK: topK,
-          retrieval_mode: "files_via_content",
-        }),
-        metadata: {
-          name: "document_retriever",
-          description: `Document retriever that retrieves entire documents from the corpus.
-  ONLY use for research questions that may require searching over entire research reports.
-  Will be slower and more expensive than chunk-level retrieval but may be necessary.`,
-        },
-      }),
-      new QueryEngineTool({
-        queryEngine: index.asQueryEngine({
-          similarityTopK: topK,
-          retrieval_mode: "chunks",
-        }),
-        metadata: {
-          name: "chunk_retriever",
-          description: `Retrieves a small set of relevant document chunks from the corpus.
-      Use for research questions that want to look up specific facts from the knowledge corpus,
-      and need entire documents.`,
-        },
-      }),
-    ];
-  } else {
-    return [
-      new QueryEngineTool({
-        queryEngine: index.asQueryEngine({
-          similarityTopK: topK,
-        }),
-        metadata: {
-          name: "retriever",
-          description: `Use this tool to retrieve information about the text corpus from the index.`,
-        },
-      }),
-    ];
-  }
+
+  return createQueryEngineTool(index);
 };
 
 /**

templates/components/reflex/contract_review/README-template.md

@@ -0,0 +1,59 @@
+This is a [LlamaIndex](https://www.llamaindex.ai/) project using [Reflex](https://reflex.dev/) bootstrapped with [`create-llama`](https://github.com/run-llama/LlamaIndexTS/tree/main/packages/create-llama) featuring automated contract review and compliance analysis use case.
+
+## Getting Started
+
+First, setup the environment with poetry:
+
+> **_Note:_** This step is not needed if you are using the dev-container.
+
+```shell
+poetry install
+```
+
+Then check the parameters that have been pre-configured in the `.env` file in this directory. (E.g. you might need to configure an `OPENAI_API_KEY` if you're using OpenAI as model provider).
+
+Second, generate the embeddings of the example document in the `./data` directory:
+
+```shell
+poetry run generate
+```
+
+Third, start app with `reflex` command:
+
+```shell
+poetry run reflex run
+```
+
+To deploy the application, refer to the Reflex deployment guide: https://reflex.dev/docs/hosting/deploy-quick-start/
+
+### UI
+
+The application provides an interactive web interface accessible at http://localhost:3000 for testing the contract review workflow.
+
+To get started:
+
+1. Upload a contract document:
+
+   - Use the provided [example_vendor_agreement.md](./example_vendor_agreement.md) for testing
+   - Or upload your own document (supported formats: PDF, TXT, Markdown, DOCX)
+
+2. Review Process:
+   - The system will automatically analyze your document against compliance guidelines
+   - By default, it uses [GDPR](./data/gdpr.pdf) as the compliance benchmark
+   - Custom guidelines can be used by adding your policy documents to the `./data` directory and running `poetry run generate` to update the embeddings
+
+The interface will display the analysis results for the compliance of the contract document.
+
+### Development
+
+You can start editing the backend workflow by modifying the [`ContractReviewWorkflow`](./app/services/contract_reviewer.py).
+
+For UI, you can start looking at the [`AppState`](./app/ui/states/app.py) code and navigating to the appropriate components.
+
+## Learn More
+
+To learn more about LlamaIndex, take a look at the following resources:
+
+- [LlamaIndex Documentation](https://docs.llamaindex.ai) - learn about LlamaIndex.
+
+You can check out [the LlamaIndex GitHub repository](https://github.com/run-llama/llama_index) - your feedback and contributions are welcome!

templates/components/reflex/contract_review/app/config.py

@@ -0,0 +1,47 @@
+DATA_DIR = "data"
+UPLOADED_DIR = "output/uploaded"
+
+# Workflow prompts
+CONTRACT_EXTRACT_PROMPT = """\
+You are given contract data below. \
+Please extract out relevant information from the contract into the defined schema - the schema is defined as a function call.\
+
+{contract_data}
+"""
+
+CONTRACT_MATCH_PROMPT = """\
+Given the following contract clause and the corresponding relevant guideline text, evaluate the compliance \
+and provide a JSON object that matches the ClauseComplianceCheck schema.
+
+**Contract Clause:**
+{clause_text}
+
+**Matched Guideline Text(s):**
+{guideline_text}
+"""
+
+
+COMPLIANCE_REPORT_SYSTEM_PROMPT = """\
+You are a compliance reporting assistant. Your task is to generate a final compliance report \
+based on the results of clause compliance checks against \
+a given set of guidelines. 
+
+Analyze the provided compliance results and produce a structured report according to the specified schema. 
+Ensure that if there are no noncompliant clauses, the report clearly indicates full compliance.
+"""
+
+COMPLIANCE_REPORT_USER_PROMPT = """\
+A set of clauses within a contract were checked against GDPR compliance guidelines for the following vendor: {vendor_name}. 
+The set of noncompliant clauses are given below.
+
+Each section includes:
+- **Clause:** The exact text of the contract clause.
+- **Guideline:** The relevant GDPR guideline text.
+- **Compliance Status:** Should be `False` for noncompliant clauses.
+- **Notes:** Additional information or explanations.
+
+{compliance_results}
+
+Based on the above compliance results, generate a final compliance report following the `ComplianceReport` schema below. 
+If there are no noncompliant clauses, the report should indicate that the contract is fully compliant.
+"""

templates/components/reflex/contract_review/app/models.py

@@ -0,0 +1,85 @@
+from typing import List, Optional
+
+from pydantic import BaseModel, Field
+
+
+class ContractClause(BaseModel):
+    clause_text: str = Field(..., description="The exact text of the clause.")
+    mentions_data_processing: bool = Field(
+        False,
+        description="True if the clause involves personal data collection or usage.",
+    )
+    mentions_data_transfer: bool = Field(
+        False,
+        description="True if the clause involves transferring personal data, especially to third parties or across borders.",
+    )
+    requires_consent: bool = Field(
+        False,
+        description="True if the clause explicitly states that user consent is needed for data activities.",
+    )
+    specifies_purpose: bool = Field(
+        False,
+        description="True if the clause specifies a clear purpose for data handling or transfer.",
+    )
+    mentions_safeguards: bool = Field(
+        False,
+        description="True if the clause mentions security measures or other safeguards for data.",
+    )
+
+
+class ContractExtraction(BaseModel):
+    vendor_name: Optional[str] = Field(
+        None, description="The vendor's name if identifiable."
+    )
+    effective_date: Optional[str] = Field(
+        None, description="The effective date of the agreement, if available."
+    )
+    governing_law: Optional[str] = Field(
+        None, description="The governing law of the contract, if stated."
+    )
+    clauses: List[ContractClause] = Field(
+        ..., description="List of extracted clauses and their relevant indicators."
+    )
+
+
+class GuidelineMatch(BaseModel):
+    guideline_text: str = Field(
+        ...,
+        description="The single most relevant guideline excerpt related to this clause.",
+    )
+    similarity_score: float = Field(
+        ...,
+        description="Similarity score indicating how closely the guideline matches the clause, e.g., between 0 and 1.",
+    )
+    relevance_explanation: Optional[str] = Field(
+        None, description="Brief explanation of why this guideline is relevant."
+    )
+
+
+class ClauseComplianceCheck(BaseModel):
+    clause_text: str = Field(
+        ..., description="The exact text of the clause from the contract."
+    )
+    matched_guideline: Optional[GuidelineMatch] = Field(
+        None, description="The most relevant guideline extracted via vector retrieval."
+    )
+    compliant: bool = Field(
+        ...,
+        description="Indicates whether the clause is considered compliant with the referenced guideline.",
+    )
+    notes: Optional[str] = Field(
+        None, description="Additional commentary or recommendations."
+    )
+
+
+class ComplianceReport(BaseModel):
+    vendor_name: Optional[str] = Field(
+        None, description="The vendor's name if identified from the contract."
+    )
+    overall_compliant: bool = Field(
+        ..., description="Indicates if the contract is considered overall compliant."
+    )
+    summary_notes: str = Field(
+        ...,
+        description="Always give a general summary or recommendations for achieving full compliance.",
+    )

templates/components/reflex/contract_review/app/services/contract_reviewer.py

@@ -0,0 +1,361 @@
+import logging
+import os
+import uuid
+from enum import Enum
+from pathlib import Path
+from typing import List
+
+from llama_index.core import SimpleDirectoryReader
+from llama_index.core.llms import LLM
+from llama_index.core.prompts import ChatPromptTemplate
+from llama_index.core.retrievers import BaseRetriever
+from llama_index.core.settings import Settings
+from llama_index.core.workflow import (
+    Context,
+    Event,
+    StartEvent,
+    StopEvent,
+    Workflow,
+    step,
+)
+
+from app.config import (
+    COMPLIANCE_REPORT_SYSTEM_PROMPT,
+    COMPLIANCE_REPORT_USER_PROMPT,
+    CONTRACT_EXTRACT_PROMPT,
+    CONTRACT_MATCH_PROMPT,
+)
+from app.engine.index import get_index
+from app.models import (
+    ClauseComplianceCheck,
+    ComplianceReport,
+    ContractClause,
+    ContractExtraction,
+)
+
+logger = logging.getLogger(__name__)
+
+
+def get_workflow():
+    index = get_index()
+    if index is None:
+        raise RuntimeError(
+            "Index not found! Please run `poetry run generate` to populate an index first."
+        )
+    return ContractReviewWorkflow(
+        guideline_retriever=index.as_retriever(),
+        llm=Settings.llm,
+        verbose=True,
+        timeout=120,
+    )
+
+
+class Step(Enum):
+    PARSE_CONTRACT = "parse_contract"
+    ANALYZE_CLAUSES = "analyze_clauses"
+    HANDLE_CLAUSE = "handle_clause"
+    GENERATE_REPORT = "generate_report"
+
+
+class ContractExtractionEvent(Event):
+    contract_extraction: ContractExtraction
+
+
+class MatchGuidelineEvent(Event):
+    request_id: str
+    clause: ContractClause
+    vendor_name: str
+
+
+class MatchGuidelineResultEvent(Event):
+    result: ClauseComplianceCheck
+
+
+class GenerateReportEvent(Event):
+    match_results: List[ClauseComplianceCheck]
+
+
+class LogEvent(Event):
+    msg: str
+    step: Step
+    data: dict = {}
+    is_step_completed: bool = False
+
+
+class ContractReviewWorkflow(Workflow):
+    """Contract review workflow."""
+
+    def __init__(
+        self,
+        guideline_retriever: BaseRetriever,
+        llm: LLM | None = None,
+        similarity_top_k: int = 20,
+        **kwargs,
+    ) -> None:
+        """Init params."""
+        super().__init__(**kwargs)
+
+        self.guideline_retriever = guideline_retriever
+
+        self.llm = llm or Settings.llm
+        self.similarity_top_k = similarity_top_k
+
+        # if not exists, create
+        out_path = Path("output") / "workflow_output"
+        if not out_path.exists():
+            out_path.mkdir(parents=True, exist_ok=True)
+            os.chmod(str(out_path), 0o0777)
+        self.output_dir = out_path
+
+    @step
+    async def parse_contract(
+        self, ctx: Context, ev: StartEvent
+    ) -> ContractExtractionEvent:
+        """Parse the contract."""
+        uploaded_contract_path = Path(ev.contract_path)
+        contract_file_name = uploaded_contract_path.name
+        # Set contract file name in context
+        await ctx.set("contract_file_name", contract_file_name)
+
+        # Parse and read the contract to documents
+        docs = SimpleDirectoryReader(
+            input_files=[str(uploaded_contract_path)]
+        ).load_data()
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Loaded document: {contract_file_name}",
+                step=Step.PARSE_CONTRACT,
+                data={
+                    "saved_path": str(uploaded_contract_path),
+                    "parsed_data": None,
+                },
+            )
+        )
+
+        # Parse the contract into a structured model
+        # See the ContractExtraction model for information we want to extract
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg="Extracting information from the document",
+                step=Step.PARSE_CONTRACT,
+                data={
+                    "saved_path": str(uploaded_contract_path),
+                    "parsed_data": None,
+                },
+            )
+        )
+        prompt = ChatPromptTemplate.from_messages([("user", CONTRACT_EXTRACT_PROMPT)])
+        contract_extraction = await self.llm.astructured_predict(
+            ContractExtraction,
+            prompt,
+            contract_data="\n".join(
+                [d.get_content(metadata_mode="all") for d in docs]  # type: ignore
+            ),
+        )
+        if not isinstance(contract_extraction, ContractExtraction):
+            raise ValueError(f"Invalid extraction from contract: {contract_extraction}")
+
+        # save output template to file
+        contract_extraction_path = Path(f"{self.output_dir}/{contract_file_name}.json")
+        with open(contract_extraction_path, "w") as fp:
+            fp.write(contract_extraction.model_dump_json())
+
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg="Extracted successfully",
+                step=Step.PARSE_CONTRACT,
+                is_step_completed=True,
+                data={
+                    "saved_path": str(contract_extraction_path),
+                    "parsed_data": contract_extraction.model_dump_json(),
+                },
+            )
+        )
+
+        return ContractExtractionEvent(contract_extraction=contract_extraction)
+
+    @step
+    async def dispatch_guideline_match(  # type: ignore
+        self, ctx: Context, ev: ContractExtractionEvent
+    ) -> MatchGuidelineEvent:
+        """For each clause in the contract, find relevant guidelines.
+
+        Use a map-reduce pattern, send each parsed clause as a MatchGuidelineEvent.
+        """
+        await ctx.set("num_clauses", len(ev.contract_extraction.clauses))
+        await ctx.set("vendor_name", ev.contract_extraction.vendor_name)
+
+        for clause in ev.contract_extraction.clauses:
+            request_id = str(uuid.uuid4())
+            ctx.send_event(
+                MatchGuidelineEvent(
+                    request_id=request_id,
+                    clause=clause,
+                    vendor_name=ev.contract_extraction.vendor_name or "Not identified",
+                )
+            )
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Created {len(ev.contract_extraction.clauses)} tasks for analyzing with the guidelines",
+                step=Step.ANALYZE_CLAUSES,
+            )
+        )
+
+    @step
+    async def handle_guideline_match(
+        self, ctx: Context, ev: MatchGuidelineEvent
+    ) -> MatchGuidelineResultEvent:
+        """Handle matching clause against guideline."""
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Handling clause for request {ev.request_id}",
+                step=Step.HANDLE_CLAUSE,
+                data={
+                    "request_id": ev.request_id,
+                    "clause_text": ev.clause.clause_text,
+                    "is_compliant": None,
+                },
+            )
+        )
+
+        # retrieve matching guideline
+        query = f"""\
+Find the relevant guideline from {ev.vendor_name} that aligns with the following contract clause:
+
+{ev.clause.clause_text}
+"""
+        guideline_docs = self.guideline_retriever.retrieve(query)
+        guideline_text = "\n\n".join([g.get_content() for g in guideline_docs])
+
+        # extract compliance from contract into a structured model
+        # see ClauseComplianceCheck model for the schema
+        prompt = ChatPromptTemplate.from_messages([("user", CONTRACT_MATCH_PROMPT)])
+        compliance_output = await self.llm.astructured_predict(
+            ClauseComplianceCheck,
+            prompt,
+            clause_text=ev.clause.model_dump_json(),
+            guideline_text=guideline_text,
+        )
+
+        if not isinstance(compliance_output, ClauseComplianceCheck):
+            raise ValueError(f"Invalid compliance check: {compliance_output}")
+
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Completed compliance check for request {ev.request_id}",
+                step=Step.HANDLE_CLAUSE,
+                is_step_completed=True,
+                data={
+                    "request_id": ev.request_id,
+                    "clause_text": ev.clause.clause_text,
+                    "is_compliant": compliance_output.compliant,
+                    "result": compliance_output,
+                },
+            )
+        )
+
+        return MatchGuidelineResultEvent(result=compliance_output)
+
+    @step
+    async def gather_guideline_match(
+        self, ctx: Context, ev: MatchGuidelineResultEvent
+    ) -> GenerateReportEvent | None:
+        """Handle matching clause against guideline."""
+        num_clauses = await ctx.get("num_clauses")
+        events = ctx.collect_events(ev, [MatchGuidelineResultEvent] * num_clauses)
+        if events is None:
+            return None
+
+        match_results = [e.result for e in events]
+        # save match results
+        contract_file_name = await ctx.get("contract_file_name")
+        match_results_path = Path(
+            f"{self.output_dir}/match_results_{contract_file_name}.jsonl"
+        )
+        with open(match_results_path, "w") as fp:
+            for mr in match_results:
+                fp.write(mr.model_dump_json() + "\n")
+
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Processed {len(match_results)} clauses",
+                step=Step.ANALYZE_CLAUSES,
+                is_step_completed=True,
+                data={"saved_path": str(match_results_path)},
+            )
+        )
+        return GenerateReportEvent(match_results=[e.result for e in events])
+
+    @step
+    async def generate_output(self, ctx: Context, ev: GenerateReportEvent) -> StopEvent:
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg="Generating Compliance Report",
+                step=Step.GENERATE_REPORT,
+                data={"is_completed": False},
+            )
+        )
+
+        # if all clauses are compliant, return a compliant result
+        non_compliant_results = [r for r in ev.match_results if not r.compliant]
+
+        # generate compliance results string
+        result_tmpl = """
+1. **Clause**: {clause}
+2. **Guideline:** {guideline}
+3. **Compliance Status:** {compliance_status}
+4. **Notes:** {notes}
+"""
+        non_compliant_strings = []
+        for nr in non_compliant_results:
+            non_compliant_strings.append(
+                result_tmpl.format(
+                    clause=nr.clause_text,
+                    guideline=nr.matched_guideline.guideline_text
+                    if nr.matched_guideline is not None
+                    else "No relevant guideline found",
+                    compliance_status=nr.compliant,
+                    notes=nr.notes,
+                )
+            )
+        non_compliant_str = "\n\n".join(non_compliant_strings)
+
+        prompt = ChatPromptTemplate.from_messages(
+            [
+                ("system", COMPLIANCE_REPORT_SYSTEM_PROMPT),
+                ("user", COMPLIANCE_REPORT_USER_PROMPT),
+            ]
+        )
+        compliance_report = await self.llm.astructured_predict(
+            ComplianceReport,
+            prompt,
+            compliance_results=non_compliant_str,
+            vendor_name=await ctx.get("vendor_name"),
+        )
+
+        # Save compliance report to file
+        contract_file_name = await ctx.get("contract_file_name")
+        compliance_report_path = Path(
+            f"{self.output_dir}/report_{contract_file_name}.json"
+        )
+        with open(compliance_report_path, "w") as fp:
+            fp.write(compliance_report.model_dump_json())
+
+        ctx.write_event_to_stream(
+            LogEvent(
+                msg=f"Compliance report saved to {compliance_report_path}",
+                step=Step.GENERATE_REPORT,
+                is_step_completed=True,
+                data={
+                    "saved_path": str(compliance_report_path),
+                    "result": compliance_report,
+                },
+            )
+        )
+
+        return StopEvent(
+            result={
+                "report": compliance_report,
+                "non_compliant_results": non_compliant_results,
+            }
+        )

templates/components/reflex/contract_review/app/ui/components/__init__.py

@@ -0,0 +1,9 @@
+from .upload import upload_component
+from .workflow import guideline_component, load_contract_component, report_component
+
+__all__ = [
+    "upload_component",
+    "load_contract_component",
+    "guideline_component",
+    "report_component",
+]

templates/components/reflex/contract_review/app/ui/components/shared/__init__.py

@@ -0,0 +1,5 @@
+from .card import card_component
+
+__all__ = [
+    "card_component",
+]

templates/components/reflex/contract_review/app/ui/components/shared/card.py

@@ -0,0 +1,20 @@
+import reflex as rx
+
+
+def card_component(
+    title: str,
+    children: rx.Component,
+    show_loading: bool = False,
+) -> rx.Component:
+    return rx.card(
+        rx.hstack(
+            rx.cond(show_loading, rx.spinner(size="2")),
+            rx.text(title, size="4"),
+            align_items="center",
+            gap="2",
+        ),
+        rx.divider(orientation="horizontal"),
+        rx.container(children),
+        width="100%",
+        background_color="var(--gray-3)",
+    )

templates/components/reflex/contract_review/app/ui/components/upload.py

@@ -0,0 +1,30 @@
+import reflex as rx
+
+from app.ui.components.shared import card_component
+from app.ui.states.app import AppState
+
+
+def upload_component() -> rx.Component:
+    return card_component(
+        title="Upload",
+        children=rx.container(
+            rx.vstack(
+                rx.upload(
+                    rx.vstack(
+                        rx.text("Drag and drop files here or click to select files"),
+                    ),
+                    on_drop=AppState.handle_upload(
+                        rx.upload_files(upload_id="upload1")
+                    ),
+                    id="upload1",
+                    border="1px dotted rgb(107,99,246)",
+                    padding="1rem",
+                ),
+                rx.cond(
+                    AppState.uploaded_file != None,  # noqa: E711
+                    rx.text(AppState.uploaded_file.file_name),  # type: ignore
+                    rx.text("No file uploaded"),
+                ),
+            ),
+        ),
+    )

templates/components/reflex/contract_review/app/ui/components/workflow/__init__.py

@@ -0,0 +1,9 @@
+from .guideline import guideline_component
+from .load import load_contract_component
+from .report import report_component
+
+__all__ = [
+    "guideline_component",
+    "load_contract_component",
+    "report_component",
+]

templates/components/reflex/contract_review/app/ui/components/workflow/guideline.py

@@ -0,0 +1,113 @@
+from typing import List
+
+import reflex as rx
+
+from app.models import ClauseComplianceCheck
+from app.ui.components.shared import card_component
+from app.ui.states.workflow import GuidelineData, GuidelineHandlerState, GuidelineState
+
+
+def guideline_handler_component(item: List) -> rx.Component:
+    _id: str = item[0]
+    status: GuidelineData = item[1]
+
+    return rx.hover_card.root(
+        rx.hover_card.trigger(
+            rx.card(
+                rx.stack(
+                    rx.container(
+                        rx.cond(
+                            ~status.is_completed,
+                            rx.spinner(size="1"),
+                            rx.cond(
+                                status.is_compliant,
+                                rx.icon(tag="check", color="green"),
+                                rx.icon(tag="x", color="red"),
+                            ),
+                        ),
+                    ),
+                    rx.flex(
+                        rx.text(status.clause_text, size="1"),
+                    ),
+                ),
+            ),
+        ),
+        rx.hover_card.content(
+            rx.cond(
+                status.is_completed,
+                guideline_result_component(status.result),  # type: ignore
+                rx.spinner(size="1"),
+            ),
+            side="right",
+        ),
+    )
+
+
+def guideline_result_component(result: ClauseComplianceCheck) -> rx.Component:
+    return rx.inset(
+        rx.table.root(
+            rx.table.header(
+                rx.table.row(
+                    rx.table.cell("Clause"),
+                    rx.table.cell("Guideline"),
+                ),
+            ),
+            rx.table.body(
+                rx.table.row(
+                    # rx.table.cell("Clause"),
+                    rx.table.cell(
+                        rx.text(result.clause_text, size="1"),
+                    ),  # type: ignore
+                    rx.table.cell(
+                        rx.text(result.matched_guideline.guideline_text, size="1"),  # type: ignore
+                    ),
+                ),
+            ),
+        ),
+        rx.container(
+            rx.cond(
+                result.compliant,
+                rx.text(
+                    result.notes,  # type: ignore
+                    size="2",
+                    color="green",
+                ),
+                rx.text(
+                    result.notes,  # type: ignore
+                    size="2",
+                    color="red",
+                ),
+            )
+        ),
+    )
+
+
+def guideline_component() -> rx.Component:
+    return rx.cond(
+        GuidelineState.is_started,
+        card_component(
+            title="Analyze the document with provided guidelines",
+            children=rx.vstack(
+                rx.vstack(
+                    rx.foreach(
+                        GuidelineState.log,
+                        lambda log: rx.box(
+                            rx.text(log["msg"]),
+                        ),
+                    ),
+                ),
+                rx.cond(
+                    GuidelineHandlerState.has_data(),  # type: ignore
+                    rx.grid(
+                        rx.foreach(
+                            GuidelineHandlerState.data,
+                            guideline_handler_component,
+                        ),
+                        columns="2",
+                        spacing="1",
+                    ),
+                ),
+            ),
+            show_loading=GuidelineState.is_running,
+        ),
+    )

templates/components/reflex/contract_review/app/ui/components/workflow/load.py

@@ -0,0 +1,22 @@
+import reflex as rx
+
+from app.ui.components.shared import card_component
+from app.ui.states.workflow import ContractLoaderState
+
+
+def load_contract_component() -> rx.Component:
+    return rx.cond(
+        ContractLoaderState.is_started,
+        card_component(
+            title="Parse contract",
+            children=rx.vstack(
+                rx.foreach(
+                    ContractLoaderState.log,
+                    lambda log: rx.box(
+                        rx.text(log["msg"]),
+                    ),
+                ),
+            ),
+            show_loading=ContractLoaderState.is_running,
+        ),
+    )

templates/components/reflex/contract_review/app/ui/components/workflow/report.py

@@ -0,0 +1,48 @@
+import reflex as rx
+
+from app.ui.components.shared import card_component
+from app.ui.states.workflow import ReportState
+
+
+def report_component() -> rx.Component:
+    return rx.cond(
+        ReportState.is_running,
+        card_component(
+            title="Conclusion",
+            show_loading=~ReportState.is_completed,  # type: ignore
+            children=rx.cond(
+                ReportState.is_completed,
+                rx.vstack(
+                    rx.box(
+                        rx.inset(
+                            rx.table.root(
+                                rx.table.body(
+                                    rx.table.row(
+                                        rx.table.cell("Vendor"),
+                                        rx.table.cell(ReportState.result.vendor_name),  # type: ignore
+                                    ),
+                                    rx.table.row(
+                                        rx.table.cell("Overall Compliance"),
+                                        rx.table.cell(
+                                            rx.cond(
+                                                ReportState.result.overall_compliant,
+                                                rx.text("Compliant", color="green"),
+                                                rx.text("Non-compliant", color="red"),
+                                            )
+                                        ),
+                                    ),
+                                    rx.table.row(
+                                        rx.table.cell("Summary Notes"),
+                                        rx.table.cell(ReportState.result.summary_notes),  # type: ignore
+                                    ),
+                                ),
+                            ),
+                        )
+                    ),
+                ),
+                rx.vstack(
+                    rx.text("Analyzing compliance results for final conclusion..."),
+                ),
+            ),
+        ),
+    )

templates/components/reflex/contract_review/app/ui/pages/index.py

@@ -0,0 +1,46 @@
+import reflex as rx
+
+from app.ui.components import (
+    guideline_component,
+    load_contract_component,
+    report_component,
+    upload_component,
+)
+from app.ui.templates import template
+
+
+@template(
+    route="/",
+    title="Structure extractor",
+)
+def index() -> rx.Component:
+    """The main index page."""
+    return rx.vstack(
+        rx.vstack(
+            rx.heading("Built by LlamaIndex", size="6"),
+            rx.text(
+                "Upload a contract to start the review process.",
+            ),
+            background_color="var(--gray-3)",
+            align_items="left",
+            justify_content="left",
+            width="100%",
+            padding="1rem",
+        ),
+        rx.container(
+            rx.vstack(
+                # Upload
+                upload_component(),
+                # Workflow
+                rx.vstack(
+                    load_contract_component(),
+                    guideline_component(),
+                    report_component(),
+                    width="100%",
+                ),
+            ),
+            width="100%",
+            padding="1rem",
+        ),
+        width="100%",
+    )

templates/components/reflex/contract_review/app/ui/states/__init__.py

@@ -0,0 +1,15 @@
+from .app import AppState
+from .workflow import (
+    ContractLoaderState,
+    GuidelineHandlerState,
+    GuidelineState,
+    ReportState,
+)
+
+__all__ = [
+    "AppState",
+    "ContractLoaderState",
+    "GuidelineHandlerState",
+    "GuidelineState",
+    "ReportState",
+]

templates/components/reflex/contract_review/app/ui/states/app.py

@@ -0,0 +1,94 @@
+import logging
+import os
+from typing import List, Optional
+
+import reflex as rx
+
+from app.config import UPLOADED_DIR
+from app.services.contract_reviewer import LogEvent, Step, get_workflow
+from app.ui.states.workflow import (
+    ContractLoaderState,
+    GuidelineHandlerState,
+    GuidelineState,
+    ReportState,
+)
+
+logger = logging.getLogger(__name__)
+
+
+class UploadedFile(rx.Base):
+    file_name: str
+    size: int
+
+
+class AppState(rx.State):
+    """
+    Whole main state for the app.
+    Handle for file upload, trigger workflow and produce workflow events.
+    """
+
+    uploaded_file: Optional[UploadedFile] = None
+
+    @rx.event
+    async def handle_upload(self, files: List[rx.UploadFile]):
+        if len(files) > 1:
+            yield rx.toast.error(
+                "You can only upload one file at a time", position="top-center"
+            )
+            return
+        try:
+            file = files[0]
+            upload_data = await file.read()
+            outfile = os.path.join(UPLOADED_DIR, file.filename)
+            with open(outfile, "wb") as f:
+                f.write(upload_data)
+            self.uploaded_file = UploadedFile(
+                file_name=file.filename, size=len(upload_data)
+            )
+            yield AppState.reset_workflow
+            yield AppState.trigger_workflow
+        except Exception as e:
+            yield rx.toast.error(str(e), position="top-center")
+
+    @rx.event
+    def reset_workflow(self):
+        yield ContractLoaderState.reset_state
+        yield GuidelineState.reset_state
+        yield GuidelineHandlerState.reset_state
+        yield ReportState.reset_state
+
+    @rx.event(background=True)
+    async def trigger_workflow(self):
+        """
+        Trigger backend to start reviewing the contract in a loop.
+        Get the event from the loop and update the state.
+        """
+        if self.uploaded_file is None:
+            yield rx.toast.error("No file uploaded", position="top-center")
+        else:
+            uploaded_file_path = os.path.join(
+                UPLOADED_DIR, self.uploaded_file.file_name
+            )
+
+            try:
+                workflow = get_workflow()
+                handler = workflow.run(
+                    contract_path=uploaded_file_path,
+                )
+                async for event in handler.stream_events():
+                    if isinstance(event, LogEvent):
+                        match event.step:
+                            case Step.PARSE_CONTRACT:
+                                yield ContractLoaderState.add_log(event)
+                            case Step.ANALYZE_CLAUSES:
+                                yield GuidelineState.add_log(event)
+                            case Step.HANDLE_CLAUSE:
+                                yield GuidelineHandlerState.add_log(event)
+                            case Step.GENERATE_REPORT:
+                                yield ReportState.add_log(event)
+                # Wait for workflow completion and propagate any exceptions
+                _ = await handler
+            except Exception as e:
+                logger.error(f"Error in trigger_workflow: {e}")
+                yield rx.toast.error(str(e), position="top-center")
+                yield AppState.reset_workflow

templates/components/reflex/contract_review/app/ui/states/workflow.py

@@ -0,0 +1,137 @@
+from typing import Any, Dict, List, Optional
+
+import reflex as rx
+from pydantic import BaseModel
+
+from app.models import ClauseComplianceCheck, ComplianceReport
+from app.services.contract_reviewer import LogEvent
+from rxconfig import config as rx_config
+
+
+class ContractLoaderState(rx.State):
+    is_running: bool = False
+    is_started: bool = False
+    log: List[Dict[str, Any]] = []
+
+    @rx.event
+    async def add_log(self, log: LogEvent):
+        if not self.is_started:
+            yield ContractLoaderState.start
+        self.log.append(log.model_dump())
+        if log.is_step_completed:
+            yield ContractLoaderState.stop
+
+    def has_log(self):
+        return len(self.log) > 0
+
+    @rx.event
+    def start(self):
+        self.is_running = True
+        self.is_started = True
+
+    @rx.event
+    def stop(self):
+        self.is_running = False
+
+    @rx.event
+    def reset_state(self):
+        self.is_running = False
+        self.is_started = False
+        self.log = []
+
+
+class GuidelineState(rx.State):
+    is_running: bool = False
+    is_started: bool = False
+    log: List[Dict[str, Any]] = []
+
+    @rx.event
+    def add_log(self, log: LogEvent):
+        if not self.is_started:
+            yield GuidelineState.start
+        self.log.append(log.model_dump())
+        if log.is_step_completed:
+            yield GuidelineState.stop
+
+    def has_log(self):
+        return len(self.log) > 0
+
+    @rx.event
+    def reset_state(self):
+        self.is_running = False
+        self.is_started = False
+        self.log = []
+
+    @rx.event
+    def start(self):
+        self.is_running = True
+        self.is_started = True
+
+    @rx.event
+    def stop(self):
+        self.is_running = False
+
+
+class GuidelineData(BaseModel):
+    is_completed: bool
+    is_compliant: Optional[bool]
+    clause_text: Optional[str]
+    result: Optional[ClauseComplianceCheck] = None
+
+
+class GuidelineHandlerState(rx.State):
+    data: Dict[str, GuidelineData] = {}
+
+    def has_data(self):
+        return len(self.data) > 0
+
+    @rx.event
+    def add_log(self, log: LogEvent):
+        _id = log.data.get("request_id")
+        if _id is None:
+            return
+        is_compliant = log.data.get("is_compliant", None)
+        self.data[_id] = GuidelineData(
+            is_completed=log.is_step_completed,
+            is_compliant=is_compliant,
+            clause_text=log.data.get("clause_text", None),
+            result=log.data.get("result", None),
+        )
+        if log.is_step_completed:
+            yield self.stop(request_id=_id)
+
+    @rx.event
+    def reset_state(self):
+        self.data = {}
+
+    @rx.event
+    def stop(self, request_id: str):
+        # Update the item in the data to be completed
+        self.data[request_id].is_completed = True
+
+
+class ReportState(rx.State):
+    is_running: bool = False
+    is_completed: bool = False
+    saved_path: str = ""
+    result: Optional[ComplianceReport] = None
+
+    @rx.var()
+    def download_url(self) -> str:
+        return f"{rx_config.api_url}/api/download/{self.saved_path}"
+
+    @rx.event
+    def add_log(self, log: LogEvent):
+        if not self.is_running:
+            self.is_running = True
+        if log.is_step_completed:
+            self.is_completed = True
+            self.saved_path = log.data.get("saved_path")
+            self.result = log.data.get("result")
+
+    @rx.event
+    def reset_state(self):
+        self.is_running = False
+        self.is_completed = False
+        self.saved_path = ""
+        self.result = None

templates/components/reflex/contract_review/example_vendor_agreement.md

@@ -0,0 +1,142 @@
+# ACME Vendor Agreement
+
+**Effective Date:** January 1, 2024
+
+## Parties:
+
+- **Client:** LlamaCo ("Client")
+- **Vendor:** ACME Office Supply, Inc. ("Vendor")
+
+## 1. Overview
+
+This Vendor Agreement ("Agreement") sets forth the terms and conditions under which ACME Office Supply, Inc. will provide office supplies, consumables, related goods ("Products"), and associated data processing services to LlamaCo.
+
+## 2. Definitions
+
+- **Personal Data:** Any information relating to an identified or identifiable natural person ('data subject').
+- **Processing:** Any operation performed on Personal Data, including collection, storage, modification, transfer, or deletion.
+- **Data Controller:** LlamaCo, who determines the purposes and means of processing Personal Data.
+- **Data Processor:** ACME Office Supply, Inc., who processes Personal Data on behalf of the Controller.
+
+## 3. Data Protection and Privacy
+
+### 3.1 Scope of Processing
+
+Vendor shall process Personal Data only:
+
+- To fulfill orders and manage deliveries
+- To provide customer support services
+- To maintain business records
+- To comply with legal obligations
+
+### 3.2 Data Subject Rights
+
+Vendor shall:
+
+- Respond to data subject requests within 30 days
+- Provide data in a structured, commonly used format
+- Implement measures to facilitate data portability
+- Assist with data subject rights requests at no additional cost
+
+### 3.3 Data Transfers and Storage
+
+- Vendor maintains primary data centers in the United States
+- Vendor may transfer data to any country where it maintains operations
+- No prior notification required for new data storage locations
+- Vendor will rely on its standard data transfer mechanisms
+- Data may be processed by staff operating outside the EEA
+
+### 3.4 Subprocessors
+
+- Vendor may engage subprocessors without prior Client approval
+- Subprocessors may be located in any jurisdiction globally
+- Notice of new subprocessors provided within 30 days of engagement
+- Client has no right to object to new subprocessors
+
+## 4. Security Measures
+
+### 4.1 Technical and Organizational Measures
+
+Vendor shall implement appropriate measures including:
+
+- Encryption of Personal Data in transit and at rest
+- Access controls and authentication
+- Regular security testing and assessments
+- Employee training on data protection
+- Incident response procedures
+
+### 4.2 Data Breaches
+
+Vendor shall:
+
+- Notify Client of any Personal Data breach within 72 hours
+- Provide details necessary to meet regulatory requirements
+- Cooperate with Client's breach investigation
+- Maintain records of all data breaches
+
+## 5. Data Retention
+
+### 5.1 Retention Period
+
+- Personal Data retained only as long as necessary
+- Standard retention period of 3 years after last transaction
+- Deletion of Personal Data upon written request
+- Backup copies retained for maximum of 6 months
+
+### 5.2 Termination
+
+Upon termination of services:
+
+- Return all Personal Data in standard format
+- Delete existing copies within 30 days
+- Provide written confirmation of deletion
+- Cease all processing activities
+
+## 6. Compliance and Audit
+
+### 6.1 Documentation
+
+Vendor shall maintain:
+
+- Records of all processing activities
+- Security measure documentation
+- Data transfer mechanisms
+- Subprocessor agreements
+
+### 6.2 Audits
+
+- Annual compliance audits permitted
+- 30 days notice required for audits
+- Vendor to provide necessary documentation
+- Client bears reasonable audit costs
+
+## 7. Liability and Indemnification
+
+### 7.1 Liability
+
+- Vendor liable for data protection violations
+- Reasonable compensation for damages
+- Coverage for regulatory fines where applicable
+- Joint liability as required by law
+
+## 8. Governing Law
+
+This Agreement shall be governed by the laws of Ireland, without regard to its conflict of laws principles.
+
+---
+
+IN WITNESS WHEREOF, the parties have executed this Agreement as of the Effective Date.
+
+**LlamaCo**
+
+By: **_
+Name: [Authorized Representative]  
+Title: [Title]  
+Date: _**
+
+**ACME Office Supply, Inc.**
+
+By: **_  
+Name: [Authorized Representative]  
+Title: [Title]  
+Date: _**

templates/components/reflex/contract_review/pyproject.toml

@@ -0,0 +1,44 @@
+[tool]
+[tool.poetry]
+name = "app"
+version = "0.1.0"
+description = ""
+authors = [ "Marcus Schiesser <mail@marcusschiesser.de>" ]
+readme = "README.md"
+
+[tool.poetry.scripts]
+generate = "app.engine.generate:generate_datasource"
+
+[tool.poetry.dependencies]
+python = "^3.11,<4.0"
+fastapi = "^0.109.1"
+python-dotenv = "^1.0.0"
+pydantic = "<2.10"
+llama-index = "^0.12.1"
+cachetools = "^5.3.3"
+reflex = "^0.6.2.post1"
+
+[tool.poetry.dependencies.uvicorn]
+extras = [ "standard" ]
+version = "^0.23.2"
+
+[tool.poetry.dependencies.docx2txt]
+version = "^0.8"
+
+[tool.poetry.dependencies.llama-index-llms-openai]
+version = "^0.3.2"
+
+[tool.poetry.dependencies.llama-index-embeddings-openai]
+version = "^0.3.1"
+
+[tool.poetry.dependencies.llama-index-agent-openai]
+version = "^0.4.0"
+
+
+[tool.poetry.group.dev.dependencies]
+pytest-asyncio = "^0.25.0"
+pytest = "^8.3.4"
+
+[build-system]
+requires = [ "poetry-core" ]
+build-backend = "poetry.core.masonry.api"

templates/components/reflex/extractor/app/ui/components/upload.py

@@ -2,6 +2,7 @@ import os
 from typing import List
 
 import reflex as rx
+
 from app.engine.generate import generate_datasource
 
 
@@ -78,10 +79,10 @@ def upload_component() -> rx.Component:
             UploadedFilesState.uploaded_files,
             lambda file: rx.card(
                 rx.stack(
-                    rx.text(file.file_name, size="sm"),
+                    rx.text(file.file_name, size="2"),
                     rx.button(
                         "x",
-                        size="sm",
+                        size="2",
                         on_click=UploadedFilesState.remove_file(file.file_name),
                     ),
                     justify="between",

templates/components/reflex/extractor/app/ui/pages/__init__.py

@@ -0,0 +1 @@
+from .index import index as index

templates/components/reflex/extractor/pyproject.toml

@@ -13,7 +13,8 @@ python = "^3.11,<4.0"
 fastapi = "^0.109.1"
 uvicorn = { extras = ["standard"], version = "^0.23.2" }
 python-dotenv = "^1.0.0"
-llama-index = "^0.11.1"
+pydantic = "<2.10"
+llama-index = "^0.12.1"
 cachetools = "^5.3.3"
 reflex = "^0.6.2.post1"
 

templates/components/routers/python/sandbox.py

@@ -17,11 +17,11 @@ import logging
 import os
 import uuid
 from dataclasses import asdict
-from typing import Any, Dict, List, Optional, Union
+from typing import Any, Dict, List, Optional
 
 from app.engine.tools.artifact import CodeArtifact
 from app.services.file import FileService
-from e2b_code_interpreter import CodeInterpreter, Sandbox  # type: ignore
+from e2b_code_interpreter import Sandbox  # type: ignore
 from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel
 
@@ -60,6 +60,14 @@ class FileUpload(BaseModel):
     name: str
 
 
+SUPPORTED_TEMPLATES = [
+    "nextjs-developer",
+    "vue-developer",
+    "streamlit-developer",
+    "gradio-developer",
+]
+
+
 @sandbox_router.post("")
 async def create_sandbox(request: Request):
     request_data = await request.json()
@@ -79,33 +87,22 @@ async def create_sandbox(request: Request):
             status_code=400, detail="Could not create artifact from the request data"
         )
 
-    sbx = None
-
-    # Create an interpreter or a sandbox
-    if artifact.template == "code-interpreter-multilang":
-        sbx = CodeInterpreter(api_key=os.getenv("E2B_API_KEY"), timeout=SANDBOX_TIMEOUT)
-        logger.debug(f"Created code interpreter {sbx}")
-    else:
-        sbx = Sandbox(
-            api_key=os.getenv("E2B_API_KEY"),
-            template=artifact.template,
-            metadata={"template": artifact.template, "user_id": "default"},
-            timeout=SANDBOX_TIMEOUT,
-        )
-        logger.debug(f"Created sandbox {sbx}")
+    sbx = Sandbox(
+        api_key=os.getenv("E2B_API_KEY"),
+        template=(
+            None if artifact.template not in SUPPORTED_TEMPLATES else artifact.template
+        ),
+        metadata={"template": artifact.template, "user_id": "default"},
+        timeout=SANDBOX_TIMEOUT,
+    )
+    logger.debug(f"Created sandbox {sbx}")
 
     # Install packages
     if artifact.has_additional_dependencies:
-        if isinstance(sbx, CodeInterpreter):
-            sbx.notebook.exec_cell(artifact.install_dependencies_command)
-            logger.debug(
-                f"Installed dependencies: {', '.join(artifact.additional_dependencies)} in code interpreter {sbx}"
-            )
-        elif isinstance(sbx, Sandbox):
-            sbx.commands.run(artifact.install_dependencies_command)
-            logger.debug(
-                f"Installed dependencies: {', '.join(artifact.additional_dependencies)} in sandbox {sbx}"
-            )
+        sbx.commands.run(artifact.install_dependencies_command)
+        logger.debug(
+            f"Installed dependencies: {', '.join(artifact.additional_dependencies)} in sandbox {sbx}"
+        )
 
     # Copy files
     if len(sandbox_files) > 0:
@@ -122,7 +119,7 @@ async def create_sandbox(request: Request):
 
     # Execute code or return a URL to the running sandbox
     if artifact.template == "code-interpreter-multilang":
-        result = sbx.notebook.exec_cell(artifact.code or "")
+        result = sbx.run_code(artifact.code or "")
         output_urls = _download_cell_results(result.results)
         runtime_error = asdict(result.error) if result.error else None
         return ExecutionResult(
@@ -145,7 +142,7 @@ async def create_sandbox(request: Request):
 
 
 def _upload_files(
-    sandbox: Union[CodeInterpreter, Sandbox],
+    sandbox: Sandbox,
     sandbox_files: List[str] = [],
 ) -> None:
     for file_path in sandbox_files:

templates/components/services/python/suggestion.py

@@ -60,10 +60,12 @@ class NextQuestionSuggestion:
             return None
 
     @classmethod
-    def _extract_questions(cls, text: str) -> List[str]:
+    def _extract_questions(cls, text: str) -> List[str] | None:
         content_match = re.search(r"```(.*?)```", text, re.DOTALL)
-        content = content_match.group(1) if content_match else ""
-        return content.strip().split("\n")
+        content = content_match.group(1) if content_match else None
+        if not content:
+            return None
+        return [q.strip() for q in content.split("\n") if q.strip()]
 
     @classmethod
     async def suggest_next_questions(