Release 0.2.19 (#365)

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

.github/workflows/e2e.yml

@@ -9,8 +9,75 @@ env:
   POETRY_VERSION: "1.6.1"
 
 jobs:
-  e2e:
-    name: create-llama
+  e2e-python:
+    name: python
+    timeout-minutes: 60
+    strategy:
+      fail-fast: true
+      matrix:
+        node-version: [20]
+        python-version: ["3.11"]
+        os: [macos-latest, windows-latest, ubuntu-22.04]
+        frameworks: ["fastapi"]
+        datasources: ["--no-files", "--example-file", "--llamacloud"]
+    defaults:
+      run:
+        shell: bash
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install Poetry
+        uses: snok/install-poetry@v1
+        with:
+          version: ${{ env.POETRY_VERSION }}
+
+      - uses: pnpm/action-setup@v3
+
+      - name: Setup Node.js ${{ matrix.node-version }}
+        uses: actions/setup-node@v4
+        with:
+          node-version: ${{ matrix.node-version }}
+          cache: "pnpm"
+
+      - name: Install dependencies
+        run: pnpm install
+
+      - name: Install Playwright Browsers
+        run: pnpm exec playwright install --with-deps
+        working-directory: .
+
+      - name: Build create-llama
+        run: pnpm run build
+        working-directory: .
+
+      - name: Install
+        run: pnpm run pack-install
+        working-directory: .
+
+      - name: Run Playwright tests for Python
+        run: pnpm run e2e:python
+        env:
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
+          FRAMEWORK: ${{ matrix.frameworks }}
+          DATASOURCE: ${{ matrix.datasources }}
+        working-directory: .
+
+      - uses: actions/upload-artifact@v3
+        if: always()
+        with:
+          name: playwright-report-python
+          path: ./playwright-report/
+          retention-days: 30
+
+  e2e-typescript:
+    name: typescript
     timeout-minutes: 60
     strategy:
       fail-fast: true
@@ -18,7 +85,7 @@ jobs:
         node-version: [18, 20]
         python-version: ["3.11"]
         os: [macos-latest, windows-latest, ubuntu-22.04]
-        frameworks: ["nextjs", "express", "fastapi"]
+        frameworks: ["nextjs", "express"]
         datasources: ["--no-files", "--example-file"]
     defaults:
       run:
@@ -60,8 +127,8 @@ jobs:
         run: pnpm run pack-install
         working-directory: .
 
-      - name: Run Playwright tests
-        run: pnpm run e2e
+      - name: Run Playwright tests for TypeScript
+        run: pnpm run e2e:typescript
         env:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
@@ -72,6 +139,6 @@ jobs:
       - uses: actions/upload-artifact@v3
         if: always()
         with:
-          name: playwright-report
+          name: playwright-report-typescript
           path: ./playwright-report/
           retention-days: 30

.github/workflows/release.yml

@@ -17,6 +17,9 @@ jobs:
 
       - uses: pnpm/action-setup@v3
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:

.husky/pre-commit

@@ -1,2 +1,3 @@
 pnpm format
 pnpm lint
+uvx ruff format --check templates/

CHANGELOG.md

@@ -1,5 +1,57 @@
 # create-llama
 
+## 0.2.19
+
+### Patch Changes
+
+- 3d41488: feat: use selected llamacloud for multiagent
+
+## 0.2.18
+
+### Patch Changes
+
+- 75e1f61: Fix cannot query public document from llamacloud
+- 88220f1: fix workflow doesn't stop when user presses stop generation button
+- 75e1f61: Fix typescript templates cannot upload file to llamacloud
+- 88220f1: Bump llama_index@0.11.17
+
+## 0.2.17
+
+### Patch Changes
+
+- cd3fcd0: bump: use LlamaIndexTS 0.6.18
+- 6335de1: Fix using LlamaCloud selector does not use the configured values in the environment (Python)
+
+## 0.2.16
+
+### Patch Changes
+
+- 0e78ba4: Fix: programmatically ensure index for LlamaCloud
+- 0e78ba4: Fix .env not loaded on poetry run generate
+- 7f4ac22: Don't need to run generate script for LlamaCloud
+- 5263bde: Use selected LlamaCloud index in multi-agent template
+
+## 0.2.15
+
+### Patch Changes
+
+- 16e6124: Bump package for llamatrace observability
+- 3790ca0: Add multi-agent task selector for TS template
+- d18f039: Add e2b code artifact tool for the FastAPI template
+
+## 0.2.14
+
+### Patch Changes
+
+- 5a7216e: feat: implement artifact tool in TS
+
+## 0.2.13
+
+### Patch Changes
+
+- 04ddebc: Add publisher agent to multi-agents for generating documents (PDF and HTML)
+- 04ddebc: Allow tool selection for multi-agents (Python and TS)
+
 ## 0.2.12
 
 ### Patch Changes

e2e/python/resolve_dependencies.spec.ts

@@ -0,0 +1,237 @@
+import { expect, test } from "@playwright/test";
+import { exec } from "child_process";
+import fs from "fs";
+import path from "path";
+import util from "util";
+import { TemplateFramework, TemplateVectorDB } from "../../helpers/types";
+import { RunCreateLlamaOptions, createTestDir, runCreateLlama } from "../utils";
+
+const execAsync = util.promisify(exec);
+
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "fastapi";
+const dataSource: string = process.env.DATASOURCE
+  ? process.env.DATASOURCE
+  : "--example-file";
+
+// TODO: add support for other templates
+
+if (
+  dataSource === "--example-file" // XXX: this test provides its own data source - only trigger it on one data source (usually the CI matrix will trigger multiple data sources)
+) {
+  // vectorDBs, tools, and data source combinations to test
+  const vectorDbs: TemplateVectorDB[] = [
+    "mongo",
+    "pg",
+    "pinecone",
+    "milvus",
+    "astra",
+    "qdrant",
+    "chroma",
+    "weaviate",
+  ];
+
+  const toolOptions = [
+    "wikipedia.WikipediaToolSpec",
+    "google.GoogleSearchToolSpec",
+    "document_generator",
+    "artifact",
+  ];
+
+  const dataSources = [
+    "--example-file",
+    "--web-source https://www.example.com",
+    "--db-source mysql+pymysql://user:pass@localhost:3306/mydb",
+  ];
+
+  const observabilityOptions = ["llamatrace", "traceloop"];
+
+  test.describe("Mypy check", () => {
+    test.describe.configure({ retries: 0 });
+
+    // Test vector databases
+    for (const vectorDb of vectorDbs) {
+      test(`Mypy check for vectorDB: ${vectorDb}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb,
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (vectorDb !== "none") {
+          if (vectorDb === "pg") {
+            expect(pyprojectContent).toContain(
+              "llama-index-vector-stores-postgres",
+            );
+          } else {
+            expect(pyprojectContent).toContain(
+              `llama-index-vector-stores-${vectorDb}`,
+            );
+          }
+        }
+      });
+    }
+
+    // Test tools
+    for (const tool of toolOptions) {
+      test(`Mypy check for tool: ${tool}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb: "none",
+            tools: tool,
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (tool === "wikipedia.WikipediaToolSpec") {
+          expect(pyprojectContent).toContain("wikipedia");
+        }
+        if (tool === "google.GoogleSearchToolSpec") {
+          expect(pyprojectContent).toContain("google");
+        }
+      });
+    }
+
+    // Test data sources
+    for (const dataSource of dataSources) {
+      const dataSourceType = dataSource.split(" ")[0];
+      test(`Mypy check for data source: ${dataSourceType}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource,
+            vectorDb: "none",
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (dataSource.includes("--web-source")) {
+          expect(pyprojectContent).toContain("llama-index-readers-web");
+        }
+        if (dataSource.includes("--db-source")) {
+          expect(pyprojectContent).toContain("llama-index-readers-database");
+        }
+      });
+    }
+
+    // Test observability options
+    for (const observability of observabilityOptions) {
+      test(`Mypy check for observability: ${observability}`, async () => {
+        const cwd = await createTestDir();
+
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb: "none",
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability,
+          },
+        });
+      });
+    }
+  });
+}
+
+async function createAndCheckLlamaProject({
+  options,
+}: {
+  options: RunCreateLlamaOptions;
+}): Promise<{ pyprojectPath: string; projectPath: string }> {
+  const result = await runCreateLlama(options);
+  const name = result.projectName;
+  const projectPath = path.join(options.cwd, name);
+
+  // Check if the app folder exists
+  expect(fs.existsSync(projectPath)).toBeTruthy();
+
+  // Check if pyproject.toml exists
+  const pyprojectPath = path.join(projectPath, "pyproject.toml");
+  expect(fs.existsSync(pyprojectPath)).toBeTruthy();
+
+  const env = {
+    ...process.env,
+    POETRY_VIRTUALENVS_IN_PROJECT: "true",
+  };
+
+  // Run poetry install
+  try {
+    const { stdout: installStdout, stderr: installStderr } = await execAsync(
+      "poetry install",
+      { cwd: projectPath, env },
+    );
+    console.log("poetry install stdout:", installStdout);
+    console.error("poetry install stderr:", installStderr);
+  } catch (error) {
+    console.error("Error running poetry install:", error);
+    throw error;
+  }
+
+  // Run poetry run mypy
+  try {
+    const { stdout: mypyStdout, stderr: mypyStderr } = await execAsync(
+      "poetry run mypy .",
+      { cwd: projectPath, env },
+    );
+    console.log("poetry run mypy stdout:", mypyStdout);
+    console.error("poetry run mypy stderr:", mypyStderr);
+  } catch (error) {
+    console.error("Error running mypy:", error);
+    throw error;
+  }
+
+  // If we reach this point without throwing an error, the test passes
+  expect(true).toBeTruthy();
+
+  return { pyprojectPath, projectPath };
+}

e2e/resolve_python_dependencies.spec.ts

@@ -1,139 +0,0 @@
-import { expect, test } from "@playwright/test";
-import { exec } from "child_process";
-import fs from "fs";
-import path from "path";
-import util from "util";
-import { TemplateFramework, TemplateVectorDB } from "../helpers/types";
-import { createTestDir, runCreateLlama } from "./utils";
-
-const execAsync = util.promisify(exec);
-
-const templateFramework: TemplateFramework = process.env.FRAMEWORK
-  ? (process.env.FRAMEWORK as TemplateFramework)
-  : "fastapi";
-const dataSource: string = process.env.DATASOURCE
-  ? process.env.DATASOURCE
-  : "--example-file";
-
-if (
-  templateFramework == "fastapi" && // test is only relevant for fastapi
-  process.version.startsWith("v20.") && // XXX: Only run for Node.js version 20 (CI matrix will trigger other versions)
-  dataSource === "--example-file" // XXX: this test provides its own data source - only trigger it on one data source (usually the CI matrix will trigger multiple data sources)
-) {
-  // vectorDBs, tools, and data source combinations to test
-  const vectorDbs: TemplateVectorDB[] = [
-    "mongo",
-    "pg",
-    "pinecone",
-    "milvus",
-    "astra",
-    "qdrant",
-    "chroma",
-    "weaviate",
-  ];
-
-  const toolOptions = [
-    "wikipedia.WikipediaToolSpec",
-    "google.GoogleSearchToolSpec",
-  ];
-
-  const dataSources = [
-    "--example-file",
-    "--web-source https://www.example.com",
-    "--db-source mysql+pymysql://user:pass@localhost:3306/mydb",
-  ];
-
-  test.describe("Test resolve python dependencies", () => {
-    for (const vectorDb of vectorDbs) {
-      for (const tool of toolOptions) {
-        for (const dataSource of dataSources) {
-          const dataSourceType = dataSource.split(" ")[0];
-          const optionDescription = `vectorDb: ${vectorDb}, tools: ${tool}, dataSource: ${dataSourceType}`;
-
-          test(`options: ${optionDescription}`, async () => {
-            const cwd = await createTestDir();
-
-            const result = await runCreateLlama({
-              cwd,
-              templateType: "streaming",
-              templateFramework: "fastapi",
-              dataSource,
-              vectorDb,
-              port: 3000, // port
-              externalPort: 8000, // externalPort
-              postInstallAction: "none", // postInstallAction
-              templateUI: undefined, // ui
-              appType: "--no-frontend", // appType
-              llamaCloudProjectName: undefined, // llamaCloudProjectName
-              llamaCloudIndexName: undefined, // llamaCloudIndexName
-              tools: tool,
-            });
-            const name = result.projectName;
-
-            // Check if the app folder exists
-            const dirExists = fs.existsSync(path.join(cwd, name));
-            expect(dirExists).toBeTruthy();
-
-            // Check if pyproject.toml exists
-            const pyprojectPath = path.join(cwd, name, "pyproject.toml");
-            const pyprojectExists = fs.existsSync(pyprojectPath);
-            expect(pyprojectExists).toBeTruthy();
-
-            // Run poetry lock
-            try {
-              const { stdout, stderr } = await execAsync(
-                "poetry config virtualenvs.in-project true && poetry lock --no-update",
-                {
-                  cwd: path.join(cwd, name),
-                },
-              );
-              console.log("poetry lock stdout:", stdout);
-              console.error("poetry lock stderr:", stderr);
-            } catch (error) {
-              console.error("Error running poetry lock:", error);
-              throw error;
-            }
-
-            // Check if poetry.lock file was created
-            const poetryLockExists = fs.existsSync(
-              path.join(cwd, name, "poetry.lock"),
-            );
-            expect(poetryLockExists).toBeTruthy();
-
-            // Verify that specific dependencies are in pyproject.toml
-            const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
-            if (vectorDb !== "none") {
-              if (vectorDb === "pg") {
-                expect(pyprojectContent).toContain(
-                  "llama-index-vector-stores-postgres",
-                );
-              } else {
-                expect(pyprojectContent).toContain(
-                  `llama-index-vector-stores-${vectorDb}`,
-                );
-              }
-            }
-            if (tool !== "none") {
-              if (tool === "wikipedia.WikipediaToolSpec") {
-                expect(pyprojectContent).toContain("wikipedia");
-              }
-              if (tool === "google.GoogleSearchToolSpec") {
-                expect(pyprojectContent).toContain("google");
-              }
-            }
-
-            // Check for data source specific dependencies
-            if (dataSource.includes("--web-source")) {
-              expect(pyprojectContent).toContain("llama-index-readers-web");
-            }
-            if (dataSource.includes("--db-source")) {
-              expect(pyprojectContent).toContain(
-                "llama-index-readers-database ",
-              );
-            }
-          });
-        }
-      }
-    }
-  });
-}

e2e/resolve_ts_dependencies.spec.ts

@@ -1,102 +0,0 @@
-import { expect, test } from "@playwright/test";
-import { exec } from "child_process";
-import fs from "fs";
-import path from "path";
-import util from "util";
-import { TemplateFramework, TemplateVectorDB } from "../helpers/types";
-import { createTestDir, runCreateLlama } from "./utils";
-
-const execAsync = util.promisify(exec);
-
-const templateFramework: TemplateFramework = process.env.FRAMEWORK
-  ? (process.env.FRAMEWORK as TemplateFramework)
-  : "nextjs";
-const dataSource: string = process.env.DATASOURCE
-  ? process.env.DATASOURCE
-  : "--example-file";
-
-if (
-  templateFramework == "nextjs" ||
-  templateFramework == "express" // test is only relevant for TS projects
-) {
-  const llamaParseOptions = [true, false];
-  // vectorDBs combinations to test
-  const vectorDbs: TemplateVectorDB[] = [
-    "mongo",
-    "pg",
-    "qdrant",
-    "pinecone",
-    "milvus",
-    "astra",
-    "chroma",
-    "llamacloud",
-    "weaviate",
-  ];
-
-  test.describe("Test resolve TS dependencies", () => {
-    for (const llamaParseOpt of llamaParseOptions) {
-      for (const vectorDb of vectorDbs) {
-        const optionDescription = `vectorDb: ${vectorDb}, dataSource: ${dataSource}, llamaParse: ${llamaParseOpt}`;
-
-        test(`options: ${optionDescription}`, async () => {
-          const cwd = await createTestDir();
-
-          const result = await runCreateLlama({
-            cwd: cwd,
-            templateType: "streaming",
-            templateFramework: templateFramework,
-            dataSource: dataSource,
-            vectorDb: vectorDb,
-            port: 3000,
-            externalPort: 8000,
-            postInstallAction: "none",
-            templateUI: undefined,
-            appType: templateFramework === "nextjs" ? "" : "--no-frontend",
-            llamaCloudProjectName: undefined,
-            llamaCloudIndexName: undefined,
-            tools: undefined,
-            useLlamaParse: llamaParseOpt,
-          });
-          const name = result.projectName;
-
-          // Check if the app folder exists
-          const appDir = path.join(cwd, name);
-          const dirExists = fs.existsSync(appDir);
-          expect(dirExists).toBeTruthy();
-
-          // Install dependencies using pnpm
-          try {
-            const { stderr: installStderr } = await execAsync(
-              "pnpm install --prefer-offline",
-              {
-                cwd: appDir,
-              },
-            );
-            expect(installStderr).toBeFalsy();
-          } catch (error) {
-            console.error("Error installing dependencies:", error);
-            throw error;
-          }
-
-          // Run tsc type check and capture the output
-          try {
-            const { stdout, stderr } = await execAsync(
-              "pnpm exec tsc -b --diagnostics",
-              {
-                cwd: appDir,
-              },
-            );
-            // Check if there's any error output
-            expect(stderr).toBeFalsy();
-
-            // Log the stdout for debugging purposes
-            console.log("TypeScript type-check output:", stdout);
-          } catch (error) {
-            console.error("Error running tsc:", error);
-            throw error;
-          }
-        });
-      }
-    }
-  });
-}

e2e/shared/extractor_template.spec.ts

@@ -3,8 +3,8 @@ 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";
+import { TemplateFramework } from "../../helpers";
+import { createTestDir, runCreateLlama } from "../utils";
 
 const templateFramework: TemplateFramework = process.env.FRAMEWORK
   ? (process.env.FRAMEWORK as TemplateFramework)
@@ -16,9 +16,8 @@ const dataSource: string = process.env.DATASOURCE
 // The extractor template currently only works with FastAPI and files (and not on Windows)
 if (
   process.platform !== "win32" &&
-  templateFramework !== "nextjs" &&
-  templateFramework !== "express" &&
-  dataSource !== "--no-files"
+  templateFramework === "fastapi" &&
+  dataSource === "--example-file"
 ) {
   test.describe("Test extractor template", async () => {
     let frontendPort: number;

e2e/shared/multiagent_template.spec.ts

@@ -7,8 +7,8 @@ import type {
   TemplateFramework,
   TemplatePostInstallAction,
   TemplateUI,
-} from "../helpers";
-import { createTestDir, runCreateLlama, type AppType } from "./utils";
+} from "../../helpers";
+import { createTestDir, runCreateLlama, type AppType } from "../utils";
 
 const templateFramework: TemplateFramework = process.env.FRAMEWORK
   ? (process.env.FRAMEWORK as TemplateFramework)
@@ -66,7 +66,7 @@ test.describe(`Test multiagent template ${templateFramework} ${dataSource} ${tem
     page,
   }) => {
     await page.goto(`http://localhost:${port}`);
-    await page.fill("form input", userMessage);
+    await page.fill("form textarea", userMessage);
 
     const responsePromise = page.waitForResponse((res) =>
       res.url().includes("/api/chat"),

e2e/shared/streaming_template.spec.ts

@@ -7,8 +7,8 @@ import type {
   TemplateFramework,
   TemplatePostInstallAction,
   TemplateUI,
-} from "../helpers";
-import { createTestDir, runCreateLlama, type AppType } from "./utils";
+} from "../../helpers";
+import { createTestDir, runCreateLlama, type AppType } from "../utils";
 
 const templateFramework: TemplateFramework = process.env.FRAMEWORK
   ? (process.env.FRAMEWORK as TemplateFramework)
@@ -27,6 +27,13 @@ const userMessage =
   dataSource !== "--no-files" ? "Physical standard for letters" : "Hello";
 
 test.describe(`Test streaming template ${templateFramework} ${dataSource} ${templateUI} ${appType} ${templatePostInstallAction}`, async () => {
+  const isNode18 = process.version.startsWith("v18");
+  const isLlamaCloud = dataSource === "--llamacloud";
+  // llamacloud is using File API which is not supported on node 18
+  if (isNode18 && isLlamaCloud) {
+    test.skip(true, "Skipping tests for Node 18 and LlamaCloud data source");
+  }
+
   let port: number;
   let externalPort: number;
   let cwd: string;
@@ -72,7 +79,7 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
   }) => {
     test.skip(templatePostInstallAction !== "runApp");
     await page.goto(`http://localhost:${port}`);
-    await page.fill("form input", userMessage);
+    await page.fill("form textarea", userMessage);
     const [response] = await Promise.all([
       page.waitForResponse(
         (res) => {

e2e/typescript/resolve_dependencies.spec.ts

@@ -0,0 +1,106 @@
+import { expect, test } from "@playwright/test";
+import { exec } from "child_process";
+import fs from "fs";
+import path from "path";
+import util from "util";
+import { TemplateFramework, TemplateVectorDB } from "../../helpers/types";
+import { createTestDir, runCreateLlama } from "../utils";
+
+const execAsync = util.promisify(exec);
+
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "nextjs";
+const dataSource: string = process.env.DATASOURCE
+  ? process.env.DATASOURCE
+  : "--example-file";
+
+// vectorDBs combinations to test
+const vectorDbs: TemplateVectorDB[] = [
+  "mongo",
+  "pg",
+  "qdrant",
+  "pinecone",
+  "milvus",
+  "astra",
+  "chroma",
+  "llamacloud",
+  "weaviate",
+];
+
+test.describe("Test resolve TS dependencies", () => {
+  // Test vector DBs without LlamaParse
+  for (const vectorDb of vectorDbs) {
+    const optionDescription = `vectorDb: ${vectorDb}, dataSource: ${dataSource}`;
+
+    test(`Vector DB test - ${optionDescription}`, async () => {
+      await runTest(vectorDb, false);
+    });
+  }
+
+  // Test LlamaParse with vectorDB 'none'
+  test(`LlamaParse test - vectorDb: none, dataSource: ${dataSource}, llamaParse: true`, async () => {
+    await runTest("none", true);
+  });
+
+  async function runTest(
+    vectorDb: TemplateVectorDB | "none",
+    useLlamaParse: boolean,
+  ) {
+    const cwd = await createTestDir();
+
+    const result = await runCreateLlama({
+      cwd: cwd,
+      templateType: "streaming",
+      templateFramework: templateFramework,
+      dataSource: dataSource,
+      vectorDb: vectorDb,
+      port: 3000,
+      externalPort: 8000,
+      postInstallAction: "none",
+      templateUI: undefined,
+      appType: templateFramework === "nextjs" ? "" : "--no-frontend",
+      llamaCloudProjectName: undefined,
+      llamaCloudIndexName: undefined,
+      tools: undefined,
+      useLlamaParse: useLlamaParse,
+    });
+    const name = result.projectName;
+
+    // Check if the app folder exists
+    const appDir = path.join(cwd, name);
+    const dirExists = fs.existsSync(appDir);
+    expect(dirExists).toBeTruthy();
+
+    // Install dependencies using pnpm
+    try {
+      const { stderr: installStderr } = await execAsync(
+        "pnpm install --prefer-offline",
+        {
+          cwd: appDir,
+        },
+      );
+    } catch (error) {
+      console.error("Error installing dependencies:", error);
+      throw error;
+    }
+
+    // Run tsc type check and capture the output
+    try {
+      const { stdout, stderr } = await execAsync(
+        "pnpm exec tsc -b --diagnostics",
+        {
+          cwd: appDir,
+        },
+      );
+      // Check if there's any error output
+      expect(stderr).toBeFalsy();
+
+      // Log the stdout for debugging purposes
+      console.log("TypeScript type-check output:", stdout);
+    } catch (error) {
+      console.error("Error running tsc:", error);
+      throw error;
+    }
+  }
+});

e2e/utils.ts

@@ -33,6 +33,7 @@ export type RunCreateLlamaOptions = {
   llamaCloudIndexName?: string;
   tools?: string;
   useLlamaParse?: boolean;
+  observability?: string;
 };
 
 export async function runCreateLlama({
@@ -50,6 +51,7 @@ export async function runCreateLlama({
   llamaCloudIndexName,
   tools,
   useLlamaParse,
+  observability,
 }: RunCreateLlamaOptions): Promise<CreateLlamaResult> {
   if (!process.env.OPENAI_API_KEY || !process.env.LLAMA_CLOUD_API_KEY) {
     throw new Error(
@@ -109,9 +111,14 @@ export async function runCreateLlama({
   if (appType) {
     commandArgs.push(appType);
   }
-  if (!useLlamaParse) {
+  if (useLlamaParse) {
+    commandArgs.push("--use-llama-parse");
+  } else {
     commandArgs.push("--no-llama-parse");
   }
+  if (observability) {
+    commandArgs.push("--observability", observability);
+  }
 
   const command = commandArgs.join(" ");
   console.log(`running command '${command}' in ${cwd}`);

helpers/env-variables.ts

@@ -65,7 +65,7 @@ const getVectorDBEnvs = (
         {
           name: "PG_CONNECTION_STRING",
           description:
-            "For generating a connection URI, see https://docs.timescale.com/use-timescale/latest/services/create-a-service\nThe PostgreSQL connection string.",
+            "For generating a connection URI, see https://supabase.com/vector\nThe PostgreSQL connection string.",
         },
       ];
 
@@ -397,12 +397,6 @@ const getEngineEnvs = (): EnvVar[] => {
       description:
         "The number of similar embeddings to return when retrieving documents.",
     },
-    {
-      name: "STREAM_TIMEOUT",
-      description:
-        "The time in milliseconds to wait for the stream to return a response.",
-      value: "60000",
-    },
   ];
 };
 
@@ -426,34 +420,35 @@ const getToolEnvs = (tools?: Tool[]): EnvVar[] => {
 const getSystemPromptEnv = (
   tools?: Tool[],
   dataSources?: TemplateDataSource[],
-  framework?: TemplateFramework,
+  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
-  let toolSystemPrompt = "";
-  tools?.forEach((tool) => {
-    const toolSystemPromptEnv = tool.envVars?.find(
-      (env) => env.name === TOOL_SYSTEM_PROMPT_ENV_VAR,
-    );
-    if (toolSystemPromptEnv) {
-      toolSystemPrompt += toolSystemPromptEnv.value + "\n";
-    }
-  });
+  // multiagent template doesn't need system prompt
+  if (template !== "multiagent") {
+    let toolSystemPrompt = "";
+    tools?.forEach((tool) => {
+      const toolSystemPromptEnv = tool.envVars?.find(
+        (env) => env.name === TOOL_SYSTEM_PROMPT_ENV_VAR,
+      );
+      if (toolSystemPromptEnv) {
+        toolSystemPrompt += toolSystemPromptEnv.value + "\n";
+      }
+    });
 
-  const systemPrompt = toolSystemPrompt
-    ? `\"${toolSystemPrompt}\"`
-    : defaultSystemPrompt;
+    const systemPrompt = toolSystemPrompt
+      ? `\"${toolSystemPrompt}\"`
+      : defaultSystemPrompt;
 
-  const systemPromptEnv = [
-    {
+    systemPromptEnv.push({
       name: "SYSTEM_PROMPT",
       description: "The system prompt for the AI model.",
       value: systemPrompt,
-    },
-  ];
-
+    });
+  }
   if (tools?.length == 0 && (dataSources?.length ?? 0 > 0)) {
     const citationPrompt = `'You have provided information from a knowledge base that has been passed to you in nodes of information.
 Each node has useful metadata such as node ID, file name, page, etc.
@@ -559,7 +554,7 @@ export const createBackendEnvFile = async (
     ...getToolEnvs(opts.tools),
     ...getTemplateEnvs(opts.template),
     ...getObservabilityEnvs(opts.observability),
-    ...getSystemPromptEnv(opts.tools, opts.dataSources, opts.framework),
+    ...getSystemPromptEnv(opts.tools, opts.dataSources, opts.template),
   ];
   // Render and write env file
   const content = renderEnvVar(envVars);

helpers/python.ts

@@ -123,7 +123,7 @@ const getAdditionalDependencies = (
             extras: ["rsa"],
           });
           dependencies.push({
-            name: "psycopg2",
+            name: "psycopg2-binary",
             version: "^2.9.9",
           });
           break;
@@ -280,6 +280,17 @@ const mergePoetryDependencies = (
   }
 };
 
+const copyRouterCode = async (root: string, tools: Tool[]) => {
+  // Copy sandbox router if the artifact tool is selected
+  if (tools?.some((t) => t.name === "artifact")) {
+    await copy("sandbox.py", path.join(root, "app", "api", "routers"), {
+      parents: true,
+      cwd: path.join(templatesDir, "components", "routers", "python"),
+      rename: assetRelocator,
+    });
+  }
+};
+
 export const addDependencies = async (
   projectDir: string,
   dependencies: Dependency[],
@@ -364,7 +375,12 @@ export const installPythonTemplate = async ({
   | "modelConfig"
 >) => {
   console.log("\nInitializing Python project with template:", template, "\n");
-  const templatePath = path.join(templatesDir, "types", template, framework);
+  let templatePath;
+  if (template === "extractor") {
+    templatePath = path.join(templatesDir, "types", "extractor", framework);
+  } else {
+    templatePath = path.join(templatesDir, "types", "streaming", framework);
+  }
   await copy("**", root, {
     parents: true,
     cwd: templatePath,
@@ -401,21 +417,43 @@ export const installPythonTemplate = async ({
       cwd: path.join(compPath, "services", "python"),
     });
   }
-
-  if (template === "streaming") {
-    // For the streaming template only:
+  // Copy engine code
+  if (template === "streaming" || template === "multiagent") {
     // Select and copy engine code based on data sources and tools
     let engine;
-    if (dataSources.length > 0 && (!tools || tools.length === 0)) {
-      console.log("\nNo tools selected - use optimized context chat engine\n");
-      engine = "chat";
-    } else {
+    // Multiagent always uses agent engine
+    if (template === "multiagent") {
       engine = "agent";
+    } else {
+      // For streaming, use chat engine by default
+      // Unless tools are selected, in which case use agent engine
+      if (dataSources.length > 0 && (!tools || tools.length === 0)) {
+        console.log(
+          "\nNo tools selected - use optimized context chat engine\n",
+        );
+        engine = "chat";
+      } else {
+        engine = "agent";
+      }
     }
+
+    // Copy engine code
     await copy("**", enginePath, {
       parents: true,
       cwd: path.join(compPath, "engines", "python", engine),
     });
+
+    // Copy router code
+    await copyRouterCode(root, tools ?? []);
+  }
+
+  if (template === "multiagent") {
+    // Copy multi-agent code
+    await copy("**", path.join(root), {
+      parents: true,
+      cwd: path.join(compPath, "multiagent", "python"),
+      rename: assetRelocator,
+    });
   }
 
   console.log("Adding additional dependencies");
@@ -439,7 +477,7 @@ export const installPythonTemplate = async ({
     if (observability === "llamatrace") {
       addOnDependencies.push({
         name: "llama-index-callbacks-arize-phoenix",
-        version: "^0.1.6",
+        version: "^0.2.1",
       });
     }
 

helpers/tools.ts

@@ -110,13 +110,36 @@ For better results, you can specify the region parameter to get results from a s
       },
     ],
   },
+  {
+    display: "Document generator",
+    name: "document_generator",
+    supportedFrameworks: ["fastapi", "nextjs", "express"],
+    dependencies: [
+      {
+        name: "xhtml2pdf",
+        version: "^0.2.14",
+      },
+      {
+        name: "markdown",
+        version: "^3.7",
+      },
+    ],
+    type: ToolType.LOCAL,
+    envVars: [
+      {
+        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
+        description: "System prompt for document generator tool.",
+        value: `If user request for a report or a post, use document generator tool to create a file and reply with the link to the file.`,
+      },
+    ],
+  },
   {
     display: "Code Interpreter",
     name: "interpreter",
     dependencies: [
       {
         name: "e2b_code_interpreter",
-        version: "0.0.7",
+        version: "0.0.10",
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
@@ -139,6 +162,33 @@ For better results, you can specify the region parameter to get results from a s
       },
     ],
   },
+  {
+    display: "Artifact Code Generator",
+    name: "artifact",
+    // Using pre-release version of e2b_code_interpreter
+    // TODO: Update to stable version when 0.0.11 is released
+    dependencies: [
+      {
+        name: "e2b_code_interpreter",
+        version: "^0.0.11b38",
+      },
+    ],
+    supportedFrameworks: ["fastapi", "express", "nextjs"],
+    type: ToolType.LOCAL,
+    envVars: [
+      {
+        name: "E2B_API_KEY",
+        description:
+          "E2B_API_KEY key is required to run artifact code generator tool. Get it here: https://e2b.dev/docs/getting-started/api-key",
+      },
+      {
+        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
+        description: "System prompt for artifact code generator tool.",
+        value:
+          "You are a code assistant that can generate and execute code using its tools. Don't generate code yourself, use the provided tools instead. Do not show the code or sandbox url in chat, just describe the steps to build the application based on the code that is generated by your tools. Do not describe how to run the code, just the steps to build the application.",
+      },
+    ],
+  },
   {
     display: "OpenAPI action",
     name: "openapi_action.OpenAPIActionToolSpec",

helpers/typescript.ts

@@ -157,7 +157,10 @@ export const installTSTemplate = async ({
   // Select and copy engine code based on data sources and tools
   let engine;
   tools = tools ?? [];
-  if (dataSources.length > 0 && tools.length === 0) {
+  // multiagent template always uses agent engine
+  if (template === "multiagent") {
+    engine = "agent";
+  } else if (dataSources.length > 0 && tools.length === 0) {
     console.log("\nNo tools selected - use optimized context chat engine\n");
     engine = "chat";
   } else {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "create-llama",
-  "version": "0.2.12",
+  "version": "0.2.19",
   "description": "Create LlamaIndex-powered apps with one command",
   "keywords": [
     "rag",
@@ -25,6 +25,8 @@
     "clean": "rimraf --glob ./dist ./templates/**/__pycache__ ./templates/**/node_modules ./templates/**/poetry.lock",
     "dev": "ncc build ./index.ts -w -o dist/",
     "e2e": "playwright test",
+    "e2e:python": "playwright test e2e/shared e2e/python",
+    "e2e:typescript": "playwright test e2e/shared e2e/typescript",
     "format": "prettier --ignore-unknown --cache --check .",
     "format:write": "prettier --ignore-unknown --write .",
     "lint": "eslint . --ignore-pattern dist --ignore-pattern e2e/cache",

questions.ts

@@ -141,12 +141,10 @@ export const getDataSourceChoices = (
     });
   }
   if (selectedDataSource === undefined || selectedDataSource.length === 0) {
-    if (template !== "multiagent") {
-      choices.push({
-        title: "No datasource",
-        value: "none",
-      });
-    }
+    choices.push({
+      title: "No datasource",
+      value: "none",
+    });
     choices.push({
       title:
         process.platform !== "linux"
@@ -734,8 +732,10 @@ export const askQuestions = async (
     }
   }
 
-  if (!program.tools && program.template === "streaming") {
-    // TODO: allow to select tools also for multi-agent framework
+  if (
+    !program.tools &&
+    (program.template === "streaming" || program.template === "multiagent")
+  ) {
     if (ciInfo.isCI) {
       program.tools = getPrefOrDefault("tools");
     } else {

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

@@ -1,17 +1,19 @@
 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
 
 
-def get_chat_engine(filters=None, params=None, event_handlers=None):
+def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
     system_prompt = os.getenv("SYSTEM_PROMPT")
     top_k = int(os.getenv("TOP_K", 0))
-    tools = []
+    tools: List[BaseTool] = []
     callback_manager = CallbackManager(handlers=event_handlers or [])
 
     # Add query tool if index exists
@@ -25,7 +27,8 @@ def get_chat_engine(filters=None, params=None, event_handlers=None):
         tools.append(query_engine_tool)
 
     # Add additional tools
-    tools += ToolFactory.from_env()
+    configured_tools: List[BaseTool] = ToolFactory.from_env()
+    tools.extend(configured_tools)
 
     return AgentRunner.from_llm(
         llm=Settings.llm,

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

@@ -1,8 +1,10 @@
-import os
-import yaml
 import importlib
-from llama_index.core.tools.tool_spec.base import BaseToolSpec
+import os
+from typing import Dict, List, Union
+
+import yaml  # type: ignore
 from llama_index.core.tools.function_tool import FunctionTool
+from llama_index.core.tools.tool_spec.base import BaseToolSpec
 
 
 class ToolType:
@@ -16,7 +18,8 @@ class ToolFactory:
         ToolType.LOCAL: "app.engine.tools",
     }
 
-    def load_tools(tool_type: str, tool_name: str, config: dict) -> list[FunctionTool]:
+    @staticmethod
+    def load_tools(tool_type: str, tool_name: str, config: dict) -> List[FunctionTool]:
         source_package = ToolFactory.TOOL_SOURCE_PACKAGE_MAP[tool_type]
         try:
             if "ToolSpec" in tool_name:
@@ -40,14 +43,34 @@ class ToolFactory:
             raise ValueError(f"Failed to load tool {tool_name}: {e}")
 
     @staticmethod
-    def from_env() -> list[FunctionTool]:
-        tools = []
+    def from_env(
+        map_result: bool = False,
+    ) -> Union[Dict[str, List[FunctionTool]], List[FunctionTool]]:
+        """
+        Load tools from the configured file.
+
+        Args:
+            map_result: If True, return a map of tool names to their corresponding tools.
+
+        Returns:
+            A dictionary of tool names to lists of FunctionTools if map_result is True,
+            otherwise a list of FunctionTools.
+        """
+        tools: Union[Dict[str, List[FunctionTool]], List[FunctionTool]] = (
+            {} if map_result else []
+        )
+
         if os.path.exists("config/tools.yaml"):
             with open("config/tools.yaml", "r") as f:
                 tool_configs = yaml.safe_load(f)
                 for tool_type, config_entries in tool_configs.items():
                     for tool_name, config in config_entries.items():
-                        tools.extend(
-                            ToolFactory.load_tools(tool_type, tool_name, config)
+                        loaded_tools = ToolFactory.load_tools(
+                            tool_type, tool_name, config
                         )
+                        if map_result:
+                            tools[tool_name] = loaded_tools  # type: ignore
+                        else:
+                            tools.extend(loaded_tools)  # type: ignore
+
         return tools

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

@@ -0,0 +1,100 @@
+import logging
+from typing import Dict, List, Optional
+
+from llama_index.core.base.llms.types import ChatMessage
+from llama_index.core.settings import Settings
+from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel, Field
+
+logger = logging.getLogger(__name__)
+
+# Prompt based on https://github.com/e2b-dev/ai-artifacts
+CODE_GENERATION_PROMPT = """You are a skilled software engineer. You do not make mistakes. Generate an artifact. You can install additional dependencies. You can use one of the following templates:
+
+1. code-interpreter-multilang: "Runs code as a Jupyter notebook cell. Strong data analysis angle. Can use complex visualisation to explain results.". File: script.py. Dependencies installed: python, jupyter, numpy, pandas, matplotlib, seaborn, plotly. Port: none.
+
+2. nextjs-developer: "A Next.js 13+ app that reloads automatically. Using the pages router.". File: pages/index.tsx. Dependencies installed: nextjs@14.2.5, typescript, @types/node, @types/react, @types/react-dom, postcss, tailwindcss, shadcn. Port: 3000.
+
+3. vue-developer: "A Vue.js 3+ app that reloads automatically. Only when asked specifically for a Vue app.". File: app.vue. Dependencies installed: vue@latest, nuxt@3.13.0, tailwindcss. Port: 3000.
+
+4. streamlit-developer: "A streamlit app that reloads automatically.". File: app.py. Dependencies installed: streamlit, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 8501.
+
+5. gradio-developer: "A gradio app. Gradio Blocks/Interface should be called demo.". File: app.py. Dependencies installed: gradio, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 7860.
+
+Make sure to use the correct syntax for the programming language you're using.
+"""
+
+
+class CodeArtifact(BaseModel):
+    commentary: str = Field(
+        ...,
+        description="Describe what you're about to do and the steps you want to take for generating the artifact in great detail.",
+    )
+    template: str = Field(
+        ..., description="Name of the template used to generate the artifact."
+    )
+    title: str = Field(..., description="Short title of the artifact. Max 3 words.")
+    description: str = Field(
+        ..., description="Short description of the artifact. Max 1 sentence."
+    )
+    additional_dependencies: List[str] = Field(
+        ...,
+        description="Additional dependencies required by the artifact. Do not include dependencies that are already included in the template.",
+    )
+    has_additional_dependencies: bool = Field(
+        ...,
+        description="Detect if additional dependencies that are not included in the template are required by the artifact.",
+    )
+    install_dependencies_command: str = Field(
+        ...,
+        description="Command to install additional dependencies required by the artifact.",
+    )
+    port: Optional[int] = Field(
+        ...,
+        description="Port number used by the resulted artifact. Null when no ports are exposed.",
+    )
+    file_path: str = Field(
+        ..., description="Relative path to the file, including the file name."
+    )
+    code: str = Field(
+        ...,
+        description="Code generated by the artifact. Only runnable code is allowed.",
+    )
+
+
+class CodeGeneratorTool:
+    def __init__(self):
+        pass
+
+    def artifact(self, query: str, old_code: Optional[str] = None) -> Dict:
+        """Generate a code artifact based on the input.
+
+        Args:
+            query (str): The description of the application you want to build.
+            old_code (Optional[str], optional): The existing code to be modified. Defaults to None.
+
+        Returns:
+            Dict: A dictionary containing the generated artifact information.
+        """
+
+        if old_code:
+            user_message = f"{query}\n\nThe existing code is: \n```\n{old_code}\n```"
+        else:
+            user_message = query
+
+        messages: List[ChatMessage] = [
+            ChatMessage(role="system", content=CODE_GENERATION_PROMPT),
+            ChatMessage(role="user", content=user_message),
+        ]
+        try:
+            sllm = Settings.llm.as_structured_llm(output_cls=CodeArtifact)  # type: ignore
+            response = sllm.chat(messages)
+            data: CodeArtifact = response.raw
+            return data.model_dump()
+        except Exception as e:
+            logger.error(f"Failed to generate artifact: {str(e)}")
+            raise e
+
+
+def get_tools(**kwargs):
+    return [FunctionTool.from_defaults(fn=CodeGeneratorTool().artifact)]

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

@@ -0,0 +1,229 @@
+import logging
+import os
+import re
+from enum import Enum
+from io import BytesIO
+
+from llama_index.core.tools.function_tool import FunctionTool
+
+OUTPUT_DIR = "output/tools"
+
+
+class DocumentType(Enum):
+    PDF = "pdf"
+    HTML = "html"
+
+
+COMMON_STYLES = """
+body {
+    font-family: Arial, sans-serif;
+    line-height: 1.3;
+    color: #333;
+}
+h1, h2, h3, h4, h5, h6 {
+    margin-top: 1em;
+    margin-bottom: 0.5em;
+}
+p {
+    margin-bottom: 0.7em;
+}
+code {
+    background-color: #f4f4f4;
+    padding: 2px 4px;
+    border-radius: 4px;
+}
+pre {
+    background-color: #f4f4f4;
+    padding: 10px;
+    border-radius: 4px;
+    overflow-x: auto;
+}
+table {
+    border-collapse: collapse;
+    width: 100%;
+    margin-bottom: 1em;
+}
+th, td {
+    border: 1px solid #ddd;
+    padding: 8px;
+    text-align: left;
+}
+th {
+    background-color: #f2f2f2;
+    font-weight: bold;
+}
+"""
+
+HTML_SPECIFIC_STYLES = """
+body {
+    max-width: 800px;
+    margin: 0 auto;
+    padding: 20px;
+}
+"""
+
+PDF_SPECIFIC_STYLES = """
+@page {
+    size: letter;
+    margin: 2cm;
+}
+body {
+    font-size: 11pt;
+}
+h1 { font-size: 18pt; }
+h2 { font-size: 16pt; }
+h3 { font-size: 14pt; }
+h4, h5, h6 { font-size: 12pt; }
+pre, code {
+    font-family: Courier, monospace;
+    font-size: 0.9em;
+}
+"""
+
+HTML_TEMPLATE = """
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <style>
+        {common_styles}
+        {specific_styles}
+    </style>
+</head>
+<body>
+    {content}
+</body>
+</html>
+"""
+
+
+class DocumentGenerator:
+    @classmethod
+    def _generate_html_content(cls, original_content: str) -> str:
+        """
+        Generate HTML content from the original markdown content.
+        """
+        try:
+            import markdown
+        except ImportError:
+            raise ImportError(
+                "Failed to import required modules. Please install markdown."
+            )
+
+        # Convert markdown to HTML with fenced code and table extensions
+        html_content = markdown.markdown(
+            original_content, extensions=["fenced_code", "tables"]
+        )
+        return html_content
+
+    @classmethod
+    def _generate_pdf(cls, html_content: str) -> BytesIO:
+        """
+        Generate a PDF from the HTML content.
+        """
+        try:
+            from xhtml2pdf import pisa
+        except ImportError:
+            raise ImportError(
+                "Failed to import required modules. Please install xhtml2pdf."
+            )
+
+        pdf_html = HTML_TEMPLATE.format(
+            common_styles=COMMON_STYLES,
+            specific_styles=PDF_SPECIFIC_STYLES,
+            content=html_content,
+        )
+
+        buffer = BytesIO()
+        pdf = pisa.pisaDocument(
+            BytesIO(pdf_html.encode("UTF-8")), buffer, encoding="UTF-8"
+        )
+
+        if pdf.err:
+            logging.error(f"PDF generation failed: {pdf.err}")
+            raise ValueError("PDF generation failed")
+
+        buffer.seek(0)
+        return buffer
+
+    @classmethod
+    def _generate_html(cls, html_content: str) -> str:
+        """
+        Generate a complete HTML document with the given HTML content.
+        """
+        return HTML_TEMPLATE.format(
+            common_styles=COMMON_STYLES,
+            specific_styles=HTML_SPECIFIC_STYLES,
+            content=html_content,
+        )
+
+    @classmethod
+    def generate_document(
+        cls, original_content: str, document_type: str, file_name: str
+    ) -> str:
+        """
+        To generate document as PDF or HTML file.
+        Parameters:
+            original_content: str (markdown style)
+            document_type: str (pdf or html) specify the type of the file format based on the use case
+            file_name: str (name of the document file) must be a valid file name, no extensions needed
+        Returns:
+            str (URL to the document file): A file URL ready to serve.
+        """
+        try:
+            document_type = DocumentType(document_type.lower())
+        except ValueError:
+            raise ValueError(
+                f"Invalid document type: {document_type}. Must be 'pdf' or 'html'."
+            )
+        # Always generate html content first
+        html_content = cls._generate_html_content(original_content)
+
+        # Based on the type of document, generate the corresponding file
+        if document_type == DocumentType.PDF:
+            content = cls._generate_pdf(html_content)
+            file_extension = "pdf"
+        elif document_type == DocumentType.HTML:
+            content = BytesIO(cls._generate_html(html_content).encode("utf-8"))
+            file_extension = "html"
+        else:
+            raise ValueError(f"Unexpected document type: {document_type}")
+
+        file_name = cls._validate_file_name(file_name)
+        file_path = os.path.join(OUTPUT_DIR, f"{file_name}.{file_extension}")
+
+        cls._write_to_file(content, file_path)
+
+        file_url = f"{os.getenv('FILESERVER_URL_PREFIX')}/{file_path}"
+        return file_url
+
+    @staticmethod
+    def _write_to_file(content: BytesIO, file_path: str):
+        """
+        Write the content to a file.
+        """
+        try:
+            os.makedirs(os.path.dirname(file_path), exist_ok=True)
+            with open(file_path, "wb") as file:
+                file.write(content.getvalue())
+        except Exception as e:
+            raise e
+
+    @staticmethod
+    def _validate_file_name(file_name: str) -> str:
+        """
+        Validate the file name.
+        """
+        # Don't allow directory traversal
+        if os.path.isabs(file_name):
+            raise ValueError("File name is not allowed.")
+        # Don't allow special characters
+        if re.match(r"^[a-zA-Z0-9_.-]+$", file_name):
+            return file_name
+        else:
+            raise ValueError("File name is not allowed to contain special characters.")
+
+
+def get_tools(**kwargs):
+    return [FunctionTool.from_defaults(DocumentGenerator.generate_document)]

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

@@ -21,16 +21,50 @@ def duckduckgo_search(
             "Please install it by running: `poetry add duckduckgo_search` or `pip install duckduckgo_search`"
         )
 
-    params = {
-        "keywords": query,
-        "region": region,
-        "max_results": max_results,
-    }
     results = []
     with DDGS() as ddg:
-        results = list(ddg.text(**params))
+        results = list(
+            ddg.text(
+                keywords=query,
+                region=region,
+                max_results=max_results,
+            )
+        )
+    return results
+
+
+def duckduckgo_image_search(
+    query: str,
+    region: str = "wt-wt",
+    max_results: int = 10,
+):
+    """
+    Use this function to search for images in DuckDuckGo.
+    Args:
+        query (str): The query to search in DuckDuckGo.
+        region Optional(str): The region to be used for the search in [country-language] convention, ex us-en, uk-en, ru-ru, etc...
+        max_results Optional(int): The maximum number of results to be returned. Default is 10.
+    """
+    try:
+        from duckduckgo_search import DDGS
+    except ImportError:
+        raise ImportError(
+            "duckduckgo_search package is required to use this function."
+            "Please install it by running: `poetry add duckduckgo_search` or `pip install duckduckgo_search`"
+        )
+    with DDGS() as ddg:
+        results = list(
+            ddg.images(
+                keywords=query,
+                region=region,
+                max_results=max_results,
+            )
+        )
     return results
 
 
 def get_tools(**kwargs):
-    return [FunctionTool.from_defaults(duckduckgo_search)]
+    return [
+        FunctionTool.from_defaults(duckduckgo_search),
+        FunctionTool.from_defaults(duckduckgo_image_search),
+    ]

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

@@ -1,10 +1,11 @@
+import logging
 import os
 import uuid
-import logging
-import requests
 from typing import Optional
-from pydantic import BaseModel, Field
+
+import requests
 from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel, Field
 
 logger = logging.getLogger(__name__)
 
@@ -26,7 +27,7 @@ class ImageGeneratorToolOutput(BaseModel):
 
 class ImageGeneratorTool:
     _IMG_OUTPUT_FORMAT = "webp"
-    _IMG_OUTPUT_DIR = "output/tool"
+    _IMG_OUTPUT_DIR = "output/tools"
     _IMG_GEN_API = "https://api.stability.ai/v2beta/stable-image/generate/core"
 
     def __init__(self, api_key: str = None):

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

@@ -1,13 +1,13 @@
-import os
-import logging
 import base64
+import logging
+import os
 import uuid
-from pydantic import BaseModel
-from typing import List, Dict, Optional
-from llama_index.core.tools import FunctionTool
+from typing import Dict, List, Optional
+
 from e2b_code_interpreter import CodeInterpreter
 from e2b_code_interpreter.models import Logs
-
+from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
 
@@ -26,7 +26,7 @@ class E2BToolOutput(BaseModel):
 
 
 class E2BCodeInterpreter:
-    output_dir = "output/tool"
+    output_dir = "output/tools"
 
     def __init__(self, api_key: str = None):
         if api_key is None:

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):
+def get_chat_engine(filters=None, 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))
@@ -43,6 +43,6 @@ def get_chat_engine(filters=None, params=None, event_handlers=None):
         memory=memory,
         system_prompt=system_prompt,
         retriever=retriever,
-        node_postprocessors=node_postprocessors,
+        node_postprocessors=node_postprocessors,  # type: ignore
         callback_manager=callback_manager,
     )

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

@@ -1,6 +1,6 @@
 import {
+  BaseChatEngine,
   BaseToolWithCall,
-  ChatEngine,
   OpenAIAgent,
   QueryEngineTool,
 } from "llamaindex";
@@ -45,7 +45,7 @@ export async function createChatEngine(documentIds?: string[], params?: any) {
   const agent = new OpenAIAgent({
     tools,
     systemPrompt: process.env.SYSTEM_PROMPT,
-  }) as unknown as ChatEngine;
+  }) as unknown as BaseChatEngine;
 
   return agent;
 }

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

@@ -0,0 +1,129 @@
+import type { JSONSchemaType } from "ajv";
+import {
+  BaseTool,
+  ChatMessage,
+  JSONValue,
+  Settings,
+  ToolMetadata,
+} from "llamaindex";
+
+// prompt based on https://github.com/e2b-dev/ai-artifacts
+const CODE_GENERATION_PROMPT = `You are a skilled software engineer. You do not make mistakes. Generate an artifact. You can install additional dependencies. You can use one of the following templates:\n
+
+1. code-interpreter-multilang: "Runs code as a Jupyter notebook cell. Strong data analysis angle. Can use complex visualisation to explain results.". File: script.py. Dependencies installed: python, jupyter, numpy, pandas, matplotlib, seaborn, plotly. Port: none.
+
+2. nextjs-developer: "A Next.js 13+ app that reloads automatically. Using the pages router.". File: pages/index.tsx. Dependencies installed: nextjs@14.2.5, typescript, @types/node, @types/react, @types/react-dom, postcss, tailwindcss, shadcn. Port: 3000.
+
+3. vue-developer: "A Vue.js 3+ app that reloads automatically. Only when asked specifically for a Vue app.". File: app.vue. Dependencies installed: vue@latest, nuxt@3.13.0, tailwindcss. Port: 3000.
+
+4. streamlit-developer: "A streamlit app that reloads automatically.". File: app.py. Dependencies installed: streamlit, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 8501.
+
+5. gradio-developer: "A gradio app. Gradio Blocks/Interface should be called demo.". File: app.py. Dependencies installed: gradio, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 7860.
+
+Provide detail information about the artifact you're about to generate in the following JSON format with the following keys:
+  
+commentary: Describe what you're about to do and the steps you want to take for generating the artifact in great detail.
+template: Name of the template used to generate the artifact.
+title: Short title of the artifact. Max 3 words.
+description: Short description of the artifact. Max 1 sentence.
+additional_dependencies: Additional dependencies required by the artifact. Do not include dependencies that are already included in the template.
+has_additional_dependencies: Detect if additional dependencies that are not included in the template are required by the artifact.
+install_dependencies_command: Command to install additional dependencies required by the artifact.
+port: Port number used by the resulted artifact. Null when no ports are exposed.
+file_path: Relative path to the file, including the file name.
+code: Code generated by the artifact. Only runnable code is allowed.
+
+Make sure to use the correct syntax for the programming language you're using. Make sure to generate only one code file. If you need to use CSS, make sure to include the CSS in the code file using Tailwind CSS syntax.
+`;
+
+// detail information to execute code
+export type CodeArtifact = {
+  commentary: string;
+  template: string;
+  title: string;
+  description: string;
+  additional_dependencies: string[];
+  has_additional_dependencies: boolean;
+  install_dependencies_command: string;
+  port: number | null;
+  file_path: string;
+  code: string;
+};
+
+export type CodeGeneratorParameter = {
+  requirement: string;
+  oldCode?: string;
+};
+
+export type CodeGeneratorToolParams = {
+  metadata?: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>>;
+};
+
+const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>> =
+  {
+    name: "artifact",
+    description: `Generate a code artifact based on the input. Don't call this tool if the user has not asked for code generation. E.g. if the user asks to write a description or specification, don't call this tool.`,
+    parameters: {
+      type: "object",
+      properties: {
+        requirement: {
+          type: "string",
+          description: "The description of the application you want to build.",
+        },
+        oldCode: {
+          type: "string",
+          description: "The existing code to be modified",
+          nullable: true,
+        },
+      },
+      required: ["requirement"],
+    },
+  };
+
+export class CodeGeneratorTool implements BaseTool<CodeGeneratorParameter> {
+  metadata: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>>;
+
+  constructor(params?: CodeGeneratorToolParams) {
+    this.metadata = params?.metadata || DEFAULT_META_DATA;
+  }
+
+  async call(input: CodeGeneratorParameter) {
+    try {
+      const artifact = await this.generateArtifact(
+        input.requirement,
+        input.oldCode,
+      );
+      return artifact as JSONValue;
+    } catch (error) {
+      return { isError: true };
+    }
+  }
+
+  // Generate artifact (code, environment, dependencies, etc.)
+  async generateArtifact(
+    query: string,
+    oldCode?: string,
+  ): Promise<CodeArtifact> {
+    const userMessage = `
+    ${query}
+    ${oldCode ? `The existing code is: \n\`\`\`${oldCode}\`\`\`` : ""}
+    `;
+    const messages: ChatMessage[] = [
+      { role: "system", content: CODE_GENERATION_PROMPT },
+      { role: "user", content: userMessage },
+    ];
+    try {
+      const response = await Settings.llm.chat({ messages });
+      const content = response.message.content.toString();
+      const jsonContent = content
+        .replace(/^```json\s*|\s*```$/g, "")
+        .replace(/^`+|`+$/g, "")
+        .trim();
+      const artifact = JSON.parse(jsonContent) as CodeArtifact;
+      return artifact;
+    } catch (error) {
+      console.log("Failed to generate artifact", error);
+      throw error;
+    }
+  }
+}

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

@@ -0,0 +1,142 @@
+import { JSONSchemaType } from "ajv";
+import { BaseTool, ToolMetadata } from "llamaindex";
+import { marked } from "marked";
+import path from "node:path";
+import { saveDocument } from "../../llamaindex/documents/helper";
+
+const OUTPUT_DIR = "output/tools";
+
+type DocumentParameter = {
+  originalContent: string;
+  fileName: string;
+};
+
+const DEFAULT_METADATA: ToolMetadata<JSONSchemaType<DocumentParameter>> = {
+  name: "document_generator",
+  description:
+    "Generate HTML document from markdown content. Return a file url to the document",
+  parameters: {
+    type: "object",
+    properties: {
+      originalContent: {
+        type: "string",
+        description: "The original markdown content to convert.",
+      },
+      fileName: {
+        type: "string",
+        description: "The name of the document file (without extension).",
+      },
+    },
+    required: ["originalContent", "fileName"],
+  },
+};
+
+const COMMON_STYLES = `
+  body {
+    font-family: Arial, sans-serif;
+    line-height: 1.3;
+    color: #333;
+  }
+  h1, h2, h3, h4, h5, h6 {
+    margin-top: 1em;
+    margin-bottom: 0.5em;
+  }
+  p {
+    margin-bottom: 0.7em;
+  }
+  code {
+    background-color: #f4f4f4;
+    padding: 2px 4px;
+    border-radius: 4px;
+  }
+  pre {
+    background-color: #f4f4f4;
+    padding: 10px;
+    border-radius: 4px;
+    overflow-x: auto;
+  }
+  table {
+    border-collapse: collapse;
+    width: 100%;
+    margin-bottom: 1em;
+  }
+  th, td {
+    border: 1px solid #ddd;
+    padding: 8px;
+    text-align: left;
+  }
+  th {
+    background-color: #f2f2f2;
+    font-weight: bold;
+  }
+  img {
+    max-width: 90%;
+    height: auto;
+    display: block;
+    margin: 1em auto;
+    border-radius: 10px;
+  }
+`;
+
+const HTML_SPECIFIC_STYLES = `
+  body {
+    max-width: 800px;
+    margin: 0 auto;
+    padding: 20px;
+  }
+`;
+
+const HTML_TEMPLATE = `
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <style>
+        ${COMMON_STYLES}
+        ${HTML_SPECIFIC_STYLES}
+    </style>
+</head>
+<body>
+    {{content}}
+</body>
+</html>
+`;
+
+export interface DocumentGeneratorParams {
+  metadata?: ToolMetadata<JSONSchemaType<DocumentParameter>>;
+}
+
+export class DocumentGenerator implements BaseTool<DocumentParameter> {
+  metadata: ToolMetadata<JSONSchemaType<DocumentParameter>>;
+
+  constructor(params: DocumentGeneratorParams) {
+    this.metadata = params.metadata ?? DEFAULT_METADATA;
+  }
+
+  private static async generateHtmlContent(
+    originalContent: string,
+  ): Promise<string> {
+    return await marked(originalContent);
+  }
+
+  private static generateHtmlDocument(htmlContent: string): string {
+    return HTML_TEMPLATE.replace("{{content}}", htmlContent);
+  }
+
+  async call(input: DocumentParameter): Promise<string> {
+    const { originalContent, fileName } = input;
+
+    const htmlContent =
+      await DocumentGenerator.generateHtmlContent(originalContent);
+    const fileContent = DocumentGenerator.generateHtmlDocument(htmlContent);
+
+    const filePath = path.join(OUTPUT_DIR, `${fileName}.html`);
+
+    return `URL: ${await saveDocument(filePath, fileContent)}`;
+  }
+}
+
+export function getTools(): BaseTool[] {
+  return [new DocumentGenerator({})];
+}

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

@@ -5,15 +5,19 @@ import { BaseTool, ToolMetadata } from "llamaindex";
 export type DuckDuckGoParameter = {
   query: string;
   region?: string;
+  maxResults?: number;
 };
 
 export type DuckDuckGoToolParams = {
   metadata?: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>>;
 };
 
-const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>> = {
-  name: "duckduckgo",
-  description: "Use this function to search for any query in DuckDuckGo.",
+const DEFAULT_SEARCH_METADATA: ToolMetadata<
+  JSONSchemaType<DuckDuckGoParameter>
+> = {
+  name: "duckduckgo_search",
+  description:
+    "Use this function to search for information (only text) in the internet using DuckDuckGo.",
   parameters: {
     type: "object",
     properties: {
@@ -27,6 +31,12 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>> = {
           "Optional, The region to be used for the search in [country-language] convention, ex us-en, uk-en, ru-ru, etc...",
         nullable: true,
       },
+      maxResults: {
+        type: "number",
+        description:
+          "Optional, The maximum number of results to be returned. Default is 10.",
+        nullable: true,
+      },
     },
     required: ["query"],
   },
@@ -42,15 +52,18 @@ export class DuckDuckGoSearchTool implements BaseTool<DuckDuckGoParameter> {
   metadata: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>>;
 
   constructor(params: DuckDuckGoToolParams) {
-    this.metadata = params.metadata ?? DEFAULT_META_DATA;
+    this.metadata = params.metadata ?? DEFAULT_SEARCH_METADATA;
   }
 
   async call(input: DuckDuckGoParameter) {
-    const { query, region } = input;
+    const { query, region, maxResults = 10 } = input;
     const options = region ? { region } : {};
+    // Temporarily sleep to reduce overloading the DuckDuckGo
+    await new Promise((resolve) => setTimeout(resolve, 1000));
+
     const searchResults = await search(query, options);
 
-    return searchResults.results.map((result) => {
+    return searchResults.results.slice(0, maxResults).map((result) => {
       return {
         title: result.title,
         description: result.description,
@@ -59,3 +72,7 @@ export class DuckDuckGoSearchTool implements BaseTool<DuckDuckGoParameter> {
     });
   }
 }
+
+export function getTools() {
+  return [new DuckDuckGoSearchTool({})];
+}

templates/components/engines/typescript/agent/tools/img-gen.ts

@@ -37,7 +37,7 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<ImgGeneratorParameter>> = {
 
 export class ImgGeneratorTool implements BaseTool<ImgGeneratorParameter> {
   readonly IMG_OUTPUT_FORMAT = "webp";
-  readonly IMG_OUTPUT_DIR = "output/tool";
+  readonly IMG_OUTPUT_DIR = "output/tools";
   readonly IMG_GEN_API =
     "https://api.stability.ai/v2beta/stable-image/generate/core";
 

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

@@ -1,5 +1,10 @@
 import { BaseToolWithCall } from "llamaindex";
 import { ToolsFactory } from "llamaindex/tools/ToolsFactory";
+import { CodeGeneratorTool, CodeGeneratorToolParams } from "./code-generator";
+import {
+  DocumentGenerator,
+  DocumentGeneratorParams,
+} from "./document-generator";
 import { DuckDuckGoSearchTool, DuckDuckGoToolParams } from "./duckduckgo";
 import { ImgGeneratorTool, ImgGeneratorToolParams } from "./img-gen";
 import { InterpreterTool, InterpreterToolParams } from "./interpreter";
@@ -43,6 +48,12 @@ const toolFactory: Record<string, ToolCreator> = {
   img_gen: async (config: unknown) => {
     return [new ImgGeneratorTool(config as ImgGeneratorToolParams)];
   },
+  artifact: async (config: unknown) => {
+    return [new CodeGeneratorTool(config as CodeGeneratorToolParams)];
+  },
+  document_generator: async (config: unknown) => {
+    return [new DocumentGenerator(config as DocumentGeneratorParams)];
+  },
 };
 
 async function createLocalTools(

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

@@ -56,7 +56,7 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<InterpreterParameter>> = {
 };
 
 export class InterpreterTool implements BaseTool<InterpreterParameter> {
-  private readonly outputDir = "output/tool";
+  private readonly outputDir = "output/tools";
   private apiKey?: string;
   private fileServerURLPrefix?: string;
   metadata: ToolMetadata<JSONSchemaType<InterpreterParameter>>;

templates/components/llamaindex/typescript/documents/helper.ts

@@ -1,4 +1,5 @@
-import fs from "fs";
+import fs from "node:fs";
+import path from "node:path";
 import { getExtractors } from "../../engine/loader";
 
 const MIME_TYPE_TO_EXT: Record<string, string> = {
@@ -15,8 +16,12 @@ export async function storeAndParseFile(
   fileBuffer: Buffer,
   mimeType: string,
 ) {
+  const fileExt = MIME_TYPE_TO_EXT[mimeType];
+  if (!fileExt) throw new Error(`Unsupported document type: ${mimeType}`);
+
   const documents = await loadDocuments(fileBuffer, mimeType);
-  await saveDocument(filename, fileBuffer, mimeType);
+  const filepath = path.join(UPLOADED_FOLDER, filename);
+  await saveDocument(filepath, fileBuffer);
   for (const document of documents) {
     document.metadata = {
       ...document.metadata,
@@ -38,26 +43,31 @@ async function loadDocuments(fileBuffer: Buffer, mimeType: string) {
   return await reader.loadDataAsContent(fileBuffer);
 }
 
-async function saveDocument(
-  filename: string,
-  fileBuffer: Buffer,
-  mimeType: string,
-) {
-  const fileExt = MIME_TYPE_TO_EXT[mimeType];
-  if (!fileExt) throw new Error(`Unsupported document type: ${mimeType}`);
-
-  const filepath = `${UPLOADED_FOLDER}/${filename}`;
-  const fileurl = `${process.env.FILESERVER_URL_PREFIX}/${filepath}`;
-
-  if (!fs.existsSync(UPLOADED_FOLDER)) {
-    fs.mkdirSync(UPLOADED_FOLDER, { recursive: true });
+// Save document to file server and return the file url
+export async function saveDocument(filepath: string, content: string | Buffer) {
+  if (path.isAbsolute(filepath)) {
+    throw new Error("Absolute file paths are not allowed.");
+  }
+  const fileName = path.basename(filepath);
+  if (!/^[a-zA-Z0-9_.-]+$/.test(fileName)) {
+    throw new Error(
+      "File name is not allowed to contain any special characters.",
+    );
+  }
+  if (!process.env.FILESERVER_URL_PREFIX) {
+    throw new Error("FILESERVER_URL_PREFIX environment variable is not set.");
   }
-  await fs.promises.writeFile(filepath, fileBuffer);
 
-  console.log(`Saved document file to ${filepath}.\nURL: ${fileurl}`);
-  return {
-    filename,
-    filepath,
-    fileurl,
-  };
+  const dirPath = path.dirname(filepath);
+  await fs.promises.mkdir(dirPath, { recursive: true });
+
+  if (typeof content === "string") {
+    await fs.promises.writeFile(filepath, content, "utf-8");
+  } else {
+    await fs.promises.writeFile(filepath, content);
+  }
+
+  const fileurl = `${process.env.FILESERVER_URL_PREFIX}/${filepath}`;
+  console.log(`Saved document to ${filepath}. Reachable at URL: ${fileurl}`);
+  return fileurl;
 }

templates/components/llamaindex/typescript/documents/upload.ts

@@ -16,14 +16,26 @@ export async function uploadDocument(
     // trigger LlamaCloudIndex API to upload the file and run the pipeline
     const projectId = await index.getProjectId();
     const pipelineId = await index.getPipelineId();
-    return [
-      await LLamaCloudFileService.addFileToPipeline(
-        projectId,
-        pipelineId,
-        new File([fileBuffer], filename, { type: mimeType }),
-        { private: "true" },
-      ),
-    ];
+    try {
+      return [
+        await LLamaCloudFileService.addFileToPipeline(
+          projectId,
+          pipelineId,
+          new File([fileBuffer], filename, { type: mimeType }),
+          { private: "true" },
+        ),
+      ];
+    } catch (error) {
+      if (
+        error instanceof ReferenceError &&
+        error.message.includes("File is not defined")
+      ) {
+        throw new Error(
+          "File class is not supported in the current Node.js version. Please use Node.js 20 or higher.",
+        );
+      }
+      throw error;
+    }
   }
 
   // run the pipeline for other vector store indexes

templates/components/llamaindex/typescript/streaming/annotations.ts

@@ -1,4 +1,4 @@
-import { JSONValue } from "ai";
+import { JSONValue, Message } from "ai";
 import { MessageContent, MessageContentDetail } from "llamaindex";
 
 export type DocumentFileType = "csv" | "pdf" | "txt" | "docx";
@@ -21,13 +21,20 @@ type Annotation = {
   data: object;
 };
 
-export function retrieveDocumentIds(annotations?: JSONValue[]): string[] {
-  if (!annotations) return [];
+export function isValidMessages(messages: Message[]): boolean {
+  const lastMessage =
+    messages && messages.length > 0 ? messages[messages.length - 1] : null;
+  return lastMessage !== null && lastMessage.role === "user";
+}
+
+export function retrieveDocumentIds(messages: Message[]): string[] {
+  // retrieve document Ids from the annotations of all messages (if any)
+  const annotations = getAllAnnotations(messages);
+  if (annotations.length === 0) return [];
 
   const ids: string[] = [];
 
-  for (const annotation of annotations) {
-    const { type, data } = getValidAnnotation(annotation);
+  for (const { type, data } of annotations) {
     if (
       type === "document_file" &&
       "files" in data &&
@@ -37,9 +44,7 @@ export function retrieveDocumentIds(annotations?: JSONValue[]): string[] {
       for (const file of files) {
         if (Array.isArray(file.content.value)) {
           // it's an array, so it's an array of doc IDs
-          for (const id of file.content.value) {
-            ids.push(id);
-          }
+          ids.push(...file.content.value);
         }
       }
     }
@@ -48,24 +53,69 @@ export function retrieveDocumentIds(annotations?: JSONValue[]): string[] {
   return ids;
 }
 
-export function convertMessageContent(
-  content: string,
-  annotations?: JSONValue[],
-): MessageContent {
-  if (!annotations) return content;
+export function retrieveMessageContent(messages: Message[]): MessageContent {
+  const userMessage = messages[messages.length - 1];
   return [
     {
       type: "text",
-      text: content,
+      text: userMessage.content,
     },
-    ...convertAnnotations(annotations),
+    ...retrieveLatestArtifact(messages),
+    ...convertAnnotations(messages),
   ];
 }
 
-function convertAnnotations(annotations: JSONValue[]): MessageContentDetail[] {
+function getAllAnnotations(messages: Message[]): Annotation[] {
+  return messages.flatMap((message) =>
+    (message.annotations ?? []).map((annotation) =>
+      getValidAnnotation(annotation),
+    ),
+  );
+}
+
+// get latest artifact from annotations to append to the user message
+function retrieveLatestArtifact(messages: Message[]): MessageContentDetail[] {
+  const annotations = getAllAnnotations(messages);
+  if (annotations.length === 0) return [];
+
+  for (const { type, data } of annotations.reverse()) {
+    if (
+      type === "tools" &&
+      "toolCall" in data &&
+      "toolOutput" in data &&
+      typeof data.toolCall === "object" &&
+      typeof data.toolOutput === "object" &&
+      data.toolCall !== null &&
+      data.toolOutput !== null &&
+      "name" in data.toolCall &&
+      data.toolCall.name === "artifact"
+    ) {
+      const toolOutput = data.toolOutput as { output?: { code?: string } };
+      if (toolOutput.output?.code) {
+        return [
+          {
+            type: "text",
+            text: `The existing code is:\n\`\`\`\n${toolOutput.output.code}\n\`\`\``,
+          },
+        ];
+      }
+    }
+  }
+  return [];
+}
+
+function convertAnnotations(messages: Message[]): MessageContentDetail[] {
+  // annotations from the last user message that has annotations
+  const annotations: Annotation[] =
+    messages
+      .slice()
+      .reverse()
+      .find((message) => message.role === "user" && message.annotations)
+      ?.annotations?.map(getValidAnnotation) || [];
+  if (annotations.length === 0) return [];
+
   const content: MessageContentDetail[] = [];
-  annotations.forEach((annotation: JSONValue) => {
-    const { type, data } = getValidAnnotation(annotation);
+  annotations.forEach(({ type, data }) => {
     // convert image
     if (type === "image" && "url" in data && typeof data.url === "string") {
       content.push({
@@ -122,3 +172,26 @@ function getValidAnnotation(annotation: JSONValue): Annotation {
   }
   return { type: annotation.type, data: annotation.data };
 }
+
+// validate and get all annotations of a specific type or role from the frontend messages
+export function getAnnotations<
+  T extends Annotation["data"] = Annotation["data"],
+>(
+  messages: Message[],
+  options?: {
+    role?: Message["role"]; // message role
+    type?: Annotation["type"]; // annotation type
+  },
+): {
+  type: string;
+  data: T;
+}[] {
+  const messagesByRole = options?.role
+    ? messages.filter((msg) => msg.role === options?.role)
+    : messages;
+  const annotations = getAllAnnotations(messagesByRole);
+  const annotationsByType = options?.type
+    ? annotations.filter((a) => a.type === options.type)
+    : annotations;
+  return annotationsByType as { type: string; data: T }[];
+}

templates/components/llamaindex/typescript/streaming/events.ts

@@ -69,22 +69,13 @@ export function appendToolData(
   });
 }
 
-export function createStreamTimeout(stream: StreamData) {
-  const timeout = Number(process.env.STREAM_TIMEOUT ?? 1000 * 60 * 5); // default to 5 minutes
-  const t = setTimeout(() => {
-    appendEventData(stream, `Stream timed out after ${timeout / 1000} seconds`);
-    stream.close();
-  }, timeout);
-  return t;
-}
-
 export function createCallbackManager(stream: StreamData) {
   const callbackManager = new CallbackManager();
 
   callbackManager.on("retrieve-end", (data) => {
     const { nodes, query } = data.detail;
     appendSourceData(stream, nodes);
-    appendEventData(stream, `Retrieving context for query: '${query}'`);
+    appendEventData(stream, `Retrieving context for query: '${query.query}'`);
     appendEventData(
       stream,
       `Retrieved ${nodes.length} sources to use as context for the query`,

templates/components/loaders/python/__init__.py

@@ -1,20 +1,22 @@
 import logging
+from typing import Any, Dict, List
 
-import yaml
+import yaml  # type: ignore
 from app.engine.loaders.db import DBLoaderConfig, get_db_documents
 from app.engine.loaders.file import FileLoaderConfig, get_file_documents
 from app.engine.loaders.web import WebLoaderConfig, get_web_documents
+from llama_index.core import Document
 
 logger = logging.getLogger(__name__)
 
 
-def load_configs():
+def load_configs() -> Dict[str, Any]:
     with open("config/loaders.yaml") as f:
         configs = yaml.safe_load(f)
     return configs
 
 
-def get_documents():
+def get_documents() -> List[Document]:
     documents = []
     config = load_configs()
     for loader_type, loader_config in config.items():

templates/components/loaders/python/db.py

@@ -1,5 +1,6 @@
 import logging
 from typing import List
+
 from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
@@ -11,7 +12,13 @@ class DBLoaderConfig(BaseModel):
 
 
 def get_db_documents(configs: list[DBLoaderConfig]):
-    from llama_index.readers.database import DatabaseReader
+    try:
+        from llama_index.readers.database import DatabaseReader
+    except ImportError:
+        logger.error(
+            "Failed to import DatabaseReader. Make sure llama_index is installed."
+        )
+        raise
 
     docs = []
     for entry in configs:

templates/components/loaders/python/web.py

@@ -1,3 +1,5 @@
+from typing import List, Optional
+
 from pydantic import BaseModel, Field
 
 
@@ -8,8 +10,8 @@ class CrawlUrl(BaseModel):
 
 
 class WebLoaderConfig(BaseModel):
-    driver_arguments: list[str] = Field(default=None)
-    urls: list[CrawlUrl]
+    driver_arguments: Optional[List[str]] = Field(default_factory=list)
+    urls: List[CrawlUrl]
 
 
 def get_web_documents(config: WebLoaderConfig):

templates/components/multiagent/python/app/agents/multi.py

@@ -8,7 +8,7 @@ from app.agents.single import (
 )
 from llama_index.core.tools.types import ToolMetadata, ToolOutput
 from llama_index.core.tools.utils import create_schema_from_function
-from llama_index.core.workflow import Context, Workflow
+from llama_index.core.workflow import Context, StopEvent, Workflow
 
 
 class AgentCallTool(ContextAwareTool):
@@ -25,7 +25,11 @@ class AgentCallTool(ContextAwareTool):
             name=name,
             description=(
                 f"Use this tool to delegate a sub task to the {agent.name} agent."
-                + (f" The agent is an {agent.role}." if agent.role else "")
+                + (
+                    f" The agent is an {agent.description}."
+                    if agent.description
+                    else ""
+                )
             ),
             fn_schema=fn_schema,
         )
@@ -35,7 +39,8 @@ class AgentCallTool(ContextAwareTool):
         handler = self.agent.run(input=input)
         # bubble all events while running the agent to the calling agent
         async for ev in handler.stream_events():
-            ctx.write_event_to_stream(ev)
+            if type(ev) is not StopEvent:
+                ctx.write_event_to_stream(ev)
         ret: AgentRunResult = await handler
         response = ret.response.message.content
         return ToolOutput(

templates/components/multiagent/python/app/agents/planner.py

@@ -11,6 +11,7 @@ from llama_index.core.agent.runner.planner import (
     SubTask,
 )
 from llama_index.core.bridge.pydantic import ValidationError
+from llama_index.core.chat_engine.types import ChatMessage
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.prompts import PromptTemplate
 from llama_index.core.settings import Settings
@@ -24,6 +25,18 @@ from llama_index.core.workflow import (
     step,
 )
 
+INITIAL_PLANNER_PROMPT = """\
+Think step-by-step. Given a conversation, set of tools and a user request. Your responsibility is to create a plan to complete the task.
+The plan must adapt with the user request and the conversation.
+
+The tools available are:
+{tools_str}
+
+Conversation: {chat_history}
+
+Overall Task: {task}
+"""
+
 
 class ExecutePlanEvent(Event):
     pass
@@ -62,14 +75,21 @@ class StructuredPlannerAgent(Workflow):
         tools: List[BaseTool] | None = None,
         timeout: float = 360.0,
         refine_plan: bool = False,
+        chat_history: Optional[List[ChatMessage]] = None,
         **kwargs: Any,
     ) -> None:
         super().__init__(*args, timeout=timeout, **kwargs)
         self.name = name
         self.refine_plan = refine_plan
+        self.chat_history = chat_history
 
         self.tools = tools or []
-        self.planner = Planner(llm=llm, tools=self.tools, verbose=self._verbose)
+        self.planner = Planner(
+            llm=llm,
+            tools=self.tools,
+            initial_plan_prompt=INITIAL_PLANNER_PROMPT,
+            verbose=self._verbose,
+        )
         # The executor is keeping the memory of all tool calls and decides to call the right tool for the task
         self.executor = FunctionCallingAgent(
             name="executor",
@@ -89,7 +109,9 @@ class StructuredPlannerAgent(Workflow):
         ctx.data["streaming"] = getattr(ev, "streaming", False)
         ctx.data["task"] = ev.input
 
-        plan_id, plan = await self.planner.create_plan(input=ev.input)
+        plan_id, plan = await self.planner.create_plan(
+            input=ev.input, chat_history=self.chat_history
+        )
         ctx.data["act_plan_id"] = plan_id
 
         # inform about the new plan
@@ -106,11 +128,12 @@ class StructuredPlannerAgent(Workflow):
             ctx.data["act_plan_id"]
         )
 
-        ctx.data["num_sub_tasks"] = len(upcoming_sub_tasks)
-        # send an event per sub task
-        events = [SubTaskEvent(sub_task=sub_task) for sub_task in upcoming_sub_tasks]
-        for event in events:
-            ctx.send_event(event)
+        if upcoming_sub_tasks:
+            # Execute only the first sub-task
+            # otherwise the executor will get over-lapping messages
+            # alternatively, we could use one executor for all sub tasks
+            next_sub_task = upcoming_sub_tasks[0]
+            return SubTaskEvent(sub_task=next_sub_task)
 
         return None
 
@@ -120,7 +143,7 @@ class StructuredPlannerAgent(Workflow):
     ) -> SubTaskResultEvent:
         if self._verbose:
             print(f"=== Executing sub task: {ev.sub_task.name} ===")
-        is_last_tasks = ctx.data["num_sub_tasks"] == self.get_remaining_subtasks(ctx)
+        is_last_tasks = self.get_remaining_subtasks(ctx) == 1
         # TODO: streaming only works without plan refining
         streaming = is_last_tasks and ctx.data["streaming"] and not self.refine_plan
         handler = self.executor.run(
@@ -142,22 +165,17 @@ class StructuredPlannerAgent(Workflow):
     async def gather_results(
         self, ctx: Context, ev: SubTaskResultEvent
     ) -> ExecutePlanEvent | StopEvent:
-        # wait for all sub tasks to finish
-        num_sub_tasks = ctx.data["num_sub_tasks"]
-        results = ctx.collect_events(ev, [SubTaskResultEvent] * num_sub_tasks)
-        if results is None:
-            return None
+        result = ev
 
         upcoming_sub_tasks = self.get_upcoming_sub_tasks(ctx)
         # if no more tasks to do, stop workflow and send result of last step
         if upcoming_sub_tasks == 0:
-            return StopEvent(result=results[-1].result)
+            return StopEvent(result=result.result)
 
         if self.refine_plan:
-            # store all results for refining the plan
+            # store the result for refining the plan
             ctx.data["results"] = ctx.data.get("results", {})
-            for result in results:
-                ctx.data["results"][result.sub_task.name] = result.result
+            ctx.data["results"][result.sub_task.name] = result.result
 
             new_plan = await self.planner.refine_plan(
                 ctx.data["task"], ctx.data["act_plan_id"], ctx.data["results"]
@@ -213,7 +231,9 @@ class Planner:
             plan_refine_prompt = PromptTemplate(plan_refine_prompt)
         self.plan_refine_prompt = plan_refine_prompt
 
-    async def create_plan(self, input: str) -> Tuple[str, Plan]:
+    async def create_plan(
+        self, input: str, chat_history: Optional[List[ChatMessage]] = None
+    ) -> Tuple[str, Plan]:
         tools = self.tools
         tools_str = ""
         for tool in tools:
@@ -225,6 +245,7 @@ class Planner:
                 self.initial_plan_prompt,
                 tools_str=tools_str,
                 task=input,
+                chat_history=chat_history,
             )
         except (ValueError, ValidationError):
             if self.verbose:

templates/components/multiagent/python/app/agents/single.py

@@ -5,10 +5,8 @@ from llama_index.core.llms import ChatMessage, ChatResponse
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.settings import Settings
-from llama_index.core.tools import ToolOutput, ToolSelection
+from llama_index.core.tools import FunctionTool, ToolOutput, ToolSelection
 from llama_index.core.tools.types import BaseTool
-from llama_index.core.tools import FunctionTool
-
 from llama_index.core.workflow import (
     Context,
     Event,
@@ -64,14 +62,14 @@ class FunctionCallingAgent(Workflow):
         timeout: float = 360.0,
         name: str,
         write_events: bool = True,
-        role: Optional[str] = None,
+        description: str | None = None,
         **kwargs: Any,
     ) -> None:
         super().__init__(*args, verbose=verbose, timeout=timeout, **kwargs)
         self.tools = tools or []
         self.name = name
-        self.role = role
         self.write_events = write_events
+        self.description = description
 
         if llm is None:
             llm = Settings.llm

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

@@ -0,0 +1,44 @@
+import logging
+
+from app.api.routers.models import (
+    ChatData,
+)
+from app.api.routers.vercel_response import VercelStreamResponse
+from app.engine.engine import get_chat_engine
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Request, status
+
+chat_router = r = APIRouter()
+
+logger = logging.getLogger("uvicorn")
+
+
+@r.post("")
+async def chat(
+    request: Request,
+    data: ChatData,
+    background_tasks: BackgroundTasks,
+):
+    try:
+        last_message_content = data.get_last_message_content()
+        messages = data.get_history_messages(include_agent_messages=True)
+
+        # The chat API supports passing private document filters and chat params
+        # but agent workflow does not support them yet
+        # ignore chat params and use all documents for now
+        # TODO: generate filters based on doc_ids
+        params = data.data or {}
+        engine = get_chat_engine(chat_history=messages, params=params)
+
+        event_handler = engine.run(input=last_message_content, streaming=True)
+        return VercelStreamResponse(
+            request=request,
+            chat_data=data,
+            event_handler=event_handler,
+            events=engine.stream_events(),
+        )
+    except Exception as e:
+        logger.exception("Error in chat engine", exc_info=True)
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"Error in chat engine: {e}",
+        ) from e

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

@@ -1,6 +1,6 @@
+import asyncio
 import json
 import logging
-from asyncio import Task
 from typing import AsyncGenerator, List
 
 from aiostream import stream
@@ -15,12 +15,94 @@ logger = logging.getLogger("uvicorn")
 
 class VercelStreamResponse(StreamingResponse):
     """
-    Class to convert the response from the chat engine to the streaming format expected by Vercel
+    Base class to convert the response from the chat engine to the streaming format expected by Vercel
     """
 
     TEXT_PREFIX = "0:"
     DATA_PREFIX = "8:"
 
+    def __init__(self, request: Request, chat_data: ChatData, *args, **kwargs):
+        self.request = request
+        self.chat_data = chat_data
+        content = self.content_generator(*args, **kwargs)
+        super().__init__(content=content)
+
+    async def content_generator(self, event_handler, events):
+        logger.info("Starting content_generator")
+        stream = self._create_stream(
+            self.request, self.chat_data, event_handler, events
+        )
+        is_stream_started = False
+        try:
+            async with stream.stream() as streamer:
+                async for output in streamer:
+                    if not is_stream_started:
+                        is_stream_started = True
+                        # Stream a blank message to start the stream
+                        yield self.convert_text("")
+
+                    yield output
+        except asyncio.CancelledError:
+            logger.info("Stopping workflow")
+            await event_handler.cancel_run()
+        except Exception as e:
+            logger.error(
+                f"Unexpected error in content_generator: {str(e)}", exc_info=True
+            )
+        finally:
+            logger.info("The stream has been stopped!")
+
+    def _create_stream(
+        self,
+        request: Request,
+        chat_data: ChatData,
+        event_handler: AgentRunResult | AsyncGenerator,
+        events: AsyncGenerator[AgentRunEvent, None],
+        verbose: bool = True,
+    ):
+        # Yield the text response
+        async def _chat_response_generator():
+            result = await event_handler
+            final_response = ""
+
+            if isinstance(result, AgentRunResult):
+                for token in result.response.message.content:
+                    final_response += token
+                    yield self.convert_text(token)
+
+            if isinstance(result, AsyncGenerator):
+                async for token in result:
+                    final_response += token.delta
+                    yield self.convert_text(token.delta)
+
+            # Generate next questions if next question prompt is configured
+            question_data = await self._generate_next_questions(
+                chat_data.messages, final_response
+            )
+            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:
+                event_response = self._event_to_response(event)
+                if verbose:
+                    logger.debug(event_response)
+                if event_response is not None:
+                    yield self.convert_data(event_response)
+
+        combine = stream.merge(_chat_response_generator(), _event_generator())
+        return combine
+
+    @staticmethod
+    def _event_to_response(event: AgentRunEvent) -> dict:
+        return {
+            "type": "agent",
+            "data": {"agent": event.name, "text": event.msg},
+        }
+
     @classmethod
     def convert_text(cls, token: str):
         # Escape newlines and double quotes to avoid breaking the stream
@@ -32,82 +114,6 @@ class VercelStreamResponse(StreamingResponse):
         data_str = json.dumps(data)
         return f"{cls.DATA_PREFIX}[{data_str}]\n"
 
-    def __init__(
-        self,
-        request: Request,
-        task: Task[AgentRunResult | AsyncGenerator],
-        events: AsyncGenerator[AgentRunEvent, None],
-        chat_data: ChatData,
-        verbose: bool = True,
-    ):
-        content = VercelStreamResponse.content_generator(
-            request, task, events, chat_data, verbose
-        )
-        super().__init__(content=content)
-
-    @classmethod
-    async def content_generator(
-        cls,
-        request: Request,
-        task: Task[AgentRunResult | AsyncGenerator],
-        events: AsyncGenerator[AgentRunEvent, None],
-        chat_data: ChatData,
-        verbose: bool = True,
-    ):
-        # Yield the text response
-        async def _chat_response_generator():
-            result = await task
-            final_response = ""
-
-            if isinstance(result, AgentRunResult):
-                for token in result.response.message.content:
-                    final_response += token
-                    yield cls.convert_text(token)
-
-            if isinstance(result, AsyncGenerator):
-                async for token in result:
-                    final_response += token.delta
-                    yield cls.convert_text(token.delta)
-
-            # Generate next questions if next question prompt is configured
-            question_data = await cls._generate_next_questions(
-                chat_data.messages, final_response
-            )
-            if question_data:
-                yield cls.convert_data(question_data)
-
-            # TODO: stream sources
-
-        # Yield the events from the event handler
-        async def _event_generator():
-            async for event in events():
-                event_response = cls._event_to_response(event)
-                if verbose:
-                    logger.debug(event_response)
-                if event_response is not None:
-                    yield cls.convert_data(event_response)
-
-        combine = stream.merge(_chat_response_generator(), _event_generator())
-
-        is_stream_started = False
-        async with combine.stream() as streamer:
-            if not is_stream_started:
-                is_stream_started = True
-                # Stream a blank message to start the stream
-                yield cls.convert_text("")
-
-            async for output in streamer:
-                yield output
-                if await request.is_disconnected():
-                    break
-
-    @staticmethod
-    def _event_to_response(event: AgentRunEvent) -> dict:
-        return {
-            "type": "agent",
-            "data": {"agent": event.name, "text": event.msg},
-        }
-
     @staticmethod
     async def _generate_next_questions(chat_history: List[Message], response: str):
         questions = await NextQuestionSuggestion.suggest_next_questions(

templates/components/multiagent/python/app/engine/engine.py

@@ -1,28 +1,28 @@
 import logging
+import os
 from typing import List, Optional
+
 from app.examples.choreography import create_choreography
 from app.examples.orchestrator import create_orchestrator
 from app.examples.workflow import create_workflow
-
-
-from llama_index.core.workflow import Workflow
 from llama_index.core.chat_engine.types import ChatMessage
-
-
-import os
+from llama_index.core.workflow import Workflow
 
 logger = logging.getLogger("uvicorn")
 
 
-def create_agent(chat_history: Optional[List[ChatMessage]] = None) -> Workflow:
+def get_chat_engine(
+    chat_history: Optional[List[ChatMessage]] = None, **kwargs
+) -> Workflow:
+    # TODO: the EXAMPLE_TYPE could be passed as a chat config parameter?
     agent_type = os.getenv("EXAMPLE_TYPE", "").lower()
     match agent_type:
         case "choreography":
-            agent = create_choreography(chat_history)
+            agent = create_choreography(chat_history, **kwargs)
         case "orchestrator":
-            agent = create_orchestrator(chat_history)
+            agent = create_orchestrator(chat_history, **kwargs)
         case _:
-            agent = create_workflow(chat_history)
+            agent = create_workflow(chat_history, **kwargs)
 
     logger.info(f"Using agent pattern: {agent_type}")
 

templates/components/multiagent/python/app/examples/choreography.py

@@ -0,0 +1,34 @@
+from textwrap import dedent
+from typing import List, Optional
+
+from app.agents.multi import AgentCallingAgent
+from app.agents.single import FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+
+
+def create_choreography(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(chat_history, **kwargs)
+    publisher = create_publisher(chat_history)
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written post to review",
+        system_prompt="You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. If the post is good, you can say 'The post is good.'",
+        chat_history=chat_history,
+    )
+    return AgentCallingAgent(
+        name="writer",
+        agents=[researcher, reviewer, publisher],
+        description="expert in writing blog posts, needs researched information and images to write a blog post",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts. You are given a task to write a blog post. Before starting to write the post, consult the researcher agent to get the information you need. Don't make up any information yourself.
+            After creating a draft for the post, send it to the reviewer agent to receive feedback and make sure to incorporate the feedback from the reviewer.
+            You can consult the reviewer and researcher a maximum of two times. Your output should contain only the blog post.
+            Finally, always request the publisher to create a document (PDF, HTML) and publish the blog post.
+        """
+        ),
+        # TODO: add chat_history support to AgentCallingAgent
+        # chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/orchestrator.py

@@ -0,0 +1,44 @@
+from textwrap import dedent
+from typing import List, Optional
+
+from app.agents.multi import AgentOrchestrator
+from app.agents.single import FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+
+
+def create_orchestrator(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(chat_history, **kwargs)
+    writer = FunctionCallingAgent(
+        name="writer",
+        description="expert in writing blog posts, need information and images to write a post",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts.
+            You are given a task to write a blog post. Do not make up any information yourself.
+            If you don't have the necessary information to write a blog post, reply "I need information about the topic to write the blog post".
+            If you need to use images, reply "I need images about the topic to write the blog post". Do not use any dummy images made up by you.
+            If you have all the information needed, write the blog post.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written blog post to review",
+        system_prompt=dedent(
+            """
+            You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post and fix any issues found yourself. You must output a final blog post.
+            A post must include at least one valid image. If not, reply "I need images about the topic to write the blog post". An image URL starting with "example" or "your website" is not valid.
+            Especially check for logical inconsistencies and proofread the post for grammar and spelling errors.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    publisher = create_publisher(chat_history)
+    return AgentOrchestrator(
+        agents=[writer, reviewer, researcher, publisher],
+        refine_plan=False,
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/publisher.py

@@ -0,0 +1,35 @@
+from textwrap import dedent
+from typing import List, Tuple
+
+from app.agents.single import FunctionCallingAgent
+from app.engine.tools import ToolFactory
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.tools import FunctionTool
+
+
+def get_publisher_tools() -> Tuple[List[FunctionTool], str, str]:
+    tools = []
+    # Get configured tools from the tools.yaml file
+    configured_tools = ToolFactory.from_env(map_result=True)
+    if "document_generator" in configured_tools.keys():
+        tools.extend(configured_tools["document_generator"])
+        prompt_instructions = dedent("""
+            Normally, reply the blog post content to the user directly. 
+            But if user requested to generate a file, use the document_generator tool to generate the file and reply the link to the file.
+        """)
+        description = "Expert in publishing the blog post, able to publish the blog post in PDF or HTML format."
+    else:
+        prompt_instructions = "You don't have a tool to generate document. Please reply the content directly."
+        description = "Expert in publishing the blog post"
+    return tools, prompt_instructions, description
+
+
+def create_publisher(chat_history: List[ChatMessage]):
+    tools, prompt_instructions, description = get_publisher_tools()
+    return FunctionCallingAgent(
+        name="publisher",
+        tools=tools,
+        description=description,
+        system_prompt=prompt_instructions,
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/researcher.py

@@ -0,0 +1,86 @@
+import os
+from textwrap import dedent
+from typing import List
+
+from app.agents.single import FunctionCallingAgent
+from app.engine.index import IndexConfig, get_index
+from app.engine.tools import ToolFactory
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.tools import QueryEngineTool, ToolMetadata
+
+
+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:
+    """
+    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)
+    researcher_tool_names = ["duckduckgo", "wikipedia.WikipediaToolSpec"]
+    configured_tools = ToolFactory.from_env(map_result=True)
+    for tool_name, tool in configured_tools.items():
+        if tool_name in researcher_tool_names:
+            tools.extend(tool)
+    return tools
+
+
+def create_researcher(chat_history: List[ChatMessage], **kwargs):
+    """
+    Researcher is an agent that take responsibility for using tools to complete a given task.
+    """
+    tools = _get_research_tools(**kwargs)
+    return FunctionCallingAgent(
+        name="researcher",
+        tools=tools,
+        description="expert in retrieving any unknown content or searching for images from the internet",
+        system_prompt=dedent(
+            """
+            You are a researcher agent. You are given a research task.
+            
+            If the conversation already includes the information and there is no new request for additional information from the user, you should return the appropriate content to the writer.
+            Otherwise, you must use tools to retrieve information or images needed for the task.
+
+            It's normal for the task to include some ambiguity. You must always think carefully about the context of the user's request to understand what are the main content needs to be retrieved.
+            Example:
+                Request: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+                ->Though: The main content is "history of the internet", while "write in English and publish in PDF format" is a requirement for other agents.
+                Your task: Look for information in English about the history of the Internet.
+                This is not your task: Create a blog post or look for how to create a PDF.
+
+                Next request: "Publish the blog post in HTML format."
+                ->Though: User just asking for a format change, the previous content is still valid.
+                Your task: Return the previous content of the post to the writer. No need to do any research.
+                This is not your task: Look for how to create an HTML file.
+
+            If you use the tools but don't find any related information, please return "I didn't find any new information for {the topic}." along with the content you found. Don't try to make up information yourself.
+            If the request doesn't need any new information because it was in the conversation history, please return "The task doesn't need any new information. Please reuse the existing content in the conversation history."
+        """
+        ),
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/workflow.py

@@ -0,0 +1,265 @@
+from textwrap import dedent
+from typing import AsyncGenerator, List, Optional
+
+from app.agents.single import AgentRunEvent, AgentRunResult, FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.prompts import PromptTemplate
+from llama_index.core.settings import Settings
+from llama_index.core.workflow import (
+    Context,
+    Event,
+    StartEvent,
+    StopEvent,
+    Workflow,
+    step,
+)
+
+
+def create_workflow(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(
+        chat_history=chat_history,
+        **kwargs,
+    )
+    publisher = create_publisher(
+        chat_history=chat_history,
+    )
+    writer = FunctionCallingAgent(
+        name="writer",
+        description="expert in writing blog posts, need information and images to write a post.",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts.
+            You are given the task of writing a blog post based on research content provided by the researcher agent. Do not invent any information yourself. 
+            It's important to read the entire conversation history to write the blog post accurately.
+            If you receive a review from the reviewer, update the post according to the feedback and return the new post content.
+            If the content is not valid (e.g., broken link, broken image, etc.), do not use it.
+            It's normal for the task to include some ambiguity, so you must define the user's initial request to write the post correctly.
+            If you update the post based on the reviewer's feedback, first explain what changes you made to the post, then provide the new post content. Do not include the reviewer's comments.
+            Example:
+                Task: "Here is the information I found about the history of the internet: 
+                Create a blog post about the history of the internet, write in English, and publish in PDF format."
+                -> Your task: Use the research content {...} to write a blog post in English.
+                -> This is not your task: Create a PDF
+                Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written blog post to review.",
+        system_prompt=dedent(
+            """
+            You are an expert in reviewing blog posts.
+            You are given a task to review a blog post. As a reviewer, it's important that your review aligns with the user's request. Please focus on the user's request when reviewing the post.
+            Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement.
+            Furthermore, proofread the post for grammar and spelling errors.
+            Only if the post is good enough for publishing should you return 'The post is good.' In all other cases, return your review.
+            It's normal for the task to include some ambiguity, so you must define the user's initial request to review the post correctly.
+            Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+            Example:
+                Task: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+                -> Your task: Review whether the main content of the post is about the history of the internet and if it is written in English.
+                -> This is not your task: Create blog post, create PDF, write in English.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    workflow = BlogPostWorkflow(
+        timeout=360, chat_history=chat_history
+    )  # Pass chat_history here
+    workflow.add_workflows(
+        researcher=researcher,
+        writer=writer,
+        reviewer=reviewer,
+        publisher=publisher,
+    )
+    return workflow
+
+
+class ResearchEvent(Event):
+    input: str
+
+
+class WriteEvent(Event):
+    input: str
+    is_good: bool = False
+
+
+class ReviewEvent(Event):
+    input: str
+
+
+class PublishEvent(Event):
+    input: str
+
+
+class BlogPostWorkflow(Workflow):
+    def __init__(
+        self, timeout: int = 360, chat_history: Optional[List[ChatMessage]] = None
+    ):
+        super().__init__(timeout=timeout)
+        self.chat_history = chat_history or []
+
+    @step()
+    async def start(self, ctx: Context, ev: StartEvent) -> ResearchEvent | PublishEvent:
+        # set streaming
+        ctx.data["streaming"] = getattr(ev, "streaming", False)
+        # start the workflow with researching about a topic
+        ctx.data["task"] = ev.input
+        ctx.data["user_input"] = ev.input
+
+        # Decision-making process
+        decision = await self._decide_workflow(ev.input, self.chat_history)
+
+        if decision != "publish":
+            return ResearchEvent(input=f"Research for this task: {ev.input}")
+        else:
+            chat_history_str = "\n".join(
+                [f"{msg.role}: {msg.content}" for msg in self.chat_history]
+            )
+            return PublishEvent(
+                input=f"Please publish content based on the chat history\n{chat_history_str}\n\n and task: {ev.input}"
+            )
+
+    async def _decide_workflow(
+        self, input: str, chat_history: List[ChatMessage]
+    ) -> str:
+        prompt_template = PromptTemplate(
+            dedent(
+                """
+                You are an expert in decision-making, helping people write and publish blog posts.
+                If the user is asking for a file or to publish content, respond with 'publish'.
+                If the user requests to write or update a blog post, respond with 'not_publish'.
+
+                Here is the chat history:
+                {chat_history}
+
+                The current user request is:
+                {input}
+
+                Given the chat history and the new user request, decide whether to publish based on existing information.
+                Decision (respond with either 'not_publish' or 'publish'):
+            """
+            )
+        )
+
+        chat_history_str = "\n".join(
+            [f"{msg.role}: {msg.content}" for msg in chat_history]
+        )
+        prompt = prompt_template.format(chat_history=chat_history_str, input=input)
+
+        output = await Settings.llm.acomplete(prompt)
+        decision = output.text.strip().lower()
+
+        return "publish" if decision == "publish" else "research"
+
+    @step()
+    async def research(
+        self, ctx: Context, ev: ResearchEvent, researcher: FunctionCallingAgent
+    ) -> WriteEvent:
+        result: AgentRunResult = await self.run_agent(ctx, researcher, ev.input)
+        content = result.response.message.content
+        return WriteEvent(
+            input=f"Write a blog post given this task: {ctx.data['task']} using this research content: {content}"
+        )
+
+    @step()
+    async def write(
+        self, ctx: Context, ev: WriteEvent, writer: FunctionCallingAgent
+    ) -> ReviewEvent | StopEvent:
+        MAX_ATTEMPTS = 2
+        ctx.data["attempts"] = ctx.data.get("attempts", 0) + 1
+        too_many_attempts = ctx.data["attempts"] > MAX_ATTEMPTS
+        if too_many_attempts:
+            ctx.write_event_to_stream(
+                AgentRunEvent(
+                    name=writer.name,
+                    msg=f"Too many attempts ({MAX_ATTEMPTS}) to write the blog post. Proceeding with the current version.",
+                )
+            )
+        if ev.is_good or too_many_attempts:
+            # too many attempts or the blog post is good - stream final response if requested
+            result = await self.run_agent(
+                ctx,
+                writer,
+                f"Based on the reviewer's feedback, refine the post and return only the final version of the post. Here's the current version: {ev.input}",
+                streaming=ctx.data["streaming"],
+            )
+            return StopEvent(result=result)
+        result: AgentRunResult = await self.run_agent(ctx, writer, ev.input)
+        ctx.data["result"] = result
+        return ReviewEvent(input=result.response.message.content)
+
+    @step()
+    async def review(
+        self, ctx: Context, ev: ReviewEvent, reviewer: FunctionCallingAgent
+    ) -> WriteEvent:
+        result: AgentRunResult = await self.run_agent(ctx, reviewer, ev.input)
+        review = result.response.message.content
+        old_content = ctx.data["result"].response.message.content
+        post_is_good = "post is good" in review.lower()
+        ctx.write_event_to_stream(
+            AgentRunEvent(
+                name=reviewer.name,
+                msg=f"The post is {'not ' if not post_is_good else ''}good enough for publishing. Sending back to the writer{' for publication.' if post_is_good else '.'}",
+            )
+        )
+        if post_is_good:
+            return WriteEvent(
+                input=f"You're blog post is ready for publication. Please respond with just the blog post. Blog post: ```{old_content}```",
+                is_good=True,
+            )
+        else:
+            return WriteEvent(
+                input=dedent(
+                    f"""
+                    Improve the writing of a given blog post by using a given review.
+                    Blog post:
+                    ```
+                    {old_content}
+                    ``` 
+
+                    Review: 
+                    ```
+                    {review}
+                    ```
+                    """
+                ),
+            )
+
+    @step()
+    async def publish(
+        self,
+        ctx: Context,
+        ev: PublishEvent,
+        publisher: FunctionCallingAgent,
+    ) -> StopEvent:
+        try:
+            result: AgentRunResult = await self.run_agent(ctx, publisher, ev.input)
+            return StopEvent(result=result)
+        except Exception as e:
+            ctx.write_event_to_stream(
+                AgentRunEvent(
+                    name=publisher.name,
+                    msg=f"Error publishing: {e}",
+                )
+            )
+            return StopEvent(result=None)
+
+    async def run_agent(
+        self,
+        ctx: Context,
+        agent: FunctionCallingAgent,
+        input: str,
+        streaming: bool = False,
+    ) -> AgentRunResult | AsyncGenerator:
+        handler = agent.run(input=input, streaming=streaming)
+        # bubble all events while running the executor to the planner
+        async for event in handler.stream_events():
+            # Don't write the StopEvent from sub task to the stream
+            if type(event) is not StopEvent:
+                ctx.write_event_to_stream(event)
+        return await handler

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

@@ -1,13 +1,13 @@
 import { StopEvent } from "@llamaindex/core/workflow";
 import { Message, streamToResponse } from "ai";
 import { Request, Response } from "express";
-import { ChatMessage, ChatResponseChunk } from "llamaindex";
+import { ChatResponseChunk } from "llamaindex";
 import { createWorkflow } from "./workflow/factory";
 import { toDataStream, workflowEventsToStreamData } from "./workflow/stream";
 
 export const chat = async (req: Request, res: Response) => {
   try {
-    const { messages }: { messages: Message[] } = req.body;
+    const { messages, data }: { messages: Message[]; data?: any } = req.body;
     const userMessage = messages.pop();
     if (!messages || !userMessage || userMessage.role !== "user") {
       return res.status(400).json({
@@ -16,8 +16,7 @@ export const chat = async (req: Request, res: Response) => {
       });
     }
 
-    const chatHistory = messages as ChatMessage[];
-    const agent = createWorkflow(chatHistory);
+    const agent = createWorkflow(messages, data);
     const result = agent.run<AsyncGenerator<ChatResponseChunk>>(
       userMessage.content,
     ) as unknown as Promise<StopEvent<AsyncGenerator<ChatResponseChunk>>>;

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

@@ -1,7 +1,7 @@
 import { initObservability } from "@/app/observability";
 import { StopEvent } from "@llamaindex/core/workflow";
 import { Message, StreamingTextResponse } from "ai";
-import { ChatMessage, ChatResponseChunk } from "llamaindex";
+import { ChatResponseChunk } from "llamaindex";
 import { NextRequest, NextResponse } from "next/server";
 import { initSettings } from "./engine/settings";
 import { createWorkflow } from "./workflow/factory";
@@ -16,7 +16,7 @@ export const dynamic = "force-dynamic";
 export async function POST(request: NextRequest) {
   try {
     const body = await request.json();
-    const { messages }: { messages: Message[] } = body;
+    const { messages, data }: { messages: Message[]; data?: any } = body;
     const userMessage = messages.pop();
     if (!messages || !userMessage || userMessage.role !== "user") {
       return NextResponse.json(
@@ -28,8 +28,7 @@ export async function POST(request: NextRequest) {
       );
     }
 
-    const chatHistory = messages as ChatMessage[];
-    const agent = createWorkflow(chatHistory);
+    const agent = createWorkflow(messages, data);
     // TODO: fix type in agent.run in LITS
     const result = agent.run<AsyncGenerator<ChatResponseChunk>>(
       userMessage.content,

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

@@ -1,33 +1,43 @@
-import { ChatMessage, QueryEngineTool } from "llamaindex";
-import { getDataSource } from "../engine";
+import { ChatMessage } from "llamaindex";
 import { FunctionCallingAgent } from "./single-agent";
+import { getQueryEngineTool, lookupTools } from "./tools";
 
-const getQueryEngineTool = async () => {
-  const index = await getDataSource();
-  if (!index) {
-    throw new Error(
-      "StorageContext is empty - call 'npm run generate' to generate the storage first.",
-    );
-  }
+export const createResearcher = async (
+  chatHistory: ChatMessage[],
+  params?: any,
+) => {
+  const queryEngineTool = await getQueryEngineTool(params);
+  const tools = (
+    await lookupTools([
+      "wikipedia_tool",
+      "duckduckgo_search",
+      "image_generator",
+    ])
+  ).concat(queryEngineTool ? [queryEngineTool] : []);
 
-  const topK = process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined;
-  return new QueryEngineTool({
-    queryEngine: index.asQueryEngine({
-      similarityTopK: topK,
-    }),
-    metadata: {
-      name: "query_index",
-      description: `Use this tool to retrieve information about the text corpus from the index.`,
-    },
-  });
-};
-
-export const createResearcher = async (chatHistory: ChatMessage[]) => {
   return new FunctionCallingAgent({
     name: "researcher",
-    tools: [await getQueryEngineTool()],
-    systemPrompt:
-      "You are a researcher agent. You are given a researching task. You must use your tools to complete the research.",
+    tools: tools,
+    systemPrompt: `You are a researcher agent. You are given a research task.
+            
+If the conversation already includes the information and there is no new request for additional information from the user, you should return the appropriate content to the writer.
+Otherwise, you must use tools to retrieve information or images needed for the task.
+
+It's normal for the task to include some ambiguity. You must always think carefully about the context of the user's request to understand what are the main content needs to be retrieved.
+Example:
+    Request: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+    ->Though: The main content is "history of the internet", while "write in English and publish in PDF format" is a requirement for other agents.
+    Your task: Look for information in English about the history of the Internet.
+    This is not your task: Create a blog post or look for how to create a PDF.
+
+    Next request: "Publish the blog post in HTML format."
+    ->Though: User just asking for a format change, the previous content is still valid.
+    Your task: Return the previous content of the post to the writer. No need to do any research.
+    This is not your task: Look for how to create an HTML file.
+
+If you use the tools but don't find any related information, please return "I didn't find any new information for {the topic}." along with the content you found. Don't try to make up information yourself.
+If the request doesn't need any new information because it was in the conversation history, please return "The task doesn't need any new information. Please reuse the existing content in the conversation history.
+`,
     chatHistory,
   });
 };
@@ -35,8 +45,19 @@ export const createResearcher = async (chatHistory: ChatMessage[]) => {
 export const createWriter = (chatHistory: ChatMessage[]) => {
   return new FunctionCallingAgent({
     name: "writer",
-    systemPrompt:
-      "You are an expert in writing blog posts. You are given a task to write a blog post. Don't make up any information yourself.",
+    systemPrompt: `You are an expert in writing blog posts.
+You are given the task of writing a blog post based on research content provided by the researcher agent. Do not invent any information yourself. 
+It's important to read the entire conversation history to write the blog post accurately.
+If you receive a review from the reviewer, update the post according to the feedback and return the new post content.
+If the content is not valid (e.g., broken link, broken image, etc.), do not use it.
+It's normal for the task to include some ambiguity, so you must define the user's initial request to write the post correctly.
+If you update the post based on the reviewer's feedback, first explain what changes you made to the post, then provide the new post content. Do not include the reviewer's comments.
+Example:
+    Task: "Here is the information I found about the history of the internet: 
+    Create a blog post about the history of the internet, write in English, and publish in PDF format."
+    -> Your task: Use the research content {...} to write a blog post in English.
+    -> This is not your task: Create a PDF
+    Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.`,
     chatHistory,
   });
 };
@@ -44,8 +65,34 @@ export const createWriter = (chatHistory: ChatMessage[]) => {
 export const createReviewer = (chatHistory: ChatMessage[]) => {
   return new FunctionCallingAgent({
     name: "reviewer",
-    systemPrompt:
-      "You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. Only if the post is good enough for publishing, then you MUST return 'The post is good.'. In all other cases return your review.",
+    systemPrompt: `You are an expert in reviewing blog posts.
+You are given a task to review a blog post. As a reviewer, it's important that your review aligns with the user's request. Please focus on the user's request when reviewing the post.
+Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement.
+Furthermore, proofread the post for grammar and spelling errors.
+Only if the post is good enough for publishing should you return 'The post is good.' In all other cases, return your review.
+It's normal for the task to include some ambiguity, so you must define the user's initial request to review the post correctly.
+Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+Example:
+    Task: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+    -> Your task: Review whether the main content of the post is about the history of the internet and if it is written in English.
+    -> This is not your task: Create blog post, create PDF, write in English.`,
+    chatHistory,
+  });
+};
+
+export const createPublisher = async (chatHistory: ChatMessage[]) => {
+  const tools = await lookupTools(["document_generator"]);
+  let systemPrompt = `You are an expert in publishing blog posts. You are given a task to publish a blog post. 
+If the writer says that there was an error, you should reply with the error and not publish the post.`;
+  if (tools.length > 0) {
+    systemPrompt = `${systemPrompt}. 
+If the user requests to generate a file, use the document_generator tool to generate the file and reply with the link to the file.
+Otherwise, simply return the content of the post.`;
+  }
+  return new FunctionCallingAgent({
+    name: "publisher",
+    tools: tools,
+    systemPrompt: systemPrompt,
     chatHistory,
   });
 };

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

@@ -5,8 +5,15 @@ import {
   Workflow,
   WorkflowEvent,
 } from "@llamaindex/core/workflow";
-import { ChatMessage, ChatResponseChunk } from "llamaindex";
-import { createResearcher, createReviewer, createWriter } from "./agents";
+import { Message } from "ai";
+import { ChatMessage, ChatResponseChunk, Settings } from "llamaindex";
+import { getAnnotations } from "../llamaindex/streaming/annotations";
+import {
+  createPublisher,
+  createResearcher,
+  createReviewer,
+  createWriter,
+} from "./agents";
 import { AgentInput, AgentRunEvent } from "./type";
 
 const TIMEOUT = 360 * 1000;
@@ -18,8 +25,45 @@ class WriteEvent extends WorkflowEvent<{
   isGood: boolean;
 }> {}
 class ReviewEvent extends WorkflowEvent<{ input: string }> {}
+class PublishEvent extends WorkflowEvent<{ input: string }> {}
 
-export const createWorkflow = (chatHistory: ChatMessage[]) => {
+const prepareChatHistory = (chatHistory: Message[]): ChatMessage[] => {
+  // By default, the chat history only contains the assistant and user messages
+  // all the agents messages are stored in annotation data which is not visible to the LLM
+
+  const MAX_AGENT_MESSAGES = 10;
+  const agentAnnotations = getAnnotations<{ agent: string; text: string }>(
+    chatHistory,
+    { role: "assistant", type: "agent" },
+  ).slice(-MAX_AGENT_MESSAGES);
+
+  const agentMessages = agentAnnotations
+    .map(
+      (annotation) =>
+        `\n<${annotation.data.agent}>\n${annotation.data.text}\n</${annotation.data.agent}>`,
+    )
+    .join("\n");
+
+  const agentContent = agentMessages
+    ? "Here is the previous conversation of agents:\n" + agentMessages
+    : "";
+
+  if (agentContent) {
+    const agentMessage: ChatMessage = {
+      role: "assistant",
+      content: agentContent,
+    };
+    return [
+      ...chatHistory.slice(0, -1),
+      agentMessage,
+      chatHistory.slice(-1)[0],
+    ] as ChatMessage[];
+  }
+  return chatHistory as ChatMessage[];
+};
+
+export const createWorkflow = (messages: Message[], params?: any) => {
+  const chatHistoryWithAgentMessages = prepareChatHistory(messages);
   const runAgent = async (
     context: Context,
     agent: Workflow,
@@ -36,13 +80,51 @@ export const createWorkflow = (chatHistory: ChatMessage[]) => {
 
   const start = async (context: Context, ev: StartEvent) => {
     context.set("task", ev.data.input);
-    return new ResearchEvent({
-      input: `Research for this task: ${ev.data.input}`,
-    });
+
+    const chatHistoryStr = chatHistoryWithAgentMessages
+      .map((msg) => `${msg.role}: ${msg.content}`)
+      .join("\n");
+
+    // Decision-making process
+    const decision = await decideWorkflow(ev.data.input, chatHistoryStr);
+
+    if (decision !== "publish") {
+      return new ResearchEvent({
+        input: `Research for this task: ${ev.data.input}`,
+      });
+    } else {
+      return new PublishEvent({
+        input: `Publish content based on the chat history\n${chatHistoryStr}\n\n and task: ${ev.data.input}`,
+      });
+    }
+  };
+
+  const decideWorkflow = async (task: string, chatHistoryStr: string) => {
+    const llm = Settings.llm;
+
+    const prompt = `You are an expert in decision-making, helping people write and publish blog posts.
+If the user is asking for a file or to publish content, respond with 'publish'.
+If the user requests to write or update a blog post, respond with 'not_publish'.
+
+Here is the chat history:
+${chatHistoryStr}
+
+The current user request is:
+${task}
+
+Given the chat history and the new user request, decide whether to publish based on existing information.
+Decision (respond with either 'not_publish' or 'publish'):`;
+
+    const output = await llm.complete({ prompt: prompt });
+    const decision = output.text.trim().toLowerCase();
+    return decision === "publish" ? "publish" : "research";
   };
 
   const research = async (context: Context, ev: ResearchEvent) => {
-    const researcher = await createResearcher(chatHistory);
+    const researcher = await createResearcher(
+      chatHistoryWithAgentMessages,
+      params,
+    );
     const researchRes = await runAgent(context, researcher, {
       message: ev.data.input,
     });
@@ -54,6 +136,8 @@ export const createWorkflow = (chatHistory: ChatMessage[]) => {
   };
 
   const write = async (context: Context, ev: WriteEvent) => {
+    const writer = createWriter(chatHistoryWithAgentMessages);
+
     context.set("attempts", context.get("attempts", 0) + 1);
     const tooManyAttempts = context.get("attempts") > MAX_ATTEMPTS;
     if (tooManyAttempts) {
@@ -66,17 +150,15 @@ export const createWorkflow = (chatHistory: ChatMessage[]) => {
     }
 
     if (ev.data.isGood || tooManyAttempts) {
-      // The text is ready for publication, we just use the writer to stream the output
-      const writer = createWriter(chatHistory);
-      const content = context.get("result");
-
-      return (await runAgent(context, writer, {
-        message: `You're blog post is ready for publication. Please respond with just the blog post. Blog post: \`\`\`${content}\`\`\``,
+      // the blog post is good or too many attempts
+      // stream the final content
+      const result = await runAgent(context, writer, {
+        message: `Based on the reviewer's feedback, refine the post and return only the final version of the post. Here's the current version: ${ev.data.input}`,
         streaming: true,
-      })) as unknown as StopEvent<AsyncGenerator<ChatResponseChunk>>;
+      });
+      return result as unknown as StopEvent<AsyncGenerator<ChatResponseChunk>>;
     }
 
-    const writer = createWriter(chatHistory);
     const writeRes = await runAgent(context, writer, {
       message: ev.data.input,
     });
@@ -86,7 +168,7 @@ export const createWorkflow = (chatHistory: ChatMessage[]) => {
   };
 
   const review = async (context: Context, ev: ReviewEvent) => {
-    const reviewer = createReviewer(chatHistory);
+    const reviewer = createReviewer(chatHistoryWithAgentMessages);
     const reviewRes = await reviewer.run(
       new StartEvent<AgentInput>({ input: { message: ev.data.input } }),
     );
@@ -123,11 +205,26 @@ export const createWorkflow = (chatHistory: ChatMessage[]) => {
     });
   };
 
+  const publish = async (context: Context, ev: PublishEvent) => {
+    const publisher = await createPublisher(chatHistoryWithAgentMessages);
+
+    const publishResult = await runAgent(context, publisher, {
+      message: `${ev.data.input}`,
+      streaming: true,
+    });
+    return publishResult as unknown as StopEvent<
+      AsyncGenerator<ChatResponseChunk>
+    >;
+  };
+
   const workflow = new Workflow({ timeout: TIMEOUT, validate: true });
-  workflow.addStep(StartEvent, start, { outputs: ResearchEvent });
+  workflow.addStep(StartEvent, start, {
+    outputs: [ResearchEvent, PublishEvent],
+  });
   workflow.addStep(ResearchEvent, research, { outputs: WriteEvent });
   workflow.addStep(WriteEvent, write, { outputs: [ReviewEvent, StopEvent] });
   workflow.addStep(ReviewEvent, review, { outputs: WriteEvent });
+  workflow.addStep(PublishEvent, publish, { outputs: StopEvent });
 
   return workflow;
 };

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

@@ -143,7 +143,7 @@ export class FunctionCallingAgent extends Workflow {
         fullResponse = chunk;
       }
 
-      if (fullResponse) {
+      if (fullResponse?.options && Object.keys(fullResponse.options).length) {
         memory.put({
           role: "assistant",
           content: "",

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

@@ -0,0 +1,54 @@
+import fs from "fs/promises";
+import { BaseToolWithCall, QueryEngineTool } from "llamaindex";
+import path from "path";
+import { getDataSource } from "../engine";
+import { createTools } from "../engine/tools/index";
+
+export const getQueryEngineTool = async (
+  params?: any,
+): Promise<QueryEngineTool | null> => {
+  const index = await getDataSource(params);
+  if (!index) {
+    return null;
+  }
+
+  const topK = process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined;
+  return new QueryEngineTool({
+    queryEngine: index.asQueryEngine({
+      similarityTopK: topK,
+    }),
+    metadata: {
+      name: "query_index",
+      description: `Use this tool to retrieve information about the text corpus from the index.`,
+    },
+  });
+};
+
+export const getAvailableTools = async () => {
+  const configFile = path.join("config", "tools.json");
+  let toolConfig: any;
+  const tools: BaseToolWithCall[] = [];
+  try {
+    toolConfig = JSON.parse(await fs.readFile(configFile, "utf8"));
+  } catch (e) {
+    console.info(`Could not read ${configFile} file. Using no tools.`);
+  }
+  if (toolConfig) {
+    tools.push(...(await createTools(toolConfig)));
+  }
+  const queryEngineTool = await getQueryEngineTool();
+  if (queryEngineTool) {
+    tools.push(queryEngineTool);
+  }
+
+  return tools;
+};
+
+export const lookupTools = async (
+  toolNames: string[],
+): Promise<BaseToolWithCall[]> => {
+  const availableTools = await getAvailableTools();
+  return availableTools.filter((tool) =>
+    toolNames.includes(tool.metadata.name),
+  );
+};

templates/components/routers/python/sandbox.py

@@ -0,0 +1,158 @@
+# Copyright 2024 FoundryLabs, Inc. and LlamaIndex, Inc.
+# Portions of this file are copied from the e2b project (https://github.com/e2b-dev/ai-artifacts) and then converted to Python
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import base64
+import logging
+import os
+import uuid
+from typing import Dict, List, Optional, Union
+
+from app.engine.tools.artifact import CodeArtifact
+from app.engine.utils.file_helper import save_file
+from e2b_code_interpreter import CodeInterpreter, Sandbox
+from fastapi import APIRouter, HTTPException, Request
+from pydantic import BaseModel
+
+logger = logging.getLogger("uvicorn")
+
+sandbox_router = APIRouter()
+
+SANDBOX_TIMEOUT = 10 * 60  # timeout in seconds
+MAX_DURATION = 60  # max duration in seconds
+
+
+class ExecutionResult(BaseModel):
+    template: str
+    stdout: List[str]
+    stderr: List[str]
+    runtime_error: Optional[Dict[str, Union[str, List[str]]]] = None
+    output_urls: List[Dict[str, str]]
+    url: Optional[str]
+
+    def to_response(self):
+        """
+        Convert the execution result to a response object (camelCase)
+        """
+        return {
+            "template": self.template,
+            "stdout": self.stdout,
+            "stderr": self.stderr,
+            "runtimeError": self.runtime_error,
+            "outputUrls": self.output_urls,
+            "url": self.url,
+        }
+
+
+@sandbox_router.post("")
+async def create_sandbox(request: Request):
+    request_data = await request.json()
+
+    try:
+        artifact = CodeArtifact(**request_data["artifact"])
+    except Exception:
+        logger.error(f"Could not create artifact from request data: {request_data}")
+        return HTTPException(
+            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}")
+
+    # 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}"
+            )
+
+    # Copy code to disk
+    if isinstance(artifact.code, list):
+        for file in artifact.code:
+            sbx.files.write(file.file_path, file.file_content)
+            logger.debug(f"Copied file to {file.file_path}")
+    else:
+        sbx.files.write(artifact.file_path, artifact.code)
+        logger.debug(f"Copied file to {artifact.file_path}")
+
+    # Execute code or return a URL to the running sandbox
+    if artifact.template == "code-interpreter-multilang":
+        result = sbx.notebook.exec_cell(artifact.code or "")
+        output_urls = _download_cell_results(result.results)
+        return ExecutionResult(
+            template=artifact.template,
+            stdout=result.logs.stdout,
+            stderr=result.logs.stderr,
+            runtime_error=result.error,
+            output_urls=output_urls,
+            url=None,
+        ).to_response()
+    else:
+        return ExecutionResult(
+            template=artifact.template,
+            stdout=[],
+            stderr=[],
+            runtime_error=None,
+            output_urls=[],
+            url=f"https://{sbx.get_host(artifact.port or 80)}",
+        ).to_response()
+
+
+def _download_cell_results(cell_results: Optional[List]) -> List[Dict[str, str]]:
+    """
+    To pull results from code interpreter cell and save them to disk for serving
+    """
+    if not cell_results:
+        return []
+
+    output = []
+    for result in cell_results:
+        try:
+            formats = result.formats()
+            for ext in formats:
+                data = result[ext]
+
+                if ext in ["png", "svg", "jpeg", "pdf"]:
+                    file_path = f"output/tools/{uuid.uuid4()}.{ext}"
+                    base64_data = data
+                    buffer = base64.b64decode(base64_data)
+                    file_meta = save_file(content=buffer, file_path=file_path)
+                    output.append(
+                        {
+                            "type": ext,
+                            "filename": file_meta.filename,
+                            "url": file_meta.url,
+                        }
+                    )
+        except Exception as e:
+            logger.error(f"Error processing result: {str(e)}")
+
+    return output

templates/components/settings/python/llmhub.py

@@ -1,7 +1,11 @@
-from llama_index.embeddings.openai import OpenAIEmbedding
-from llama_index.core.settings import Settings
-from typing import Dict
+import logging
 import os
+from typing import Dict
+
+from llama_index.core.settings import Settings
+from llama_index.embeddings.openai import OpenAIEmbedding
+
+logger = logging.getLogger(__name__)
 
 DEFAULT_MODEL = "gpt-3.5-turbo"
 DEFAULT_EMBEDDING_MODEL = "text-embedding-3-large"
@@ -50,7 +54,11 @@ def embedding_config_from_env() -> Dict:
 
 
 def init_llmhub():
-    from llama_index.llms.openai_like import OpenAILike
+    try:
+        from llama_index.llms.openai_like import OpenAILike
+    except ImportError:
+        logger.error("Failed to import OpenAILike. Make sure llama_index is installed.")
+        raise
 
     llm_configs = llm_config_from_env()
     embedding_configs = embedding_config_from_env()

templates/components/settings/python/settings.py

@@ -33,8 +33,13 @@ def init_settings():
 
 
 def init_ollama():
-    from llama_index.embeddings.ollama import OllamaEmbedding
-    from llama_index.llms.ollama.base import DEFAULT_REQUEST_TIMEOUT, Ollama
+    try:
+        from llama_index.embeddings.ollama import OllamaEmbedding
+        from llama_index.llms.ollama.base import DEFAULT_REQUEST_TIMEOUT, Ollama
+    except ImportError:
+        raise ImportError(
+            "Ollama support is not installed. Please install it with `poetry add llama-index-llms-ollama` and `poetry add llama-index-embeddings-ollama`"
+        )
 
     base_url = os.getenv("OLLAMA_BASE_URL") or "http://127.0.0.1:11434"
     request_timeout = float(
@@ -55,25 +60,29 @@ def init_openai():
     from llama_index.llms.openai import OpenAI
 
     max_tokens = os.getenv("LLM_MAX_TOKENS")
-    config = {
-        "model": os.getenv("MODEL"),
-        "temperature": float(os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)),
-        "max_tokens": int(max_tokens) if max_tokens is not None else None,
-    }
-    Settings.llm = OpenAI(**config)
+    Settings.llm = OpenAI(
+        model=os.getenv("MODEL", "gpt-4o-mini"),
+        temperature=float(os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)),
+        max_tokens=int(max_tokens) if max_tokens is not None else None,
+    )
 
     dimensions = os.getenv("EMBEDDING_DIM")
-    config = {
-        "model": os.getenv("EMBEDDING_MODEL"),
-        "dimensions": int(dimensions) if dimensions is not None else None,
-    }
-    Settings.embed_model = OpenAIEmbedding(**config)
+    Settings.embed_model = OpenAIEmbedding(
+        model=os.getenv("EMBEDDING_MODEL", "text-embedding-3-small"),
+        dimensions=int(dimensions) if dimensions is not None else None,
+    )
 
 
 def init_azure_openai():
     from llama_index.core.constants import DEFAULT_TEMPERATURE
-    from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
-    from llama_index.llms.azure_openai import AzureOpenAI
+
+    try:
+        from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
+        from llama_index.llms.azure_openai import AzureOpenAI
+    except ImportError:
+        raise ImportError(
+            "Azure OpenAI support is not installed. Please install it with `poetry add llama-index-llms-azure-openai` and `poetry add llama-index-embeddings-azure-openai`"
+        )
 
     llm_deployment = os.environ["AZURE_OPENAI_LLM_DEPLOYMENT"]
     embedding_deployment = os.environ["AZURE_OPENAI_EMBEDDING_DEPLOYMENT"]
@@ -105,26 +114,37 @@ def init_azure_openai():
 
 
 def init_fastembed():
-    """
-    Use Qdrant Fastembed as the local embedding provider.
-    """
-    from llama_index.embeddings.fastembed import FastEmbedEmbedding
+    try:
+        from llama_index.embeddings.fastembed import FastEmbedEmbedding
+    except ImportError:
+        raise ImportError(
+            "FastEmbed support is not installed. Please install it with `poetry add llama-index-embeddings-fastembed`"
+        )
 
     embed_model_map: Dict[str, str] = {
         # Small and multilingual
         "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
         # Large and multilingual
-        "paraphrase-multilingual-mpnet-base-v2": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",  # noqa: E501
+        "paraphrase-multilingual-mpnet-base-v2": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
     }
 
+    embedding_model = os.getenv("EMBEDDING_MODEL")
+    if embedding_model is None:
+        raise ValueError("EMBEDDING_MODEL environment variable is not set")
+
     # This will download the model automatically if it is not already downloaded
     Settings.embed_model = FastEmbedEmbedding(
-        model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
+        model_name=embed_model_map[embedding_model]
     )
 
 
 def init_groq():
-    from llama_index.llms.groq import Groq
+    try:
+        from llama_index.llms.groq import Groq
+    except ImportError:
+        raise ImportError(
+            "Groq support is not installed. Please install it with `poetry add llama-index-llms-groq`"
+        )
 
     Settings.llm = Groq(model=os.getenv("MODEL"))
     # Groq does not provide embeddings, so we use FastEmbed instead
@@ -132,7 +152,12 @@ def init_groq():
 
 
 def init_anthropic():
-    from llama_index.llms.anthropic import Anthropic
+    try:
+        from llama_index.llms.anthropic import Anthropic
+    except ImportError:
+        raise ImportError(
+            "Anthropic support is not installed. Please install it with `poetry add llama-index-llms-anthropic`"
+        )
 
     model_map: Dict[str, str] = {
         "claude-3-opus": "claude-3-opus-20240229",
@@ -148,8 +173,13 @@ def init_anthropic():
 
 
 def init_gemini():
-    from llama_index.embeddings.gemini import GeminiEmbedding
-    from llama_index.llms.gemini import Gemini
+    try:
+        from llama_index.embeddings.gemini import GeminiEmbedding
+        from llama_index.llms.gemini import Gemini
+    except ImportError:
+        raise ImportError(
+            "Gemini support is not installed. Please install it with `poetry add llama-index-llms-gemini` and `poetry add llama-index-embeddings-gemini`"
+        )
 
     model_name = f"models/{os.getenv('MODEL')}"
     embed_model_name = f"models/{os.getenv('EMBEDDING_MODEL')}"

templates/components/vectordbs/python/astra/vectordb.py

@@ -15,6 +15,6 @@ def get_vector_store():
         token=token,
         api_endpoint=endpoint,
         collection_name=collection,
-        embedding_dimension=int(os.getenv("EMBEDDING_DIM")),
+        embedding_dimension=int(os.getenv("EMBEDDING_DIM", 768)),
     )
     return store

templates/components/vectordbs/python/chroma/vectordb.py

@@ -1,4 +1,5 @@
 import os
+
 from llama_index.vector_stores.chroma import ChromaVectorStore
 
 
@@ -18,7 +19,7 @@ def get_vector_store():
             )
         store = ChromaVectorStore.from_params(
             host=os.getenv("CHROMA_HOST"),
-            port=int(os.getenv("CHROMA_PORT")),
+            port=os.getenv("CHROMA_PORT", "8001"),
             collection_name=collection_name,
         )
     return store

templates/components/vectordbs/python/llamacloud/generate.py

@@ -1,22 +1,66 @@
 # flake8: noqa: E402
-from dotenv import load_dotenv
+import os
 
-from app.engine.index import get_index
+from dotenv import load_dotenv
 
 load_dotenv()
 
 import logging
-from llama_index.core.readers import SimpleDirectoryReader
+
+from app.engine.index import get_client, get_index
 from app.engine.service import LLamaCloudFileService
+from app.settings import init_settings
+from llama_cloud import PipelineType
+from llama_index.core.readers import SimpleDirectoryReader
+from llama_index.core.settings import Settings
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger()
 
 
+def ensure_index(index):
+    project_id = index._get_project_id()
+    client = get_client()
+    pipelines = client.pipelines.search_pipelines(
+        project_id=project_id,
+        pipeline_name=index.name,
+        pipeline_type=PipelineType.MANAGED.value,
+    )
+    if len(pipelines) == 0:
+        from llama_index.embeddings.openai import OpenAIEmbedding
+
+        if not isinstance(Settings.embed_model, OpenAIEmbedding):
+            raise ValueError(
+                "Creating a new pipeline with a non-OpenAI embedding model is not supported."
+            )
+        client.pipelines.upsert_pipeline(
+            project_id=project_id,
+            request={
+                "name": index.name,
+                "embedding_config": {
+                    "type": "OPENAI_EMBEDDING",
+                    "component": {
+                        "api_key": os.getenv("OPENAI_API_KEY"),  # editable
+                        "model_name": os.getenv("EMBEDDING_MODEL"),
+                    },
+                },
+                "transform_config": {
+                    "mode": "auto",
+                    "config": {
+                        "chunk_size": Settings.chunk_size,  # editable
+                        "chunk_overlap": Settings.chunk_overlap,  # editable
+                    },
+                },
+            },
+        )
+
+
 def generate_datasource():
+    init_settings()
     logger.info("Generate index for the provided data")
 
     index = get_index()
+    ensure_index(index)
     project_id = index._get_project_id()
     pipeline_id = index._get_pipeline_id()
 
@@ -34,13 +78,7 @@ def generate_datasource():
                 f"Adding file {input_file} to pipeline {index.name} in project {index.project_name}"
             )
             LLamaCloudFileService.add_file_to_pipeline(
-                project_id,
-                pipeline_id,
-                f,
-                custom_metadata={
-                    # Set private=false to mark the document as public (required for filtering)
-                    "private": "false",
-                },
+                project_id, pipeline_id, f, custom_metadata={}
             )
 
     logger.info("Finished generating the index")

templates/components/vectordbs/python/llamacloud/index.py

@@ -7,7 +7,7 @@ from llama_index.core.ingestion.api_utils import (
     get_client as llama_cloud_get_client,
 )
 from llama_index.indices.managed.llama_cloud import LlamaCloudIndex
-from pydantic import BaseModel, Field, validator
+from pydantic import BaseModel, Field, field_validator
 
 logger = logging.getLogger("uvicorn")
 
@@ -15,31 +15,39 @@ logger = logging.getLogger("uvicorn")
 class LlamaCloudConfig(BaseModel):
     # Private attributes
     api_key: str = Field(
-        default=os.getenv("LLAMA_CLOUD_API_KEY"),
         exclude=True,  # Exclude from the model representation
     )
     base_url: Optional[str] = Field(
-        default=os.getenv("LLAMA_CLOUD_BASE_URL"),
         exclude=True,
     )
     organization_id: Optional[str] = Field(
-        default=os.getenv("LLAMA_CLOUD_ORGANIZATION_ID"),
         exclude=True,
     )
     # Configuration attributes, can be set by the user
     pipeline: str = Field(
         description="The name of the pipeline to use",
-        default=os.getenv("LLAMA_CLOUD_INDEX_NAME"),
     )
     project: str = Field(
         description="The name of the LlamaCloud project",
-        default=os.getenv("LLAMA_CLOUD_PROJECT_NAME"),
     )
 
+    def __init__(self, **kwargs):
+        if "api_key" not in kwargs:
+            kwargs["api_key"] = os.getenv("LLAMA_CLOUD_API_KEY")
+        if "base_url" not in kwargs:
+            kwargs["base_url"] = os.getenv("LLAMA_CLOUD_BASE_URL")
+        if "organization_id" not in kwargs:
+            kwargs["organization_id"] = os.getenv("LLAMA_CLOUD_ORGANIZATION_ID")
+        if "pipeline" not in kwargs:
+            kwargs["pipeline"] = os.getenv("LLAMA_CLOUD_INDEX_NAME")
+        if "project" not in kwargs:
+            kwargs["project"] = os.getenv("LLAMA_CLOUD_PROJECT_NAME")
+        super().__init__(**kwargs)
+
     # Validate and throw error if the env variables are not set before starting the app
-    @validator("pipeline", "project", "api_key", pre=True, always=True)
+    @field_validator("pipeline", "project", "api_key", mode="before")
     @classmethod
-    def validate_env_vars(cls, value):
+    def validate_fields(cls, value):
         if value is None:
             raise ValueError(
                 "Please set LLAMA_CLOUD_INDEX_NAME, LLAMA_CLOUD_PROJECT_NAME and LLAMA_CLOUD_API_KEY"
@@ -56,7 +64,7 @@ class LlamaCloudConfig(BaseModel):
 
 class IndexConfig(BaseModel):
     llama_cloud_pipeline_config: LlamaCloudConfig = Field(
-        default=LlamaCloudConfig(),
+        default_factory=lambda: LlamaCloudConfig(),
         alias="llamaCloudPipeline",
     )
     callback_manager: Optional[CallbackManager] = Field(

templates/components/vectordbs/python/llamacloud/query_filter.py

@@ -5,7 +5,7 @@ def generate_filters(doc_ids):
     """
     Generate public/private document filters based on the doc_ids and the vector store.
     """
-    # Using "is_empty" filter to include the documents don't have the "private" key because they're uploaded in LlamaCloud UI
+    # public documents (ingested by "poetry run generate" or in the LlamaCloud UI) don't have the "private" field
     public_doc_filter = MetadataFilter(
         key="private",
         value=None,

templates/components/vectordbs/python/milvus/vectordb.py

@@ -1,4 +1,5 @@
 import os
+
 from llama_index.vector_stores.milvus import MilvusVectorStore
 
 
@@ -15,6 +16,6 @@ def get_vector_store():
         user=os.getenv("MILVUS_USERNAME"),
         password=os.getenv("MILVUS_PASSWORD"),
         collection_name=collection,
-        dim=int(os.getenv("EMBEDDING_DIM")),
+        dim=int(os.getenv("EMBEDDING_DIM", 768)),
     )
     return store

templates/components/vectordbs/python/none/index.py

@@ -3,7 +3,7 @@ import os
 from datetime import timedelta
 from typing import Optional
 
-from cachetools import TTLCache, cached
+from cachetools import TTLCache, cached  # type: ignore
 from llama_index.core.callbacks import CallbackManager
 from llama_index.core.indices import load_index_from_storage
 from llama_index.core.storage import StorageContext

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

@@ -25,6 +25,8 @@ async function* walk(dir: string): AsyncGenerator<string> {
 
 async function loadAndIndex() {
   const index = await getDataSource();
+  // ensure the index is available or create a new one
+  await index.ensureIndex();
   const projectId = await index.getProjectId();
   const pipelineId = await index.getPipelineId();
 
@@ -32,10 +34,23 @@ async function loadAndIndex() {
   for await (const filePath of walk(DATA_DIR)) {
     const buffer = await fs.readFile(filePath);
     const filename = path.basename(filePath);
-    const file = new File([buffer], filename);
-    await LLamaCloudFileService.addFileToPipeline(projectId, pipelineId, file, {
-      private: "false",
-    });
+    try {
+      await LLamaCloudFileService.addFileToPipeline(
+        projectId,
+        pipelineId,
+        new File([buffer], filename),
+      );
+    } catch (error) {
+      if (
+        error instanceof ReferenceError &&
+        error.message.includes("File is not defined")
+      ) {
+        throw new Error(
+          "File class is not supported in the current Node.js version. Please use Node.js 20 or higher.",
+        );
+      }
+      throw error;
+    }
   }
 
   console.log(`Successfully uploaded documents to LlamaCloud!`);

templates/components/vectordbs/typescript/llamacloud/queryFilter.ts

@@ -1,14 +1,17 @@
-import { MetadataFilter, MetadataFilters } from "llamaindex";
+import { CloudRetrieveParams, MetadataFilter } from "llamaindex";
 
-export function generateFilters(documentIds: string[]): MetadataFilters {
-  // public documents don't have the "private" field or it's set to "false"
+export function generateFilters(documentIds: string[]) {
+  // public documents (ingested by "npm run generate" or in the LlamaCloud UI) don't have the "private" field
   const publicDocumentsFilter: MetadataFilter = {
     key: "private",
     operator: "is_empty",
   };
 
   // if no documentIds are provided, only retrieve information from public documents
-  if (!documentIds.length) return { filters: [publicDocumentsFilter] };
+  if (!documentIds.length)
+    return {
+      filters: [publicDocumentsFilter],
+    } as CloudRetrieveParams["filters"];
 
   const privateDocumentsFilter: MetadataFilter = {
     key: "file_id", // Note: LLamaCloud uses "file_id" to reference private document ids as "doc_id" is a restricted field in LlamaCloud
@@ -20,5 +23,5 @@ export function generateFilters(documentIds: string[]): MetadataFilters {
   return {
     filters: [publicDocumentsFilter, privateDocumentsFilter],
     condition: "or",
-  };
+  } as CloudRetrieveParams["filters"];
 }

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

@@ -18,7 +18,9 @@ async function loadAndIndex() {
 
   // create postgres vector store
   const vectorStore = new PGVectorStore({
-    connectionString: process.env.PG_CONNECTION_STRING,
+    clientConfig: {
+      connectionString: process.env.PG_CONNECTION_STRING,
+    },
     schemaName: PGVECTOR_SCHEMA,
     tableName: PGVECTOR_TABLE,
   });

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

@@ -9,7 +9,9 @@ import {
 export async function getDataSource(params?: any) {
   checkRequiredEnvVars();
   const pgvs = new PGVectorStore({
-    connectionString: process.env.PG_CONNECTION_STRING,
+    clientConfig: {
+      connectionString: process.env.PG_CONNECTION_STRING,
+    },
     schemaName: PGVECTOR_SCHEMA,
     tableName: PGVECTOR_TABLE,
   });

templates/types/extractor/fastapi/app/engine/generate.py

@@ -6,7 +6,7 @@ load_dotenv()
 import logging
 import os
 
-from llama_index.core.ingestion import IngestionPipeline
+from llama_index.core.ingestion import DocstoreStrategy, IngestionPipeline
 from llama_index.core.node_parser import SentenceSplitter
 from llama_index.core.settings import Settings
 from llama_index.core.storage import StorageContext
@@ -41,7 +41,7 @@ def run_pipeline(docstore, vector_store, documents):
             Settings.embed_model,
         ],
         docstore=docstore,
-        docstore_strategy="upserts_and_delete",
+        docstore_strategy=DocstoreStrategy.UPSERTS_AND_DELETE,  # type: ignore
         vector_store=vector_store,
     )
 

templates/types/extractor/fastapi/app/engine/index.py

@@ -16,7 +16,7 @@ class IndexConfig(BaseModel):
     )
 
 
-def get_index(config: IndexConfig = None):
+def get_index(config: Optional[IndexConfig] = None) -> VectorStoreIndex:
     if config is None:
         config = IndexConfig()
     logger.info("Connecting vector store...")

templates/types/multiagent/fastapi/app/api/routers/chat.py

@@ -1,39 +0,0 @@
-import logging
-
-from app.api.routers.models import (
-    ChatData,
-)
-from app.api.routers.vercel_response import VercelStreamResponse
-from app.examples.factory import create_agent
-from fastapi import APIRouter, HTTPException, Request, status
-from llama_index.core.workflow import Workflow
-
-chat_router = r = APIRouter()
-
-logger = logging.getLogger("uvicorn")
-
-
-@r.post("")
-async def chat(
-    request: Request,
-    data: ChatData,
-):
-    try:
-        last_message_content = data.get_last_message_content()
-        messages = data.get_history_messages()
-        # TODO: generate filters based on doc_ids
-        # for now just use all documents
-        # doc_ids = data.get_chat_document_ids()
-        # TODO: use params
-        # params = data.data or {}
-
-        agent: Workflow = create_agent(chat_history=messages)
-        handler = agent.run(input=last_message_content, streaming=True)
-
-        return VercelStreamResponse(request, handler, agent.stream_events, data)
-    except Exception as e:
-        logger.exception("Error in agent", exc_info=True)
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail=f"Error in agent: {e}",
-        ) from e

templates/types/multiagent/fastapi/app/api/routers/chat_config.py

@@ -1,48 +0,0 @@
-import logging
-import os
-
-from fastapi import APIRouter
-
-from app.api.routers.models import ChatConfig
-
-
-config_router = r = APIRouter()
-
-logger = logging.getLogger("uvicorn")
-
-
-@r.get("")
-async def chat_config() -> ChatConfig:
-    starter_questions = None
-    conversation_starters = os.getenv("CONVERSATION_STARTERS")
-    if conversation_starters and conversation_starters.strip():
-        starter_questions = conversation_starters.strip().split("\n")
-    return ChatConfig(starter_questions=starter_questions)
-
-
-try:
-    from app.engine.service import LLamaCloudFileService
-
-    logger.info("LlamaCloud is configured. Adding /config/llamacloud route.")
-
-    @r.get("/llamacloud")
-    async def chat_llama_cloud_config():
-        projects = LLamaCloudFileService.get_all_projects_with_pipelines()
-        pipeline = os.getenv("LLAMA_CLOUD_INDEX_NAME")
-        project = os.getenv("LLAMA_CLOUD_PROJECT_NAME")
-        pipeline_config = None
-        if pipeline and project:
-            pipeline_config = {
-                "pipeline": pipeline,
-                "project": project,
-            }
-        return {
-            "projects": projects,
-            "pipeline": pipeline_config,
-        }
-
-except ImportError:
-    logger.debug(
-        "LlamaCloud is not configured. Skipping adding /config/llamacloud route."
-    )
-    pass

templates/types/multiagent/fastapi/app/api/routers/models.py

@@ -1,227 +0,0 @@
-import logging
-import os
-from typing import Any, Dict, List, Literal, Optional
-
-from llama_index.core.llms import ChatMessage, MessageRole
-from llama_index.core.schema import NodeWithScore
-from pydantic import BaseModel, Field, validator
-from pydantic.alias_generators import to_camel
-
-from app.config import DATA_DIR
-
-logger = logging.getLogger("uvicorn")
-
-
-class FileContent(BaseModel):
-    type: Literal["text", "ref"]
-    # If the file is pure text then the value is be a string
-    # otherwise, it's a list of document IDs
-    value: str | List[str]
-
-
-class File(BaseModel):
-    id: str
-    content: FileContent
-    filename: str
-    filesize: int
-    filetype: str
-
-
-class AnnotationFileData(BaseModel):
-    files: List[File] = Field(
-        default=[],
-        description="List of files",
-    )
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "csvFiles": [
-                    {
-                        "content": "Name, Age\nAlice, 25\nBob, 30",
-                        "filename": "example.csv",
-                        "filesize": 123,
-                        "id": "123",
-                        "type": "text/csv",
-                    }
-                ]
-            }
-        }
-        alias_generator = to_camel
-
-
-class Annotation(BaseModel):
-    type: str
-    data: AnnotationFileData | List[str]
-
-    def to_content(self) -> str | None:
-        if self.type == "document_file":
-            # We only support generating context content for CSV files for now
-            csv_files = [file for file in self.data.files if file.filetype == "csv"]
-            if len(csv_files) > 0:
-                return "Use data from following CSV raw content\n" + "\n".join(
-                    [f"```csv\n{csv_file.content.value}\n```" for csv_file in csv_files]
-                )
-        else:
-            logger.warning(
-                f"The annotation {self.type} is not supported for generating context content"
-            )
-        return None
-
-
-class Message(BaseModel):
-    role: MessageRole
-    content: str
-    annotations: List[Annotation] | None = None
-
-
-class ChatData(BaseModel):
-    messages: List[Message]
-    data: Any = None
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "messages": [
-                    {
-                        "role": "user",
-                        "content": "What standards for letters exist?",
-                    }
-                ]
-            }
-        }
-
-    @validator("messages")
-    def messages_must_not_be_empty(cls, v):
-        if len(v) == 0:
-            raise ValueError("Messages must not be empty")
-        return v
-
-    def get_last_message_content(self) -> str:
-        """
-        Get the content of the last message along with the data content if available.
-        Fallback to use data content from previous messages
-        """
-        if len(self.messages) == 0:
-            raise ValueError("There is not any message in the chat")
-        last_message = self.messages[-1]
-        message_content = last_message.content
-        for message in reversed(self.messages):
-            if message.role == MessageRole.USER and message.annotations is not None:
-                annotation_contents = filter(
-                    None,
-                    [annotation.to_content() for annotation in message.annotations],
-                )
-                if not annotation_contents:
-                    continue
-                annotation_text = "\n".join(annotation_contents)
-                message_content = f"{message_content}\n{annotation_text}"
-                break
-        return message_content
-
-    def get_history_messages(self) -> List[ChatMessage]:
-        """
-        Get the history messages
-        """
-        return [
-            ChatMessage(role=message.role, content=message.content)
-            for message in self.messages[:-1]
-        ]
-
-    def is_last_message_from_user(self) -> bool:
-        return self.messages[-1].role == MessageRole.USER
-
-    def get_chat_document_ids(self) -> List[str]:
-        """
-        Get the document IDs from the chat messages
-        """
-        document_ids: List[str] = []
-        for message in self.messages:
-            if message.role == MessageRole.USER and message.annotations is not None:
-                for annotation in message.annotations:
-                    if (
-                        annotation.type == "document_file"
-                        and annotation.data.files is not None
-                    ):
-                        for fi in annotation.data.files:
-                            if fi.content.type == "ref":
-                                document_ids += fi.content.value
-        return list(set(document_ids))
-
-
-class SourceNodes(BaseModel):
-    id: str
-    metadata: Dict[str, Any]
-    score: Optional[float]
-    text: str
-    url: Optional[str]
-
-    @classmethod
-    def from_source_node(cls, source_node: NodeWithScore):
-        metadata = source_node.node.metadata
-        url = cls.get_url_from_metadata(metadata)
-
-        return cls(
-            id=source_node.node.node_id,
-            metadata=metadata,
-            score=source_node.score,
-            text=source_node.node.text,  # type: ignore
-            url=url,
-        )
-
-    @classmethod
-    def get_url_from_metadata(cls, metadata: Dict[str, Any]) -> str:
-        url_prefix = os.getenv("FILESERVER_URL_PREFIX")
-        if not url_prefix:
-            logger.warning(
-                "Warning: FILESERVER_URL_PREFIX not set in environment variables. Can't use file server"
-            )
-        file_name = metadata.get("file_name")
-
-        if file_name and url_prefix:
-            # file_name exists and file server is configured
-            pipeline_id = metadata.get("pipeline_id")
-            if pipeline_id:
-                # file is from LlamaCloud
-                file_name = f"{pipeline_id}${file_name}"
-                return f"{url_prefix}/output/llamacloud/{file_name}"
-            is_private = metadata.get("private", "false") == "true"
-            if is_private:
-                # file is a private upload
-                return f"{url_prefix}/output/uploaded/{file_name}"
-            # file is from calling the 'generate' script
-            # Get the relative path of file_path to data_dir
-            file_path = metadata.get("file_path")
-            data_dir = os.path.abspath(DATA_DIR)
-            if file_path and data_dir:
-                relative_path = os.path.relpath(file_path, data_dir)
-                return f"{url_prefix}/data/{relative_path}"
-        # fallback to URL in metadata (e.g. for websites)
-        return metadata.get("URL")
-
-    @classmethod
-    def from_source_nodes(cls, source_nodes: List[NodeWithScore]):
-        return [cls.from_source_node(node) for node in source_nodes]
-
-
-class Result(BaseModel):
-    result: Message
-    nodes: List[SourceNodes]
-
-
-class ChatConfig(BaseModel):
-    starter_questions: Optional[List[str]] = Field(
-        default=None,
-        description="List of starter questions",
-        serialization_alias="starterQuestions",
-    )
-
-    class Config:
-        json_schema_extra = {
-            "example": {
-                "starterQuestions": [
-                    "What standards for letters exist?",
-                    "What are the requirements for a letter to be considered a letter?",
-                ]
-            }
-        }

templates/types/multiagent/fastapi/app/api/routers/upload.py

@@ -1,29 +0,0 @@
-import logging
-from typing import List, Any
-
-from fastapi import APIRouter, HTTPException
-from pydantic import BaseModel
-
-from app.api.services.file import PrivateFileService
-
-file_upload_router = r = APIRouter()
-
-logger = logging.getLogger("uvicorn")
-
-
-class FileUploadRequest(BaseModel):
-    base64: str
-    filename: str
-    params: Any = None
-
-
-@r.post("")
-def upload_file(request: FileUploadRequest) -> List[str]:
-    try:
-        logger.info("Processing file")
-        return PrivateFileService.process_file(
-            request.filename, request.base64, request.params
-        )
-    except Exception as e:
-        logger.error(f"Error processing file: {e}", exc_info=True)
-        raise HTTPException(status_code=500, detail="Error processing file")

templates/types/multiagent/fastapi/app/config.py

@@ -1 +0,0 @@
-DATA_DIR = "data"

templates/types/multiagent/fastapi/app/examples/choreography.py

@@ -1,25 +0,0 @@
-from typing import List, Optional
-from app.agents.single import FunctionCallingAgent
-from app.agents.multi import AgentCallingAgent
-from app.examples.researcher import create_researcher
-from llama_index.core.chat_engine.types import ChatMessage
-
-
-def create_choreography(chat_history: Optional[List[ChatMessage]] = None):
-    researcher = create_researcher(chat_history)
-    reviewer = FunctionCallingAgent(
-        name="reviewer",
-        role="expert in reviewing blog posts",
-        system_prompt="You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. If the post is good, you can say 'The post is good.'",
-        chat_history=chat_history,
-    )
-    return AgentCallingAgent(
-        name="writer",
-        agents=[researcher, reviewer],
-        role="expert in writing blog posts",
-        system_prompt="""You are an expert in writing blog posts. You are given a task to write a blog post. Before starting to write the post, consult the researcher agent to get the information you need. Don't make up any information yourself.
-        After creating a draft for the post, send it to the reviewer agent to receive some feedback and make sure to incorporate the feedback from the reviewer.
-        You can consult the reviewer and researcher maximal two times. Your output should just contain the blog post.""",
-        # TODO: add chat_history support to AgentCallingAgent
-        # chat_history=chat_history,
-    )

templates/types/multiagent/fastapi/app/examples/orchestrator.py

@@ -1,27 +0,0 @@
-from typing import List, Optional
-from app.agents.single import FunctionCallingAgent
-from app.agents.multi import AgentOrchestrator
-from app.examples.researcher import create_researcher
-
-from llama_index.core.chat_engine.types import ChatMessage
-
-
-def create_orchestrator(chat_history: Optional[List[ChatMessage]] = None):
-    researcher = create_researcher(chat_history)
-    writer = FunctionCallingAgent(
-        name="writer",
-        role="expert in writing blog posts",
-        system_prompt="""You are an expert in writing blog posts. You are given a task to write a blog post. Don't make up any information yourself. If you don't have the necessary information to write a blog post, reply "I need information about the topic to write the blog post". If you have all the information needed, write the blog post.""",
-        chat_history=chat_history,
-    )
-    reviewer = FunctionCallingAgent(
-        name="reviewer",
-        role="expert in reviewing blog posts",
-        system_prompt="""You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post and fix the issues found yourself. You must output a final blog post.
-        Especially check for logical inconsistencies and proofread the post for grammar and spelling errors.""",
-        chat_history=chat_history,
-    )
-    return AgentOrchestrator(
-        agents=[writer, reviewer, researcher],
-        refine_plan=False,
-    )

templates/types/multiagent/fastapi/app/examples/researcher.py

@@ -1,39 +0,0 @@
-import os
-from typing import List
-from llama_index.core.tools import QueryEngineTool, ToolMetadata
-from app.agents.single import FunctionCallingAgent
-from app.engine.index import get_index
-
-from llama_index.core.chat_engine.types import ChatMessage
-
-
-def get_query_engine_tool() -> QueryEngineTool:
-    """
-    Provide an agent worker that can be used to query the index.
-    """
-    index = get_index()
-    if index is None:
-        raise ValueError("Index not found. Please create an index first.")
-    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 create_researcher(chat_history: List[ChatMessage]):
-    return FunctionCallingAgent(
-        name="researcher",
-        tools=[get_query_engine_tool()],
-        role="expert in retrieving any unknown content",
-        system_prompt="You are a researcher agent. You are given a researching task. You must use your tools to complete the research.",
-        chat_history=chat_history,
-    )

templates/types/multiagent/fastapi/app/examples/workflow.py

@@ -1,139 +0,0 @@
-from typing import AsyncGenerator, List, Optional
-
-from app.agents.single import AgentRunEvent, AgentRunResult, FunctionCallingAgent
-from app.examples.researcher import create_researcher
-from llama_index.core.chat_engine.types import ChatMessage
-from llama_index.core.workflow import (
-    Context,
-    Event,
-    StartEvent,
-    StopEvent,
-    Workflow,
-    step,
-)
-
-
-def create_workflow(chat_history: Optional[List[ChatMessage]] = None):
-    researcher = create_researcher(
-        chat_history=chat_history,
-    )
-    writer = FunctionCallingAgent(
-        name="writer",
-        role="expert in writing blog posts",
-        system_prompt="""You are an expert in writing blog posts. You are given a task to write a blog post. Don't make up any information yourself.""",
-        chat_history=chat_history,
-    )
-    reviewer = FunctionCallingAgent(
-        name="reviewer",
-        role="expert in reviewing blog posts",
-        system_prompt="You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. Only if the post is good enough for publishing, then you MUST return 'The post is good.'. In all other cases return your review.",
-        chat_history=chat_history,
-    )
-    workflow = BlogPostWorkflow(timeout=360)
-    workflow.add_workflows(researcher=researcher, writer=writer, reviewer=reviewer)
-    return workflow
-
-
-class ResearchEvent(Event):
-    input: str
-
-
-class WriteEvent(Event):
-    input: str
-    is_good: bool = False
-
-
-class ReviewEvent(Event):
-    input: str
-
-
-class BlogPostWorkflow(Workflow):
-    @step()
-    async def start(self, ctx: Context, ev: StartEvent) -> ResearchEvent:
-        # set streaming
-        ctx.data["streaming"] = getattr(ev, "streaming", False)
-        # start the workflow with researching about a topic
-        ctx.data["task"] = ev.input
-        return ResearchEvent(input=f"Research for this task: {ev.input}")
-
-    @step()
-    async def research(
-        self, ctx: Context, ev: ResearchEvent, researcher: FunctionCallingAgent
-    ) -> WriteEvent:
-        result: AgentRunResult = await self.run_agent(ctx, researcher, ev.input)
-        content = result.response.message.content
-        return WriteEvent(
-            input=f"Write a blog post given this task: {ctx.data['task']} using this research content: {content}"
-        )
-
-    @step()
-    async def write(
-        self, ctx: Context, ev: WriteEvent, writer: FunctionCallingAgent
-    ) -> ReviewEvent | StopEvent:
-        MAX_ATTEMPTS = 2
-        ctx.data["attempts"] = ctx.data.get("attempts", 0) + 1
-        too_many_attempts = ctx.data["attempts"] > MAX_ATTEMPTS
-        if too_many_attempts:
-            ctx.write_event_to_stream(
-                AgentRunEvent(
-                    name=writer.name,
-                    msg=f"Too many attempts ({MAX_ATTEMPTS}) to write the blog post. Proceeding with the current version.",
-                )
-            )
-        if ev.is_good or too_many_attempts:
-            # too many attempts or the blog post is good - stream final response if requested
-            result = await self.run_agent(
-                ctx, writer, ev.input, streaming=ctx.data["streaming"]
-            )
-            return StopEvent(result=result)
-        result: AgentRunResult = await self.run_agent(ctx, writer, ev.input)
-        ctx.data["result"] = result
-        return ReviewEvent(input=result.response.message.content)
-
-    @step()
-    async def review(
-        self, ctx: Context, ev: ReviewEvent, reviewer: FunctionCallingAgent
-    ) -> WriteEvent:
-        result: AgentRunResult = await self.run_agent(ctx, reviewer, ev.input)
-        review = result.response.message.content
-        old_content = ctx.data["result"].response.message.content
-        post_is_good = "post is good" in review.lower()
-        ctx.write_event_to_stream(
-            AgentRunEvent(
-                name=reviewer.name,
-                msg=f"The post is {'not ' if not post_is_good else ''}good enough for publishing. Sending back to the writer{' for publication.' if post_is_good else '.'}",
-            )
-        )
-        if post_is_good:
-            return WriteEvent(
-                input=f"You're blog post is ready for publication. Please respond with just the blog post. Blog post: ```{old_content}```",
-                is_good=True,
-            )
-        else:
-            return WriteEvent(
-                input=f"""Improve the writing of a given blog post by using a given review.
-Blog post:
-```
-{old_content}
-``` 
-
-Review: 
-```
-{review}
-```"""
-            )
-
-    async def run_agent(
-        self,
-        ctx: Context,
-        agent: FunctionCallingAgent,
-        input: str,
-        streaming: bool = False,
-    ) -> AgentRunResult | AsyncGenerator:
-        handler = agent.run(input=input, streaming=streaming)
-        # bubble all events while running the executor to the planner
-        async for event in handler.stream_events():
-            # Don't write the StopEvent from sub task to the stream
-            if type(event) is not StopEvent:
-                ctx.write_event_to_stream(event)
-        return await handler

templates/types/multiagent/fastapi/app/observability.py

@@ -1,2 +0,0 @@
-def init_observability():
-    pass

templates/types/multiagent/fastapi/app/utils.py

@@ -1,8 +0,0 @@
-import os
-
-
-def load_from_env(var: str, throw_error: bool = True) -> str:
-    res = os.getenv(var)
-    if res is None and throw_error:
-        raise ValueError(f"Missing environment variable: {var}")
-    return res

templates/types/multiagent/fastapi/gitignore

@@ -1,4 +0,0 @@
-__pycache__
-storage
-.env
-output

templates/types/multiagent/fastapi/main.py

@@ -1,72 +0,0 @@
-# flake8: noqa: E402
-import os
-from dotenv import load_dotenv
-
-from app.config import DATA_DIR
-
-load_dotenv()
-
-import logging
-
-import uvicorn
-from app.api.routers.chat import chat_router
-from app.api.routers.chat_config import config_router
-from app.api.routers.upload import file_upload_router
-from app.observability import init_observability
-from app.settings import init_settings
-from fastapi import FastAPI
-from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import RedirectResponse
-from fastapi.staticfiles import StaticFiles
-
-app = FastAPI()
-
-init_settings()
-init_observability()
-
-
-environment = os.getenv("ENVIRONMENT", "dev")  # Default to 'development' if not set
-logger = logging.getLogger("uvicorn")
-
-if environment == "dev":
-    logger.warning("Running in development mode - allowing CORS for all origins")
-    app.add_middleware(
-        CORSMiddleware,
-        allow_origins=["*"],
-        allow_credentials=True,
-        allow_methods=["*"],
-        allow_headers=["*"],
-    )
-
-    # Redirect to documentation page when accessing base URL
-    @app.get("/")
-    async def redirect_to_docs():
-        return RedirectResponse(url="/docs")
-
-
-def mount_static_files(directory, path):
-    if os.path.exists(directory):
-        logger.info(f"Mounting static files '{directory}' at '{path}'")
-        app.mount(
-            path,
-            StaticFiles(directory=directory, check_dir=False),
-            name=f"{directory}-static",
-        )
-
-
-# Mount the data files to serve the file viewer
-mount_static_files(DATA_DIR, "/api/files/data")
-# Mount the output files from tools
-mount_static_files("output", "/api/files/output")
-
-app.include_router(chat_router, prefix="/api/chat")
-app.include_router(config_router, prefix="/api/chat/config")
-app.include_router(file_upload_router, prefix="/api/chat/upload")
-
-
-if __name__ == "__main__":
-    app_host = os.getenv("APP_HOST", "0.0.0.0")
-    app_port = int(os.getenv("APP_PORT", "8000"))
-    reload = True if environment == "dev" else False
-
-    uvicorn.run(app="main:app", host=app_host, port=app_port, reload=reload)

templates/types/multiagent/fastapi/pyproject.toml

@@ -1,27 +0,0 @@
-[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,<3.13"
-llama-index-agent-openai = ">=0.3.0,<0.4.0"
-llama-index = "0.11.11"
-fastapi = "^0.112.2"
-python-dotenv = "^1.0.0"
-uvicorn = { extras = ["standard"], version = "^0.23.2" }
-cachetools = "^5.3.3"
-aiostream = "^0.5.2"
-
-[tool.poetry.dependencies.docx2txt]
-version = "^0.8"
-
-[build-system]
-requires = ["poetry-core"]
-build-backend = "poetry.core.masonry.api"

templates/types/streaming/express/index.ts

@@ -2,6 +2,7 @@
 import cors from "cors";
 import "dotenv/config";
 import express, { Express, Request, Response } from "express";
+import { sandbox } from "./src/controllers/sandbox.controller";
 import { initObservability } from "./src/observability";
 import chatRouter from "./src/routes/chat.route";
 
@@ -40,6 +41,7 @@ app.get("/", (req: Request, res: Response) => {
 });
 
 app.use("/api/chat", chatRouter);
+app.use("/api/sandbox", sandbox);
 
 app.listen(port, () => {
   console.log(`⚡️[server]: Server is running at http://localhost:${port}`);

templates/types/streaming/express/package.json

@@ -21,13 +21,14 @@
     "dotenv": "^16.3.1",
     "duck-duck-scrape": "^2.2.5",
     "express": "^4.18.2",
-    "llamaindex": "0.6.2",
+    "llamaindex": "0.6.19",
     "pdf2json": "3.0.5",
     "ajv": "^8.12.0",
-    "@e2b/code-interpreter": "^0.0.5",
+    "@e2b/code-interpreter": "0.0.9-beta.3",
     "got": "^14.4.1",
     "@apidevtools/swagger-parser": "^10.1.0",
-    "formdata-node": "^6.0.3"
+    "formdata-node": "^6.0.3",
+    "marked": "^14.1.2"
   },
   "devDependencies": {
     "@types/cors": "^2.8.16",

templates/types/streaming/express/src/controllers/chat-upload.controller.ts

@@ -14,5 +14,11 @@ export const chatUpload = async (req: Request, res: Response) => {
     });
   }
   const index = await getDataSource(params);
+  if (!index) {
+    return res.status(500).json({
+      error:
+        "StorageContext is empty - call 'npm run generate' to generate the storage first",
+    });
+  }
   return res.status(200).json(await uploadDocument(index, filename, base64));
 };

templates/types/streaming/express/src/controllers/chat.controller.ts

@@ -1,64 +1,34 @@
-import {
-  JSONValue,
-  LlamaIndexAdapter,
-  Message,
-  StreamData,
-  streamToResponse,
-} from "ai";
+import { LlamaIndexAdapter, Message, StreamData, streamToResponse } from "ai";
 import { Request, Response } from "express";
 import { ChatMessage, Settings } from "llamaindex";
 import { createChatEngine } from "./engine/chat";
 import {
-  convertMessageContent,
+  isValidMessages,
   retrieveDocumentIds,
+  retrieveMessageContent,
 } from "./llamaindex/streaming/annotations";
-import {
-  createCallbackManager,
-  createStreamTimeout,
-} from "./llamaindex/streaming/events";
+import { createCallbackManager } from "./llamaindex/streaming/events";
 import { generateNextQuestions } from "./llamaindex/streaming/suggestion";
 
 export const chat = async (req: Request, res: Response) => {
   // Init Vercel AI StreamData and timeout
   const vercelStreamData = new StreamData();
-  const streamTimeout = createStreamTimeout(vercelStreamData);
   try {
     const { messages, data }: { messages: Message[]; data?: any } = req.body;
-    const userMessage = messages.pop();
-    if (!messages || !userMessage || userMessage.role !== "user") {
+    if (!isValidMessages(messages)) {
       return res.status(400).json({
         error:
           "messages are required in the request body and the last message must be from the user",
       });
     }
 
-    let annotations = userMessage.annotations;
-    if (!annotations) {
-      // the user didn't send any new annotations with the last message
-      // so use the annotations from the last user message that has annotations
-      // REASON: GPT4 doesn't consider MessageContentDetail from previous messages, only strings
-      annotations = messages
-        .slice()
-        .reverse()
-        .find(
-          (message) => message.role === "user" && message.annotations,
-        )?.annotations;
-    }
-
-    // retrieve document Ids from the annotations of all messages (if any) and create chat engine with index
-    const allAnnotations: JSONValue[] = [...messages, userMessage].flatMap(
-      (message) => {
-        return message.annotations ?? [];
-      },
-    );
-    const ids = retrieveDocumentIds(allAnnotations);
+    // retrieve document ids from the annotations of all messages (if any)
+    const ids = retrieveDocumentIds(messages);
+    // create chat engine with index using the document ids
     const chatEngine = await createChatEngine(ids, data);
 
-    // Convert message content from Vercel/AI format to LlamaIndex/OpenAI format
-    const userMessageContent = convertMessageContent(
-      userMessage.content,
-      annotations,
-    );
+    // retrieve user message content from Vercel/AI format
+    const userMessageContent = retrieveMessageContent(messages);
 
     // Setup callbacks
     const callbackManager = createCallbackManager(vercelStreamData);
@@ -96,7 +66,5 @@ export const chat = async (req: Request, res: Response) => {
     return res.status(500).json({
       detail: (error as Error).message,
     });
-  } finally {
-    clearTimeout(streamTimeout);
   }
 };

templates/types/streaming/express/src/controllers/sandbox.controller.ts

@@ -0,0 +1,140 @@
+/*
+ * Copyright 2023 FoundryLabs, Inc.
+ * Portions of this file are copied from the e2b project (https://github.com/e2b-dev/ai-artifacts)
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import {
+  CodeInterpreter,
+  ExecutionError,
+  Result,
+  Sandbox,
+} from "@e2b/code-interpreter";
+import { Request, Response } from "express";
+import { saveDocument } from "./llamaindex/documents/helper";
+
+type CodeArtifact = {
+  commentary: string;
+  template: string;
+  title: string;
+  description: string;
+  additional_dependencies: string[];
+  has_additional_dependencies: boolean;
+  install_dependencies_command: string;
+  port: number | null;
+  file_path: string;
+  code: string;
+};
+
+const sandboxTimeout = 10 * 60 * 1000; // 10 minute in ms
+
+export const maxDuration = 60;
+
+export type ExecutionResult = {
+  template: string;
+  stdout: string[];
+  stderr: string[];
+  runtimeError?: ExecutionError;
+  outputUrls: Array<{ url: string; filename: string }>;
+  url: string;
+};
+
+export const sandbox = async (req: Request, res: Response) => {
+  const { artifact }: { artifact: CodeArtifact } = req.body;
+
+  let sbx: Sandbox | CodeInterpreter | undefined = undefined;
+
+  // Create a interpreter or a sandbox
+  if (artifact.template === "code-interpreter-multilang") {
+    sbx = await CodeInterpreter.create({
+      metadata: { template: artifact.template },
+      timeoutMs: sandboxTimeout,
+    });
+    console.log("Created code interpreter", sbx.sandboxID);
+  } else {
+    sbx = await Sandbox.create(artifact.template, {
+      metadata: { template: artifact.template, userID: "default" },
+      timeoutMs: sandboxTimeout,
+    });
+    console.log("Created sandbox", sbx.sandboxID);
+  }
+
+  // Install packages
+  if (artifact.has_additional_dependencies) {
+    if (sbx instanceof CodeInterpreter) {
+      await sbx.notebook.execCell(artifact.install_dependencies_command);
+      console.log(
+        `Installed dependencies: ${artifact.additional_dependencies.join(", ")} in code interpreter ${sbx.sandboxID}`,
+      );
+    } else if (sbx instanceof Sandbox) {
+      await sbx.commands.run(artifact.install_dependencies_command);
+      console.log(
+        `Installed dependencies: ${artifact.additional_dependencies.join(", ")} in sandbox ${sbx.sandboxID}`,
+      );
+    }
+  }
+
+  // Copy code to fs
+  if (artifact.code && Array.isArray(artifact.code)) {
+    artifact.code.forEach(async (file) => {
+      await sbx.files.write(file.file_path, file.file_content);
+      console.log(`Copied file to ${file.file_path} in ${sbx.sandboxID}`);
+    });
+  } else {
+    await sbx.files.write(artifact.file_path, artifact.code);
+    console.log(`Copied file to ${artifact.file_path} in ${sbx.sandboxID}`);
+  }
+
+  // Execute code or return a URL to the running sandbox
+  if (artifact.template === "code-interpreter-multilang") {
+    const result = await (sbx as CodeInterpreter).notebook.execCell(
+      artifact.code || "",
+    );
+    await (sbx as CodeInterpreter).close();
+    const outputUrls = await downloadCellResults(result.results);
+
+    return res.status(200).json({
+      template: artifact.template,
+      stdout: result.logs.stdout,
+      stderr: result.logs.stderr,
+      runtimeError: result.error,
+      outputUrls: outputUrls,
+    });
+  } else {
+    return res.status(200).json({
+      template: artifact.template,
+      url: `https://${sbx?.getHost(artifact.port || 80)}`,
+    });
+  }
+};
+
+async function downloadCellResults(
+  cellResults?: Result[],
+): Promise<Array<{ url: string; filename: string }>> {
+  if (!cellResults) return [];
+  const results = await Promise.all(
+    cellResults.map(async (res) => {
+      const formats = res.formats(); // available formats in the result
+      const formatResults = await Promise.all(
+        formats.map(async (ext) => {
+          const filename = `${crypto.randomUUID()}.${ext}`;
+          const base64 = res[ext as keyof Result];
+          const buffer = Buffer.from(base64, "base64");
+          const fileurl = await saveDocument(filename, buffer);
+          return { url: fileurl, filename };
+        }),
+      );
+      return formatResults;
+    }),
+  );
+  return results.flat();
+}

templates/types/streaming/fastapi/app/api/routers/__init__.py

@@ -0,0 +1,18 @@
+from fastapi import APIRouter
+
+from .chat import chat_router  # noqa: F401
+from .chat_config import config_router  # noqa: F401
+from .upload import file_upload_router  # noqa: F401
+
+api_router = APIRouter()
+api_router.include_router(chat_router, prefix="/chat")
+api_router.include_router(config_router, prefix="/chat/config")
+api_router.include_router(file_upload_router, prefix="/chat/upload")
+
+# Dynamically adding additional routers if they exist
+try:
+    from .sandbox import sandbox_router  # noqa: F401
+
+    api_router.include_router(sandbox_router, prefix="/sandbox")
+except ImportError:
+    pass

templates/types/streaming/fastapi/app/api/routers/chat.py

@@ -1,8 +1,8 @@
 import logging
 from typing import List
 
-from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Request, status
-from llama_index.core.chat_engine.types import BaseChatEngine, NodeWithScore
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Request, status
+from llama_index.core.chat_engine.types import NodeWithScore
 from llama_index.core.llms import MessageRole
 
 from app.api.routers.events import EventCallbackHandler
@@ -13,7 +13,7 @@ from app.api.routers.models import (
     SourceNodes,
 )
 from app.api.routers.vercel_response import VercelStreamResponse
-from app.engine import get_chat_engine
+from app.engine.engine import get_chat_engine
 from app.engine.query_filter import generate_filters
 
 chat_router = r = APIRouter()
@@ -58,11 +58,19 @@ async def chat(
 @r.post("/request")
 async def chat_request(
     data: ChatData,
-    chat_engine: BaseChatEngine = Depends(get_chat_engine),
 ) -> Result:
     last_message_content = data.get_last_message_content()
     messages = data.get_history_messages()
 
+    doc_ids = data.get_chat_document_ids()
+    filters = generate_filters(doc_ids)
+    params = data.data or {}
+    logger.info(
+        f"Creating chat engine with filters: {str(filters)}",
+    )
+
+    chat_engine = get_chat_engine(filters=filters, params=params)
+
     response = await chat_engine.achat(last_message_content, messages)
     return Result(
         result=Message(role=MessageRole.ASSISTANT, content=response.response),

templates/types/streaming/fastapi/app/api/routers/chat_config.py

@@ -1,11 +1,10 @@
 import logging
 import os
 
-from fastapi import APIRouter
+from fastapi import APIRouter, HTTPException
 
 from app.api.routers.models import ChatConfig
 
-
 config_router = r = APIRouter()
 
 logger = logging.getLogger("uvicorn")
@@ -23,10 +22,14 @@ async def chat_config() -> ChatConfig:
 try:
     from app.engine.service import LLamaCloudFileService
 
-    logger.info("LlamaCloud is configured. Adding /config/llamacloud route.")
+    print("LlamaCloud is configured. Adding /config/llamacloud route.")
 
     @r.get("/llamacloud")
     async def chat_llama_cloud_config():
+        if not os.getenv("LLAMA_CLOUD_API_KEY"):
+            raise HTTPException(
+                status_code=500, detail="LlamaCloud API KEY is not configured"
+            )
         projects = LLamaCloudFileService.get_all_projects_with_pipelines()
         pipeline = os.getenv("LLAMA_CLOUD_INDEX_NAME")
         project = os.getenv("LLAMA_CLOUD_PROJECT_NAME")
@@ -42,7 +45,5 @@ try:
         }
 
 except ImportError:
-    logger.debug(
-        "LlamaCloud is not configured. Skipping adding /config/llamacloud route."
-    )
+    print("LlamaCloud is not configured. Skipping adding /config/llamacloud route.")
     pass

templates/types/streaming/fastapi/app/api/routers/events.py

@@ -1,13 +1,13 @@
-import json
 import asyncio
+import json
 import logging
-from typing import AsyncGenerator, Dict, Any, List, Optional
+from typing import Any, AsyncGenerator, Dict, List, Optional
+
 from llama_index.core.callbacks.base import BaseCallbackHandler
 from llama_index.core.callbacks.schema import CBEventType
 from llama_index.core.tools.types import ToolOutput
 from pydantic import BaseModel
 
-
 logger = logging.getLogger(__name__)
 
 
@@ -31,15 +31,20 @@ class CallbackEvent(BaseModel):
             return None
 
     def get_tool_message(self) -> dict | None:
+        if self.payload is None:
+            return None
         func_call_args = self.payload.get("function_call")
         if func_call_args is not None and "tool" in self.payload:
             tool = self.payload.get("tool")
+            if tool is None:
+                return None
             return {
                 "type": "events",
                 "data": {
                     "title": f"Calling tool: {tool.name} with inputs: {func_call_args}",
                 },
             }
+        return None
 
     def _is_output_serializable(self, output: Any) -> bool:
         try:
@@ -49,6 +54,8 @@ class CallbackEvent(BaseModel):
             return False
 
     def get_agent_tool_response(self) -> dict | None:
+        if self.payload is None:
+            return None
         response = self.payload.get("response")
         if response is not None:
             sources = response.sources
@@ -74,6 +81,7 @@ class CallbackEvent(BaseModel):
                             },
                         },
                     }
+        return None
 
     def to_response(self):
         try:
@@ -114,11 +122,13 @@ class EventCallbackHandler(BaseCallbackHandler):
         event_type: CBEventType,
         payload: Optional[Dict[str, Any]] = None,
         event_id: str = "",
+        parent_id: str = "",
         **kwargs: Any,
     ) -> str:
         event = CallbackEvent(event_id=event_id, event_type=event_type, payload=payload)
         if event.to_response() is not None:
             self._aqueue.put_nowait(event)
+        return event_id
 
     def on_event_end(
         self,

templates/types/streaming/fastapi/app/api/routers/models.py

@@ -1,6 +1,6 @@
 import logging
 import os
-from typing import Any, Dict, List, Literal, Optional
+from typing import Any, Dict, List, Literal, Optional, Union
 
 from llama_index.core.llms import ChatMessage, MessageRole
 from llama_index.core.schema import NodeWithScore
@@ -50,17 +50,35 @@ class AnnotationFileData(BaseModel):
         alias_generator = to_camel
 
 
+class AgentAnnotation(BaseModel):
+    agent: str
+    text: str
+
+
+class ArtifactAnnotation(BaseModel):
+    toolCall: Dict[str, Any]
+    toolOutput: Dict[str, Any]
+
+
 class Annotation(BaseModel):
     type: str
-    data: AnnotationFileData | List[str]
+    data: Union[AnnotationFileData, List[str], AgentAnnotation, ArtifactAnnotation]
 
-    def to_content(self) -> str | None:
+    def to_content(self) -> Optional[str]:
         if self.type == "document_file":
-            # We only support generating context content for CSV files for now
-            csv_files = [file for file in self.data.files if file.filetype == "csv"]
-            if len(csv_files) > 0:
-                return "Use data from following CSV raw content\n" + "\n".join(
-                    [f"```csv\n{csv_file.content.value}\n```" for csv_file in csv_files]
+            if isinstance(self.data, AnnotationFileData):
+                # We only support generating context content for CSV files for now
+                csv_files = [file for file in self.data.files if file.filetype == "csv"]
+                if len(csv_files) > 0:
+                    return "Use data from following CSV raw content\n" + "\n".join(
+                        [
+                            f"```csv\n{csv_file.content.value}\n```"
+                            for csv_file in csv_files
+                        ]
+                    )
+            else:
+                logger.warning(
+                    f"Unexpected data type for document_file annotation: {type(self.data)}"
                 )
         else:
             logger.warning(
@@ -119,14 +137,76 @@ class ChatData(BaseModel):
                 break
         return message_content
 
-    def get_history_messages(self) -> List[ChatMessage]:
+    def _get_agent_messages(self, max_messages: int = 10) -> List[str]:
+        """
+        Construct agent messages from the annotations in the chat messages
+        """
+        agent_messages = []
+        for message in self.messages:
+            if (
+                message.role == MessageRole.ASSISTANT
+                and message.annotations is not None
+            ):
+                for annotation in message.annotations:
+                    if annotation.type == "agent" and isinstance(
+                        annotation.data, AgentAnnotation
+                    ):
+                        text = annotation.data.text
+                        agent_messages.append(
+                            f"\nAgent: {annotation.data.agent}\nsaid: {text}\n"
+                        )
+                        if len(agent_messages) >= max_messages:
+                            break
+        return agent_messages
+
+    def _get_latest_code_artifact(self) -> Optional[str]:
+        """
+        Get latest code artifact from annotations to append to the user message
+        """
+        for message in reversed(self.messages):
+            if (
+                message.role == MessageRole.ASSISTANT
+                and message.annotations is not None
+            ):
+                for annotation in message.annotations:
+                    # type is tools and has `toolOutput` attribute
+                    if annotation.type == "tools" and isinstance(
+                        annotation.data, ArtifactAnnotation
+                    ):
+                        tool_output = annotation.data.toolOutput
+                        if tool_output and not tool_output.get("isError", False):
+                            return tool_output.get("output", {}).get("code", None)
+        return None
+
+    def get_history_messages(
+        self,
+        include_agent_messages: bool = False,
+        include_code_artifact: bool = True,
+    ) -> List[ChatMessage]:
         """
         Get the history messages
         """
-        return [
+        chat_messages = [
             ChatMessage(role=message.role, content=message.content)
             for message in self.messages[:-1]
         ]
+        if include_agent_messages:
+            agent_messages = self._get_agent_messages(max_messages=5)
+            if len(agent_messages) > 0:
+                message = ChatMessage(
+                    role=MessageRole.ASSISTANT,
+                    content="Previous agent events: \n" + "\n".join(agent_messages),
+                )
+                chat_messages.append(message)
+        if include_code_artifact:
+            latest_code_artifact = self._get_latest_code_artifact()
+            if latest_code_artifact:
+                message = ChatMessage(
+                    role=MessageRole.ASSISTANT,
+                    content=f"The existing code is:\n```\n{latest_code_artifact}\n```",
+                )
+                chat_messages.append(message)
+        return chat_messages
 
     def is_last_message_from_user(self) -> bool:
         return self.messages[-1].role == MessageRole.USER
@@ -141,6 +221,7 @@ class ChatData(BaseModel):
                 for annotation in message.annotations:
                     if (
                         annotation.type == "document_file"
+                        and isinstance(annotation.data, AnnotationFileData)
                         and annotation.data.files is not None
                     ):
                         for fi in annotation.data.files:
@@ -170,7 +251,7 @@ class SourceNodes(BaseModel):
         )
 
     @classmethod
-    def get_url_from_metadata(cls, metadata: Dict[str, Any]) -> str:
+    def get_url_from_metadata(cls, metadata: Dict[str, Any]) -> Optional[str]:
         url_prefix = os.getenv("FILESERVER_URL_PREFIX")
         if not url_prefix:
             logger.warning(