Release 0.1.3 (#84)

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

.changeset/config.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://unpkg.com/@changesets/config@3.0.0/schema.json",
   "changelog": "@changesets/cli/changelog",
-  "commit": true,
+  "commit": false,
   "fixed": [],
   "linked": [],
   "access": "public",

.changeset/cool-ears-train.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Use `gpt-4-turbo` model as default. Upgrade Python llama-index to 0.10.28

.changeset/curvy-eels-boil.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Remove asking for AI models and use defaults instead (OpenAIs GPT-4 Vision Preview and Embeddings v3). Use `--ask-models` CLI parameter to select models.

.changeset/dirty-cars-leave.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Add observability for Python

.changeset/four-tools-buy.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Use poetry run generate to generate embeddings for FastAPI

.changeset/khaki-phones-flash.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Use Settings object for LlamaIndex configuration

.changeset/lovely-olives-carry.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-Add Qdrant support

.github/workflows/e2e.yml

@@ -35,12 +35,13 @@ jobs:
         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 }}
-
-      - uses: pnpm/action-setup@v3
+          cache: "pnpm"
 
       - name: Install dependencies
         run: pnpm install

.github/workflows/lint_on_push_or_pull.yml

@@ -14,12 +14,13 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
+      - uses: pnpm/action-setup@v3
+
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:
           node-version-file: ".nvmrc"
-
-      - uses: pnpm/action-setup@v3
+          cache: "pnpm"
 
       - name: Install dependencies
         run: pnpm install

.github/workflows/publish_github.yml

@@ -0,0 +1,36 @@
+name: Publish to GitHub Releases
+
+on:
+  push:
+    tags:
+      - "v*"
+
+jobs:
+  build-and-publish:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout Repo
+        uses: actions/checkout@v4
+
+      - uses: pnpm/action-setup@v3
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version-file: ".nvmrc"
+          cache: "pnpm"
+
+      - name: Install dependencies
+        run: pnpm install
+
+      - name: Build tarball
+        run: |
+          pnpm pack
+
+      - name: Create release
+        uses: ncipollo/release-action@v1
+        with:
+          artifacts: "create-llama-*.tgz"
+          name: Release ${{ github.ref }}
+          bodyFile: "CHANGELOG.md"
+          token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/release.yml

@@ -15,17 +15,41 @@ jobs:
       - name: Checkout Repo
         uses: actions/checkout@v4
 
+      - uses: pnpm/action-setup@v3
+
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:
           node-version-file: ".nvmrc"
-
-      - uses: pnpm/action-setup@v3
+          cache: "pnpm"
 
       - name: Install dependencies
         run: pnpm install
 
-      - name: Create Release Pull Request
+      - name: Add auth token to .npmrc file
+        run: |
+          cat << EOF >> ".npmrc"
+            //registry.npmjs.org/:_authToken=$NPM_TOKEN
+          EOF
+        env:
+          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
+
+      - name: Get changeset status
+        id: get-changeset-status
+        run: |
+          pnpm changeset status --output .changeset/status.json
+          new_version=$(jq -r '.releases[0].newVersion' < .changeset/status.json)
+          rm -v .changeset/status.json
+          echo "new-version=${new_version}" >> "$GITHUB_OUTPUT"
+
+      - name: Create Release Pull Request or Publish to npm
+        id: changesets
         uses: changesets/action@v1
+        with:
+          commit: Release ${{ steps.get-changeset-status.outputs.new-version }}
+          title: Release ${{ steps.get-changeset-status.outputs.new-version }}
+          # build package and call changeset publish
+          publish: pnpm release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}

CHANGELOG.md

@@ -1,5 +1,41 @@
 # create-llama
 
+## 0.1.3
+
+### Patch Changes
+
+- 416073d: Directly import vector stores to work with NextJS
+
+## 0.1.2
+
+### Patch Changes
+
+- 056e376: Add support for displaying tool outputs (including weather widget as example)
+
+## 0.1.1
+
+### Patch Changes
+
+- 7bd3ed5: Support Anthropic and Gemini as model providers
+- 7bd3ed5: Support new agents from LITS 0.3
+- cfb5257: Display events (e.g. retrieving nodes) per chat message
+
+## 0.1.0
+
+### Minor Changes
+
+- f1c3e8d: Add Llama3 and Phi3 support using Ollama
+
+### Patch Changes
+
+- a0dec80: Use `gpt-4-turbo` model as default. Upgrade Python llama-index to 0.10.28
+- 753229d: Remove asking for AI models and use defaults instead (OpenAIs GPT-4 Vision Preview and Embeddings v3). Use `--ask-models` CLI parameter to select models.
+- 1d78202: Add observability for Python
+- 6acccd2: Use poetry run generate to generate embeddings for FastAPI
+- 9efcffe: Use Settings object for LlamaIndex configuration
+- 418bf9b: refactor: use tsx instead of ts-node
+- 1be69a5: Add Qdrant support
+
 ## 0.0.32
 
 ### Patch Changes

CONTRIBUTING.md

@@ -7,7 +7,7 @@ Install NodeJS. Preferably v18 using nvm or n.
 Inside the `create-llama` directory:
 
 ```
-npm i -g pnpm ts-node
+npm i -g pnpm
 pnpm install
 ```
 

create-app.ts

@@ -30,10 +30,8 @@ export async function createApp({
   appPath,
   packageManager,
   frontend,
-  openAiKey,
+  modelConfig,
   llamaCloudKey,
-  model,
-  embeddingModel,
   communityProjectConfig,
   llamapack,
   vectorDb,
@@ -77,10 +75,8 @@ export async function createApp({
     ui,
     packageManager,
     isOnline,
-    openAiKey,
+    modelConfig,
     llamaCloudKey,
-    model,
-    embeddingModel,
     communityProjectConfig,
     llamapack,
     vectorDb,

helpers/copy.ts

@@ -52,6 +52,7 @@ export const copy = async (
 export const assetRelocator = (name: string) => {
   switch (name) {
     case "gitignore":
+    case "npmrc":
     case "eslintrc.json": {
       return `.${name}`;
     }

helpers/env-variables.ts

@@ -1,6 +1,7 @@
 import fs from "fs/promises";
 import path from "path";
 import {
+  ModelConfig,
   TemplateDataSource,
   TemplateFramework,
   TemplateVectorDB,
@@ -28,7 +29,10 @@ const renderEnvVar = (envVars: EnvVar[]): string => {
   );
 };
 
-const getVectorDBEnvs = (vectorDb: TemplateVectorDB) => {
+const getVectorDBEnvs = (vectorDb?: TemplateVectorDB): EnvVar[] => {
+  if (!vectorDb) {
+    return [];
+  }
   switch (vectorDb) {
     case "mongo":
       return [
@@ -130,14 +134,111 @@ const getVectorDBEnvs = (vectorDb: TemplateVectorDB) => {
   }
 };
 
+const getModelEnvs = (modelConfig: ModelConfig): EnvVar[] => {
+  return [
+    {
+      name: "MODEL_PROVIDER",
+      description: "The provider for the AI models to use.",
+      value: modelConfig.provider,
+    },
+    {
+      name: "MODEL",
+      description: "The name of LLM model to use.",
+      value: modelConfig.model,
+    },
+    {
+      name: "EMBEDDING_MODEL",
+      description: "Name of the embedding model to use.",
+      value: modelConfig.embeddingModel,
+    },
+    {
+      name: "EMBEDDING_DIM",
+      description: "Dimension of the embedding model to use.",
+      value: modelConfig.dimensions.toString(),
+    },
+    ...(modelConfig.provider === "openai"
+      ? [
+          {
+            name: "OPENAI_API_KEY",
+            description: "The OpenAI API key to use.",
+            value: modelConfig.apiKey,
+          },
+          {
+            name: "LLM_TEMPERATURE",
+            description: "Temperature for sampling from the model.",
+          },
+          {
+            name: "LLM_MAX_TOKENS",
+            description: "Maximum number of tokens to generate.",
+          },
+        ]
+      : []),
+    ...(modelConfig.provider === "anthropic"
+      ? [
+          {
+            name: "ANTHROPIC_API_KEY",
+            description: "The Anthropic API key to use.",
+            value: modelConfig.apiKey,
+          },
+        ]
+      : []),
+    ...(modelConfig.provider === "gemini"
+      ? [
+          {
+            name: "GOOGLE_API_KEY",
+            description: "The Google API key to use.",
+            value: modelConfig.apiKey,
+          },
+        ]
+      : []),
+  ];
+};
+
+const getFrameworkEnvs = (
+  framework?: TemplateFramework,
+  port?: number,
+): EnvVar[] => {
+  if (framework !== "fastapi") {
+    return [];
+  }
+  return [
+    {
+      name: "APP_HOST",
+      description: "The address to start the backend app.",
+      value: "0.0.0.0",
+    },
+    {
+      name: "APP_PORT",
+      description: "The port to start the backend app.",
+      value: port?.toString() || "8000",
+    },
+    // TODO: Once LlamaIndexTS supports string templates, move this to `getEngineEnvs`
+    {
+      name: "SYSTEM_PROMPT",
+      description: `Custom system prompt.
+Example:
+SYSTEM_PROMPT="You are a helpful assistant who helps users with their questions."`,
+    },
+  ];
+};
+
+const getEngineEnvs = (): EnvVar[] => {
+  return [
+    {
+      name: "TOP_K",
+      description:
+        "The number of similar embeddings to return when retrieving documents.",
+      value: "3",
+    },
+  ];
+};
+
 export const createBackendEnvFile = async (
   root: string,
   opts: {
-    openAiKey?: string;
     llamaCloudKey?: string;
     vectorDb?: TemplateVectorDB;
-    model?: string;
-    embeddingModel?: string;
+    modelConfig: ModelConfig;
     framework?: TemplateFramework;
     dataSources?: TemplateDataSource[];
     port?: number;
@@ -145,95 +246,20 @@ export const createBackendEnvFile = async (
 ) => {
   // Init env values
   const envFileName = ".env";
-  const defaultEnvs = [
-    {
-      render: true,
-      name: "MODEL",
-      description: "The name of LLM model to use.",
-      value: opts.model,
-    },
-    {
-      render: true,
-      name: "OPENAI_API_KEY",
-      description: "The OpenAI API key to use.",
-      value: opts.openAiKey,
-    },
+  const envVars: EnvVar[] = [
     {
       name: "LLAMA_CLOUD_API_KEY",
       description: `The Llama Cloud API key.`,
       value: opts.llamaCloudKey,
     },
-    {
-      name: "EMBEDDING_MODEL",
-      description: "Name of the embedding model to use.",
-      value: opts.embeddingModel,
-    },
-    {
-      name: "EMBEDDING_DIM",
-      description: "Dimension of the embedding model to use.",
-      value: 1536,
-    },
+    // Add model environment variables
+    ...getModelEnvs(opts.modelConfig),
+    // Add engine environment variables
+    ...getEngineEnvs(),
     // Add vector database environment variables
-    ...(opts.vectorDb ? getVectorDBEnvs(opts.vectorDb) : []),
+    ...getVectorDBEnvs(opts.vectorDb),
+    ...getFrameworkEnvs(opts.framework, opts.port),
   ];
-  let envVars: EnvVar[] = [];
-  if (opts.framework === "fastapi") {
-    envVars = [
-      ...defaultEnvs,
-      ...[
-        {
-          name: "APP_HOST",
-          description: "The address to start the backend app.",
-          value: "0.0.0.0",
-        },
-        {
-          name: "APP_PORT",
-          description: "The port to start the backend app.",
-          value: opts.port?.toString() || "8000",
-        },
-        {
-          name: "LLM_TEMPERATURE",
-          description: "Temperature for sampling from the model.",
-        },
-        {
-          name: "LLM_MAX_TOKENS",
-          description: "Maximum number of tokens to generate.",
-        },
-        {
-          name: "TOP_K",
-          description:
-            "The number of similar embeddings to return when retrieving documents.",
-          value: "3",
-        },
-        {
-          name: "SYSTEM_PROMPT",
-          description: `Custom system prompt.
-Example:
-SYSTEM_PROMPT="
-We have provided context information below.
----------------------
-{context_str}
----------------------
-Given this information, please answer the question: {query_str}
-"`,
-        },
-      ],
-    ];
-  } else {
-    envVars = [
-      ...defaultEnvs,
-      ...[
-        opts.framework === "nextjs"
-          ? {
-              name: "NEXT_PUBLIC_MODEL",
-              description:
-                "The LLM model to use (hardcode to front-end artifact).",
-              value: opts.model,
-            }
-          : {},
-      ],
-    ];
-  }
   // Render and write env file
   const content = renderEnvVar(envVars);
   await fs.writeFile(path.join(root, envFileName), content);
@@ -244,20 +270,9 @@ export const createFrontendEnvFile = async (
   root: string,
   opts: {
     customApiPath?: string;
-    model?: string;
   },
 ) => {
   const defaultFrontendEnvs = [
-    {
-      name: "MODEL",
-      description: "The OpenAI model to use.",
-      value: opts.model,
-    },
-    {
-      name: "NEXT_PUBLIC_MODEL",
-      description: "The OpenAI model to use (hardcode to front-end artifact).",
-      value: opts.model,
-    },
     {
       name: "NEXT_PUBLIC_CHAT_API",
       description: "The backend API for chat endpoint.",

helpers/index.ts

@@ -15,6 +15,7 @@ import { ConfigFileType, writeToolsConfig } from "./tools";
 import {
   FileSourceConfig,
   InstallTemplateArgs,
+  ModelConfig,
   TemplateDataSource,
   TemplateFramework,
   TemplateVectorDB,
@@ -24,8 +25,8 @@ import { installTSTemplate } from "./typescript";
 // eslint-disable-next-line max-params
 async function generateContextData(
   framework: TemplateFramework,
+  modelConfig: ModelConfig,
   packageManager?: PackageManager,
-  openAiKey?: string,
   vectorDb?: TemplateVectorDB,
   llamaCloudKey?: string,
   useLlamaParse?: boolean,
@@ -36,12 +37,12 @@ async function generateContextData(
         ? "poetry run generate"
         : `${packageManager} run generate`,
     )}`;
-    const openAiKeyConfigured = openAiKey || process.env["OPENAI_API_KEY"];
+    const modelConfigured = modelConfig.isConfigured();
     const llamaCloudKeyConfigured = useLlamaParse
       ? llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
       : true;
     const hasVectorDb = vectorDb && vectorDb !== "none";
-    if (openAiKeyConfigured && llamaCloudKeyConfigured && !hasVectorDb) {
+    if (modelConfigured && llamaCloudKeyConfigured && !hasVectorDb) {
       // If all the required environment variables are set, run the generate script
       if (framework === "fastapi") {
         if (isHavingPoetryLockFile()) {
@@ -63,7 +64,7 @@ async function generateContextData(
 
     // generate the message of what to do to run the generate script manually
     const settings = [];
-    if (!openAiKeyConfigured) settings.push("your OpenAI key");
+    if (!modelConfigured) settings.push("your model provider API key");
     if (!llamaCloudKeyConfigured) settings.push("your Llama Cloud key");
     if (hasVectorDb) settings.push("your Vector DB environment variables");
     const settingsMessage =
@@ -141,11 +142,9 @@ export const installTemplate = async (
 
     // Copy the environment file to the target directory.
     await createBackendEnvFile(props.root, {
-      openAiKey: props.openAiKey,
+      modelConfig: props.modelConfig,
       llamaCloudKey: props.llamaCloudKey,
       vectorDb: props.vectorDb,
-      model: props.model,
-      embeddingModel: props.embeddingModel,
       framework: props.framework,
       dataSources: props.dataSources,
       port: props.externalPort,
@@ -163,8 +162,8 @@ export const installTemplate = async (
       ) {
         await generateContextData(
           props.framework,
+          props.modelConfig,
           props.packageManager,
-          props.openAiKey,
           props.vectorDb,
           props.llamaCloudKey,
           props.useLlamaParse,
@@ -174,7 +173,6 @@ export const installTemplate = async (
   } else {
     // this is a frontend for a full-stack app, create .env file with model information
     await createFrontendEnvFile(props.root, {
-      model: props.model,
       customApiPath: props.customApiPath,
     });
   }

helpers/providers/anthropic.ts

@@ -0,0 +1,106 @@
+import ciInfo from "ci-info";
+import prompts from "prompts";
+import { ModelConfigParams } from ".";
+import { questionHandlers, toChoice } from "../../questions";
+
+const MODELS = [
+  "claude-3-opus",
+  "claude-3-sonnet",
+  "claude-3-haiku",
+  "claude-2.1",
+  "claude-instant-1.2",
+];
+const DEFAULT_MODEL = MODELS[0];
+
+// TODO: get embedding vector dimensions from the anthropic sdk (currently not supported)
+// Use huggingface embedding models for now
+enum HuggingFaceEmbeddingModelType {
+  XENOVA_ALL_MINILM_L6_V2 = "all-MiniLM-L6-v2",
+  XENOVA_ALL_MPNET_BASE_V2 = "all-mpnet-base-v2",
+}
+type ModelData = {
+  dimensions: number;
+};
+const EMBEDDING_MODELS: Record<HuggingFaceEmbeddingModelType, ModelData> = {
+  [HuggingFaceEmbeddingModelType.XENOVA_ALL_MINILM_L6_V2]: {
+    dimensions: 384,
+  },
+  [HuggingFaceEmbeddingModelType.XENOVA_ALL_MPNET_BASE_V2]: {
+    dimensions: 768,
+  },
+};
+const DEFAULT_EMBEDDING_MODEL = Object.keys(EMBEDDING_MODELS)[0];
+const DEFAULT_DIMENSIONS = Object.values(EMBEDDING_MODELS)[0].dimensions;
+
+type AnthropicQuestionsParams = {
+  apiKey?: string;
+  askModels: boolean;
+};
+
+export async function askAnthropicQuestions({
+  askModels,
+  apiKey,
+}: AnthropicQuestionsParams): Promise<ModelConfigParams> {
+  const config: ModelConfigParams = {
+    apiKey,
+    model: DEFAULT_MODEL,
+    embeddingModel: DEFAULT_EMBEDDING_MODEL,
+    dimensions: DEFAULT_DIMENSIONS,
+    isConfigured(): boolean {
+      if (config.apiKey) {
+        return true;
+      }
+      if (process.env["ANTHROPIC_API_KEY"]) {
+        return true;
+      }
+      return false;
+    },
+  };
+
+  if (!config.apiKey) {
+    const { key } = await prompts(
+      {
+        type: "text",
+        name: "key",
+        message:
+          "Please provide your Anthropic API key (or leave blank to use ANTHROPIC_API_KEY env variable):",
+      },
+      questionHandlers,
+    );
+    config.apiKey = key || process.env.ANTHROPIC_API_KEY;
+  }
+
+  // use default model values in CI or if user should not be asked
+  const useDefaults = ciInfo.isCI || !askModels;
+  if (!useDefaults) {
+    const { model } = await prompts(
+      {
+        type: "select",
+        name: "model",
+        message: "Which LLM model would you like to use?",
+        choices: MODELS.map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.model = model;
+
+    const { embeddingModel } = await prompts(
+      {
+        type: "select",
+        name: "embeddingModel",
+        message: "Which embedding model would you like to use?",
+        choices: Object.keys(EMBEDDING_MODELS).map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.embeddingModel = embeddingModel;
+    config.dimensions =
+      EMBEDDING_MODELS[
+        embeddingModel as HuggingFaceEmbeddingModelType
+      ].dimensions;
+  }
+
+  return config;
+}

helpers/providers/gemini.ts

@@ -0,0 +1,87 @@
+import ciInfo from "ci-info";
+import prompts from "prompts";
+import { ModelConfigParams } from ".";
+import { questionHandlers, toChoice } from "../../questions";
+
+const MODELS = ["gemini-1.5-pro-latest", "gemini-pro", "gemini-pro-vision"];
+type ModelData = {
+  dimensions: number;
+};
+const EMBEDDING_MODELS: Record<string, ModelData> = {
+  "embedding-001": { dimensions: 768 },
+  "text-embedding-004": { dimensions: 768 },
+};
+
+const DEFAULT_MODEL = MODELS[0];
+const DEFAULT_EMBEDDING_MODEL = Object.keys(EMBEDDING_MODELS)[0];
+const DEFAULT_DIMENSIONS = Object.values(EMBEDDING_MODELS)[0].dimensions;
+
+type GeminiQuestionsParams = {
+  apiKey?: string;
+  askModels: boolean;
+};
+
+export async function askGeminiQuestions({
+  askModels,
+  apiKey,
+}: GeminiQuestionsParams): Promise<ModelConfigParams> {
+  const config: ModelConfigParams = {
+    apiKey,
+    model: DEFAULT_MODEL,
+    embeddingModel: DEFAULT_EMBEDDING_MODEL,
+    dimensions: DEFAULT_DIMENSIONS,
+    isConfigured(): boolean {
+      if (config.apiKey) {
+        return true;
+      }
+      if (process.env["GOOGLE_API_KEY"]) {
+        return true;
+      }
+      return false;
+    },
+  };
+
+  if (!config.apiKey) {
+    const { key } = await prompts(
+      {
+        type: "text",
+        name: "key",
+        message:
+          "Please provide your Google API key (or leave blank to use GOOGLE_API_KEY env variable):",
+      },
+      questionHandlers,
+    );
+    config.apiKey = key || process.env.GOOGLE_API_KEY;
+  }
+
+  // use default model values in CI or if user should not be asked
+  const useDefaults = ciInfo.isCI || !askModels;
+  if (!useDefaults) {
+    const { model } = await prompts(
+      {
+        type: "select",
+        name: "model",
+        message: "Which LLM model would you like to use?",
+        choices: MODELS.map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.model = model;
+
+    const { embeddingModel } = await prompts(
+      {
+        type: "select",
+        name: "embeddingModel",
+        message: "Which embedding model would you like to use?",
+        choices: Object.keys(EMBEDDING_MODELS).map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.embeddingModel = embeddingModel;
+    config.dimensions = EMBEDDING_MODELS[embeddingModel].dimensions;
+  }
+
+  return config;
+}

helpers/providers/index.ts

@@ -0,0 +1,67 @@
+import ciInfo from "ci-info";
+import prompts from "prompts";
+import { questionHandlers } from "../../questions";
+import { ModelConfig, ModelProvider } from "../types";
+import { askAnthropicQuestions } from "./anthropic";
+import { askGeminiQuestions } from "./gemini";
+import { askOllamaQuestions } from "./ollama";
+import { askOpenAIQuestions } from "./openai";
+
+const DEFAULT_MODEL_PROVIDER = "openai";
+
+export type ModelConfigQuestionsParams = {
+  openAiKey?: string;
+  askModels: boolean;
+};
+
+export type ModelConfigParams = Omit<ModelConfig, "provider">;
+
+export async function askModelConfig({
+  askModels,
+  openAiKey,
+}: ModelConfigQuestionsParams): Promise<ModelConfig> {
+  let modelProvider: ModelProvider = DEFAULT_MODEL_PROVIDER;
+  if (askModels && !ciInfo.isCI) {
+    const { provider } = await prompts(
+      {
+        type: "select",
+        name: "provider",
+        message: "Which model provider would you like to use",
+        choices: [
+          {
+            title: "OpenAI",
+            value: "openai",
+          },
+          { title: "Ollama", value: "ollama" },
+          { title: "Anthropic", value: "anthropic" },
+          { title: "Gemini", value: "gemini" },
+        ],
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    modelProvider = provider;
+  }
+
+  let modelConfig: ModelConfigParams;
+  switch (modelProvider) {
+    case "ollama":
+      modelConfig = await askOllamaQuestions({ askModels });
+      break;
+    case "anthropic":
+      modelConfig = await askAnthropicQuestions({ askModels });
+      break;
+    case "gemini":
+      modelConfig = await askGeminiQuestions({ askModels });
+      break;
+    default:
+      modelConfig = await askOpenAIQuestions({
+        openAiKey,
+        askModels,
+      });
+  }
+  return {
+    ...modelConfig,
+    provider: modelProvider,
+  };
+}

helpers/providers/ollama.ts

@@ -0,0 +1,95 @@
+import ciInfo from "ci-info";
+import ollama, { type ModelResponse } from "ollama";
+import { red } from "picocolors";
+import prompts from "prompts";
+import { ModelConfigParams } from ".";
+import { questionHandlers, toChoice } from "../../questions";
+
+type ModelData = {
+  dimensions: number;
+};
+const MODELS = ["llama3:8b", "wizardlm2:7b", "gemma:7b", "phi3"];
+const DEFAULT_MODEL = MODELS[0];
+// TODO: get embedding vector dimensions from the ollama sdk (currently not supported)
+const EMBEDDING_MODELS: Record<string, ModelData> = {
+  "nomic-embed-text": { dimensions: 768 },
+  "mxbai-embed-large": { dimensions: 1024 },
+  "all-minilm": { dimensions: 384 },
+};
+const DEFAULT_EMBEDDING_MODEL: string = Object.keys(EMBEDDING_MODELS)[0];
+
+type OllamaQuestionsParams = {
+  askModels: boolean;
+};
+
+export async function askOllamaQuestions({
+  askModels,
+}: OllamaQuestionsParams): Promise<ModelConfigParams> {
+  const config: ModelConfigParams = {
+    model: DEFAULT_MODEL,
+    embeddingModel: DEFAULT_EMBEDDING_MODEL,
+    dimensions: EMBEDDING_MODELS[DEFAULT_EMBEDDING_MODEL].dimensions,
+    isConfigured(): boolean {
+      return true;
+    },
+  };
+
+  // use default model values in CI or if user should not be asked
+  const useDefaults = ciInfo.isCI || !askModels;
+  if (!useDefaults) {
+    const { model } = await prompts(
+      {
+        type: "select",
+        name: "model",
+        message: "Which LLM model would you like to use?",
+        choices: MODELS.map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    await ensureModel(model);
+    config.model = model;
+
+    const { embeddingModel } = await prompts(
+      {
+        type: "select",
+        name: "embeddingModel",
+        message: "Which embedding model would you like to use?",
+        choices: Object.keys(EMBEDDING_MODELS).map(toChoice),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    await ensureModel(embeddingModel);
+    config.embeddingModel = embeddingModel;
+    config.dimensions = EMBEDDING_MODELS[embeddingModel].dimensions;
+  }
+
+  return config;
+}
+
+async function ensureModel(modelName: string) {
+  try {
+    if (modelName.split(":").length === 1) {
+      // model doesn't have a version suffix, use latest
+      modelName = modelName + ":latest";
+    }
+    const { models } = await ollama.list();
+    const found =
+      models.find((model: ModelResponse) => model.name === modelName) !==
+      undefined;
+    if (!found) {
+      console.log(
+        red(
+          `Model ${modelName} was not pulled yet. Call 'ollama pull ${modelName}' and try again.`,
+        ),
+      );
+      process.exit(1);
+    }
+  } catch (error) {
+    console.log(
+      red("Listing Ollama models failed. Is 'ollama' running? " + error),
+    );
+    process.exit(1);
+  }
+}

helpers/providers/openai.ts

@@ -0,0 +1,144 @@
+import ciInfo from "ci-info";
+import got from "got";
+import ora from "ora";
+import { red } from "picocolors";
+import prompts from "prompts";
+import { ModelConfigParams, ModelConfigQuestionsParams } from ".";
+import { questionHandlers } from "../../questions";
+
+const OPENAI_API_URL = "https://api.openai.com/v1";
+
+const DEFAULT_MODEL = "gpt-3.5-turbo";
+const DEFAULT_EMBEDDING_MODEL = "text-embedding-3-large";
+
+export async function askOpenAIQuestions({
+  openAiKey,
+  askModels,
+}: ModelConfigQuestionsParams): Promise<ModelConfigParams> {
+  const config: ModelConfigParams = {
+    apiKey: openAiKey,
+    model: DEFAULT_MODEL,
+    embeddingModel: DEFAULT_EMBEDDING_MODEL,
+    dimensions: getDimensions(DEFAULT_EMBEDDING_MODEL),
+    isConfigured(): boolean {
+      if (config.apiKey) {
+        return true;
+      }
+      if (process.env["OPENAI_API_KEY"]) {
+        return true;
+      }
+      return false;
+    },
+  };
+
+  if (!config.apiKey) {
+    const { key } = await prompts(
+      {
+        type: "text",
+        name: "key",
+        message: askModels
+          ? "Please provide your OpenAI API key (or leave blank to use OPENAI_API_KEY env variable):"
+          : "Please provide your OpenAI API key (leave blank to skip):",
+        validate: (value: string) => {
+          if (askModels && !value) {
+            if (process.env.OPENAI_API_KEY) {
+              return true;
+            }
+            return "OPENAI_API_KEY env variable is not set - key is required";
+          }
+          return true;
+        },
+      },
+      questionHandlers,
+    );
+    config.apiKey = key || process.env.OPENAI_API_KEY;
+  }
+
+  // use default model values in CI or if user should not be asked
+  const useDefaults = ciInfo.isCI || !askModels;
+  if (!useDefaults) {
+    const { model } = await prompts(
+      {
+        type: "select",
+        name: "model",
+        message: "Which LLM model would you like to use?",
+        choices: await getAvailableModelChoices(false, config.apiKey),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.model = model;
+
+    const { embeddingModel } = await prompts(
+      {
+        type: "select",
+        name: "embeddingModel",
+        message: "Which embedding model would you like to use?",
+        choices: await getAvailableModelChoices(true, config.apiKey),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    config.embeddingModel = embeddingModel;
+    config.dimensions = getDimensions(embeddingModel);
+  }
+
+  return config;
+}
+
+async function getAvailableModelChoices(
+  selectEmbedding: boolean,
+  apiKey?: string,
+) {
+  if (!apiKey) {
+    throw new Error("need OpenAI key to retrieve model choices");
+  }
+  const isLLMModel = (modelId: string) => {
+    return modelId.startsWith("gpt");
+  };
+
+  const isEmbeddingModel = (modelId: string) => {
+    return modelId.includes("embedding");
+  };
+
+  const spinner = ora("Fetching available models").start();
+  try {
+    const response = await got(`${OPENAI_API_URL}/models`, {
+      headers: {
+        Authorization: "Bearer " + apiKey,
+      },
+      timeout: 5000,
+      responseType: "json",
+    });
+    const data: any = await response.body;
+    spinner.stop();
+    return data.data
+      .filter((model: any) =>
+        selectEmbedding ? isEmbeddingModel(model.id) : isLLMModel(model.id),
+      )
+      .map((el: any) => {
+        return {
+          title: el.id,
+          value: el.id,
+        };
+      });
+  } catch (error) {
+    spinner.stop();
+    if ((error as any).response?.statusCode === 401) {
+      console.log(
+        red(
+          "Invalid OpenAI API key provided! Please provide a valid key and try again!",
+        ),
+      );
+    } else {
+      console.log(red("Request failed: " + error));
+    }
+    process.exit(1);
+  }
+}
+
+function getDimensions(modelName: string) {
+  // at 2024-04-24 all OpenAI embedding models support 1536 dimensions except
+  // "text-embedding-3-large", see https://openai.com/blog/new-embedding-models-and-api-updates
+  return modelName === "text-embedding-3-large" ? 1024 : 1536;
+}

helpers/python.ts

@@ -10,6 +10,7 @@ import { isPoetryAvailable, tryPoetryInstall } from "./poetry";
 import { Tool } from "./tools";
 import {
   InstallTemplateArgs,
+  ModelConfig,
   TemplateDataSource,
   TemplateVectorDB,
 } from "./types";
@@ -21,8 +22,9 @@ interface Dependency {
 }
 
 const getAdditionalDependencies = (
+  modelConfig: ModelConfig,
   vectorDb?: TemplateVectorDB,
-  dataSource?: TemplateDataSource,
+  dataSources?: TemplateDataSource[],
   tools?: Tool[],
 ) => {
   const dependencies: Dependency[] = [];
@@ -41,6 +43,7 @@ const getAdditionalDependencies = (
         name: "llama-index-vector-stores-postgres",
         version: "^0.1.1",
       });
+      break;
     }
     case "pinecone": {
       dependencies.push({
@@ -70,44 +73,88 @@ const getAdditionalDependencies = (
   }
 
   // Add data source dependencies
-  const dataSourceType = dataSource?.type;
-  switch (dataSourceType) {
-    case "file":
-      dependencies.push({
-        name: "docx2txt",
-        version: "^0.8",
-      });
-      break;
-    case "web":
-      dependencies.push({
-        name: "llama-index-readers-web",
-        version: "^0.1.6",
-      });
-      break;
-    case "db":
-      dependencies.push({
-        name: "llama-index-readers-database",
-        version: "^0.1.3",
-      });
-      dependencies.push({
-        name: "pymysql",
-        version: "^1.1.0",
-        extras: ["rsa"],
-      });
-      dependencies.push({
-        name: "psycopg2",
-        version: "^2.9.9",
-      });
-      break;
+  if (dataSources) {
+    for (const ds of dataSources) {
+      const dsType = ds?.type;
+      switch (dsType) {
+        case "file":
+          dependencies.push({
+            name: "docx2txt",
+            version: "^0.8",
+          });
+          break;
+        case "web":
+          dependencies.push({
+            name: "llama-index-readers-web",
+            version: "^0.1.6",
+          });
+          break;
+        case "db":
+          dependencies.push({
+            name: "llama-index-readers-database",
+            version: "^0.1.3",
+          });
+          dependencies.push({
+            name: "pymysql",
+            version: "^1.1.0",
+            extras: ["rsa"],
+          });
+          dependencies.push({
+            name: "psycopg2",
+            version: "^2.9.9",
+          });
+          break;
+      }
+    }
   }
 
   // Add tools dependencies
+  console.log("Adding tools dependencies");
   tools?.forEach((tool) => {
     tool.dependencies?.forEach((dep) => {
       dependencies.push(dep);
     });
   });
 
+  switch (modelConfig.provider) {
+    case "ollama":
+      dependencies.push({
+        name: "llama-index-llms-ollama",
+        version: "0.1.2",
+      });
+      dependencies.push({
+        name: "llama-index-embeddings-ollama",
+        version: "0.1.2",
+      });
+      break;
+    case "openai":
+      dependencies.push({
+        name: "llama-index-agent-openai",
+        version: "0.2.2",
+      });
+      break;
+    case "anthropic":
+      dependencies.push({
+        name: "llama-index-llms-anthropic",
+        version: "0.1.10",
+      });
+      dependencies.push({
+        name: "llama-index-embeddings-huggingface",
+        version: "0.2.0",
+      });
+      break;
+    case "gemini":
+      dependencies.push({
+        name: "llama-index-llms-gemini",
+        version: "0.1.7",
+      });
+      dependencies.push({
+        name: "llama-index-embeddings-gemini",
+        version: "0.1.6",
+      });
+      break;
+  }
+
   return dependencies;
 };
 
@@ -205,8 +252,8 @@ export const installPythonTemplate = async ({
   dataSources,
   tools,
   postInstallAction,
-  useLlamaParse,
   observability,
+  modelConfig,
 }: Pick<
   InstallTemplateArgs,
   | "root"
@@ -215,9 +262,9 @@ export const installPythonTemplate = async ({
   | "vectorDb"
   | "dataSources"
   | "tools"
-  | "useLlamaParse"
   | "postInstallAction"
   | "observability"
+  | "modelConfig"
 >) => {
   console.log("\nInitializing Python project with template:", template, "\n");
   const templatePath = path.join(templatesDir, "types", template, framework);
@@ -257,9 +304,14 @@ export const installPythonTemplate = async ({
     cwd: path.join(compPath, "engines", "python", engine),
   });
 
-  const addOnDependencies = dataSources
-    .map((ds) => getAdditionalDependencies(vectorDb, ds, tools))
-    .flat();
+  console.log("Adding additional dependencies");
+
+  const addOnDependencies = getAdditionalDependencies(
+    modelConfig,
+    vectorDb,
+    dataSources,
+    tools,
+  );
 
   if (observability === "opentelemetry") {
     addOnDependencies.push({

helpers/tools.ts

@@ -5,12 +5,18 @@ import yaml from "yaml";
 import { makeDir } from "./make-dir";
 import { TemplateFramework } from "./types";
 
+export enum ToolType {
+  LLAMAHUB = "llamahub",
+  LOCAL = "local",
+}
+
 export type Tool = {
   display: string;
   name: string;
   config?: Record<string, any>;
   dependencies?: ToolDependencies[];
   supportedFrameworks?: Array<TemplateFramework>;
+  type: ToolType;
 };
 
 export type ToolDependencies = {
@@ -35,6 +41,7 @@ export const supportedTools: Tool[] = [
       },
     ],
     supportedFrameworks: ["fastapi"],
+    type: ToolType.LLAMAHUB,
   },
   {
     display: "Wikipedia",
@@ -46,6 +53,14 @@ export const supportedTools: Tool[] = [
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
+    type: ToolType.LLAMAHUB,
+  },
+  {
+    display: "Weather",
+    name: "weather",
+    dependencies: [],
+    supportedFrameworks: ["fastapi", "express", "nextjs"],
+    type: ToolType.LOCAL,
   },
 ];
 
@@ -90,9 +105,19 @@ export const writeToolsConfig = async (
   type: ConfigFileType = ConfigFileType.YAML,
 ) => {
   if (tools.length === 0) return; // no tools selected, no config need
-  const configContent: Record<string, any> = {};
+  const configContent: {
+    [key in ToolType]: Record<string, any>;
+  } = {
+    local: {},
+    llamahub: {},
+  };
   tools.forEach((tool) => {
-    configContent[tool.name] = tool.config ?? {};
+    if (tool.type === ToolType.LLAMAHUB) {
+      configContent.llamahub[tool.name] = tool.config ?? {};
+    }
+    if (tool.type === ToolType.LOCAL) {
+      configContent.local[tool.name] = tool.config ?? {};
+    }
   });
   const configPath = path.join(root, "config");
   await makeDir(configPath);

helpers/types.ts

@@ -1,6 +1,15 @@
 import { PackageManager } from "../helpers/get-pkg-manager";
 import { Tool } from "./tools";
 
+export type ModelProvider = "openai" | "ollama" | "anthropic" | "gemini";
+export type ModelConfig = {
+  provider: ModelProvider;
+  apiKey?: string;
+  model: string;
+  embeddingModel: string;
+  dimensions: number;
+  isConfigured(): boolean;
+};
 export type TemplateType = "streaming" | "community" | "llamapack";
 export type TemplateFramework = "nextjs" | "express" | "fastapi";
 export type TemplateUI = "html" | "shadcn";
@@ -59,11 +68,9 @@ export interface InstallTemplateArgs {
   ui: TemplateUI;
   dataSources: TemplateDataSource[];
   customApiPath?: string;
-  openAiKey?: string;
+  modelConfig: ModelConfig;
   llamaCloudKey?: string;
   useLlamaParse?: boolean;
-  model: string;
-  embeddingModel: string;
   communityProjectConfig?: CommunityProjectConfig;
   llamapack?: string;
   vectorDb?: TemplateVectorDB;

helpers/typescript.ts

@@ -105,7 +105,7 @@ export const installTSTemplate = async ({
   const enginePath = path.join(root, relativeEngineDestPath, "engine");
 
   // copy vector db component
-  console.log("\nUsing vector DB:", vectorDb, "\n");
+  console.log("\nUsing vector DB:", vectorDb ?? "none", "\n");
   await copy("**", enginePath, {
     parents: true,
     cwd: path.join(compPath, "vectordbs", "typescript", vectorDb ?? "none"),
@@ -205,7 +205,7 @@ async function updatePackageJson({
     // add generate script if using context engine
     packageJson.scripts = {
       ...packageJson.scripts,
-      generate: `ts-node ${path.join(
+      generate: `tsx ${path.join(
         relativeEngineDestPath,
         "engine",
         "generate.ts",

index.ts

@@ -107,20 +107,6 @@ const program = new Commander.Command(packageJson.name)
     `
 
   Whether to generate a frontend for your backend.
-`,
-  )
-  .option(
-    "--model <model>",
-    `
-
-  Select OpenAI model to use. E.g. gpt-3.5-turbo.
-`,
-  )
-  .option(
-    "--embedding-model <embeddingModel>",
-    `
-
-  Select OpenAI embedding model to use. E.g. text-embedding-ada-002.
 `,
   )
   .option(
@@ -201,9 +187,7 @@ if (process.argv.includes("--tools")) {
 if (process.argv.includes("--no-llama-parse")) {
   program.useLlamaParse = false;
 }
-if (process.argv.includes("--ask-models")) {
-  program.askModels = true;
-}
+program.askModels = process.argv.includes("--ask-models");
 if (process.argv.includes("--no-files")) {
   program.dataSources = [];
 } else {
@@ -290,7 +274,11 @@ async function run(): Promise<void> {
   }
 
   const preferences = (conf.get("preferences") || {}) as QuestionArgs;
-  await askQuestions(program as unknown as QuestionArgs, preferences);
+  await askQuestions(
+    program as unknown as QuestionArgs,
+    preferences,
+    program.openAiKey,
+  );
 
   await createApp({
     template: program.template,
@@ -299,10 +287,8 @@ async function run(): Promise<void> {
     appPath: resolvedProjectPath,
     packageManager,
     frontend: program.frontend,
-    openAiKey: program.openAiKey,
+    modelConfig: program.modelConfig,
     llamaCloudKey: program.llamaCloudKey,
-    model: program.model,
-    embeddingModel: program.embeddingModel,
     communityProjectConfig: program.communityProjectConfig,
     llamapack: program.llamapack,
     vectorDb: program.vectorDb,

package.json

@@ -1,12 +1,12 @@
 {
   "name": "create-llama",
-  "version": "0.0.32",
+  "version": "0.1.3",
+  "description": "Create LlamaIndex-powered apps with one command",
   "keywords": [
     "rag",
     "llamaindex",
     "next.js"
   ],
-  "description": "Create LlamaIndex-powered apps with one command",
   "repository": {
     "type": "git",
     "url": "https://github.com/run-llama/LlamaIndexTS",
@@ -20,32 +20,30 @@
     "dist"
   ],
   "scripts": {
-    "clean": "rimraf --glob ./dist ./templates/**/__pycache__ ./templates/**/node_modules ./templates/**/poetry.lock",
-    "format": "prettier --ignore-unknown --cache --check .",
-    "format:write": "prettier --ignore-unknown --write .",
-    "dev": "ncc build ./index.ts -w -o dist/",
     "build": "bash ./scripts/build.sh",
     "build:ncc": "pnpm run clean && ncc build ./index.ts -o ./dist/ --minify --no-cache --no-source-map-register",
-    "lint": "eslint . --ignore-pattern dist --ignore-pattern e2e/cache",
+    "clean": "rimraf --glob ./dist ./templates/**/__pycache__ ./templates/**/node_modules ./templates/**/poetry.lock",
+    "dev": "ncc build ./index.ts -w -o dist/",
     "e2e": "playwright test",
+    "format": "prettier --ignore-unknown --cache --check .",
+    "format:write": "prettier --ignore-unknown --write .",
+    "lint": "eslint . --ignore-pattern dist --ignore-pattern e2e/cache",
+    "new-snapshot": "pnpm run build && changeset version --snapshot",
+    "new-version": "pnpm run build && changeset version",
+    "pack-install": "bash ./scripts/pack.sh",
     "prepare": "husky",
     "release": "pnpm run build && changeset publish",
-    "new-version": "pnpm run build && changeset version",
-    "release-snapshot": "pnpm run build && changeset publish --tag snapshot",
-    "new-snapshot": "pnpm run build && changeset version --snapshot",
-    "pack-install": "bash ./scripts/pack.sh"
+    "release-snapshot": "pnpm run build && changeset publish --tag snapshot"
   },
-  "devDependencies": {
-    "@playwright/test": "^1.41.1",
+  "dependencies": {
     "@types/async-retry": "1.4.2",
     "@types/ci-info": "2.0.0",
     "@types/cross-spawn": "6.0.0",
+    "@types/fs-extra": "11.0.4",
     "@types/node": "^20.11.7",
     "@types/prompts": "2.0.1",
     "@types/tar": "6.1.5",
     "@types/validate-npm-package-name": "3.0.0",
-    "@types/fs-extra": "11.0.4",
-    "@vercel/ncc": "0.38.1",
     "async-retry": "1.3.1",
     "async-sema": "3.0.1",
     "ci-info": "github:watson/ci-info#f43f6a1cefff47fb361c88cf4b943fdbcaafe540",
@@ -53,29 +51,34 @@
     "conf": "10.2.0",
     "cross-spawn": "7.0.3",
     "fast-glob": "3.3.1",
+    "fs-extra": "11.2.0",
     "got": "10.7.0",
+    "ollama": "^0.5.0",
+    "ora": "^8.0.1",
     "picocolors": "1.0.0",
     "prompts": "2.1.0",
-    "rimraf": "^5.0.5",
     "smol-toml": "^1.1.4",
     "tar": "6.1.15",
     "terminal-link": "^3.0.0",
     "update-check": "1.5.4",
     "validate-npm-package-name": "3.0.0",
-    "wait-port": "^1.1.0",
+    "yaml": "2.4.1"
+  },
+  "devDependencies": {
     "@changesets/cli": "^2.27.1",
+    "@playwright/test": "^1.41.1",
+    "@vercel/ncc": "0.38.1",
     "eslint": "^8.56.0",
+    "eslint-config-prettier": "^8.10.0",
     "husky": "^9.0.10",
     "prettier": "^3.2.5",
     "prettier-plugin-organize-imports": "^3.2.4",
+    "rimraf": "^5.0.5",
     "typescript": "^5.3.3",
-    "eslint-config-prettier": "^8.10.0",
-    "ora": "^8.0.1",
-    "fs-extra": "11.2.0",
-    "yaml": "2.4.1"
+    "wait-port": "^1.1.0"
   },
+  "packageManager": "pnpm@9.0.5",
   "engines": {
     "node": ">=16.14.0"
-  },
-  "packageManager": "pnpm@8.15.1"
+  }
 }

questions.ts

@@ -1,8 +1,6 @@
 import { execSync } from "child_process";
 import ciInfo from "ci-info";
 import fs from "fs";
-import got from "got";
-import ora from "ora";
 import path from "path";
 import { blue, green, red } from "picocolors";
 import prompts from "prompts";
@@ -16,11 +14,10 @@ import { COMMUNITY_OWNER, COMMUNITY_REPO } from "./helpers/constant";
 import { EXAMPLE_FILE } from "./helpers/datasources";
 import { templatesDir } from "./helpers/dir";
 import { getAvailableLlamapackOptions } from "./helpers/llama-pack";
+import { askModelConfig } from "./helpers/providers";
 import { getProjectOptions } from "./helpers/repo";
 import { supportedTools, toolsRequireConfig } from "./helpers/tools";
 
-const OPENAI_API_URL = "https://api.openai.com/v1";
-
 export type QuestionArgs = Omit<
   InstallAppArgs,
   "appPath" | "packageManager"
@@ -67,16 +64,13 @@ if ($dialogResult -eq [System.Windows.Forms.DialogResult]::OK)
 }
 `;
 
-const defaults: QuestionArgs = {
+const defaults: Omit<QuestionArgs, "modelConfig"> = {
   template: "streaming",
   framework: "nextjs",
   ui: "shadcn",
   frontend: false,
-  openAiKey: "",
   llamaCloudKey: "",
   useLlamaParse: false,
-  model: "gpt-4-turbo",
-  embeddingModel: "text-embedding-3-large",
   communityProjectConfig: undefined,
   llamapack: "",
   postInstallAction: "dependencies",
@@ -84,7 +78,7 @@ const defaults: QuestionArgs = {
   tools: [],
 };
 
-const handlers = {
+export const questionHandlers = {
   onCancel: () => {
     console.error("Exiting.");
     process.exit(1);
@@ -232,63 +226,15 @@ export const onPromptState = (state: any) => {
   }
 };
 
-const getAvailableModelChoices = async (
-  selectEmbedding: boolean,
-  apiKey?: string,
-) => {
-  const isLLMModel = (modelId: string) => {
-    return modelId.startsWith("gpt");
-  };
-
-  const isEmbeddingModel = (modelId: string) => {
-    return modelId.includes("embedding");
-  };
-
-  if (apiKey) {
-    const spinner = ora("Fetching available models").start();
-    try {
-      const response = await got(`${OPENAI_API_URL}/models`, {
-        headers: {
-          Authorization: "Bearer " + apiKey,
-        },
-        timeout: 5000,
-        responseType: "json",
-      });
-      const data: any = await response.body;
-      spinner.stop();
-      return data.data
-        .filter((model: any) =>
-          selectEmbedding ? isEmbeddingModel(model.id) : isLLMModel(model.id),
-        )
-        .map((el: any) => {
-          return {
-            title: el.id,
-            value: el.id,
-          };
-        });
-    } catch (error) {
-      spinner.stop();
-      if ((error as any).response?.statusCode === 401) {
-        console.log(
-          red(
-            "Invalid OpenAI API key provided! Please provide a valid key and try again!",
-          ),
-        );
-      } else {
-        console.log(red("Request failed: " + error));
-      }
-      process.exit(1);
-    }
-  }
-};
-
 export const askQuestions = async (
   program: QuestionArgs,
   preferences: QuestionArgs,
+  openAiKey?: string,
 ) => {
-  const getPrefOrDefault = <K extends keyof QuestionArgs>(
+  const getPrefOrDefault = <K extends keyof Omit<QuestionArgs, "modelConfig">>(
     field: K,
-  ): QuestionArgs[K] => preferences[field] ?? defaults[field];
+  ): Omit<QuestionArgs, "modelConfig">[K] =>
+    preferences[field] ?? defaults[field];
 
   // Ask for next action after installation
   async function askPostInstallAction() {
@@ -311,8 +257,8 @@ export const askQuestions = async (
           },
         ];
 
-        const openAiKeyConfigured =
-          program.openAiKey || process.env["OPENAI_API_KEY"];
+        const modelConfigured =
+          !program.llamapack && program.modelConfig.isConfigured();
         // If using LlamaParse, require LlamaCloud API key
         const llamaCloudKeyConfigured = program.useLlamaParse
           ? program.llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
@@ -321,10 +267,9 @@ export const askQuestions = async (
         // Can run the app if all tools do not require configuration
         if (
           !hasVectorDb &&
-          openAiKeyConfigured &&
+          modelConfigured &&
           llamaCloudKeyConfigured &&
-          !toolsRequireConfig(program.tools) &&
-          !program.llamapack
+          !toolsRequireConfig(program.tools)
         ) {
           actionChoices.push({
             title:
@@ -341,7 +286,7 @@ export const askQuestions = async (
             choices: actionChoices,
             initial: 1,
           },
-          handlers,
+          questionHandlers,
         );
 
         program.postInstallAction = action;
@@ -374,7 +319,7 @@ export const askQuestions = async (
           ],
           initial: 0,
         },
-        handlers,
+        questionHandlers,
       );
       program.template = template;
       preferences.template = template;
@@ -397,7 +342,7 @@ export const askQuestions = async (
         })),
         initial: 0,
       },
-      handlers,
+      questionHandlers,
     );
     const projectConfig = JSON.parse(communityProjectConfig);
     program.communityProjectConfig = projectConfig;
@@ -418,7 +363,7 @@ export const askQuestions = async (
         })),
         initial: 0,
       },
-      handlers,
+      questionHandlers,
     );
     program.llamapack = llamapack;
     preferences.llamapack = llamapack;
@@ -444,7 +389,7 @@ export const askQuestions = async (
           choices,
           initial: 0,
         },
-        handlers,
+        questionHandlers,
       );
       program.framework = framework;
       preferences.framework = framework;
@@ -453,7 +398,6 @@ export const askQuestions = async (
 
   if (program.framework === "express" || program.framework === "fastapi") {
     // if a backend-only framework is selected, ask whether we should create a frontend
-    // (only for streaming backends)
     if (program.frontend === undefined) {
       if (ciInfo.isCI) {
         program.frontend = getPrefOrDefault("frontend");
@@ -504,7 +448,7 @@ export const askQuestions = async (
           ],
           initial: 0,
         },
-        handlers,
+        questionHandlers,
       );
 
       program.observability = observability;
@@ -512,67 +456,13 @@ export const askQuestions = async (
     }
   }
 
-  if (!program.openAiKey) {
-    const { key } = await prompts(
-      {
-        type: "text",
-        name: "key",
-        message: program.askModels
-          ? "Please provide your OpenAI API key (or leave blank to reuse OPENAI_API_KEY env variable):"
-          : "Please provide your OpenAI API key (leave blank to skip):",
-        validate: (value: string) => {
-          if (program.askModels && !value) {
-            if (process.env.OPENAI_API_KEY) {
-              return true;
-            }
-            return "OpenAI API key is required";
-          }
-          return true;
-        },
-      },
-      handlers,
-    );
-
-    program.openAiKey = key || process.env.OPENAI_API_KEY;
-    preferences.openAiKey = key || process.env.OPENAI_API_KEY;
-  }
-
-  if (!program.model) {
-    if (ciInfo.isCI || !program.askModels) {
-      program.model = defaults.model;
-    } else {
-      const { model } = await prompts(
-        {
-          type: "select",
-          name: "model",
-          message: "Which LLM model would you like to use?",
-          choices: await getAvailableModelChoices(false, program.openAiKey),
-          initial: 0,
-        },
-        handlers,
-      );
-      program.model = model;
-      preferences.model = model;
-    }
-  }
-
-  if (!program.embeddingModel) {
-    if (ciInfo.isCI || !program.askModels) {
-      program.embeddingModel = defaults.embeddingModel;
-    } else {
-      const { embeddingModel } = await prompts(
-        {
-          type: "select",
-          name: "embeddingModel",
-          message: "Which embedding model would you like to use?",
-          choices: await getAvailableModelChoices(true, program.openAiKey),
-          initial: 0,
-        },
-        handlers,
-      );
-      program.embeddingModel = embeddingModel;
-      preferences.embeddingModel = embeddingModel;
-    }
+  if (!program.modelConfig) {
+    const modelConfig = await askModelConfig({
+      openAiKey,
+      askModels: program.askModels ?? false,
+    });
+    program.modelConfig = modelConfig;
+    preferences.modelConfig = modelConfig;
   }
 
   if (!program.dataSources) {
@@ -596,7 +486,7 @@ export const askQuestions = async (
             ),
             initial: firstQuestion ? 1 : 0,
           },
-          handlers,
+          questionHandlers,
         );
 
         if (selectedSource === "no" || selectedSource === "none") {
@@ -642,7 +532,7 @@ export const askQuestions = async (
                   return true;
                 },
               },
-              handlers,
+              questionHandlers,
             );
 
             program.dataSources.push({
@@ -687,7 +577,7 @@ export const askQuestions = async (
             ];
             program.dataSources.push({
               type: "db",
-              config: await prompts(dbPrompts, handlers),
+              config: await prompts(dbPrompts, questionHandlers),
             });
           }
         }
@@ -714,7 +604,7 @@ export const askQuestions = async (
           active: "yes",
           inactive: "no",
         },
-        handlers,
+        questionHandlers,
       );
       program.useLlamaParse = useLlamaParse;
 
@@ -727,7 +617,7 @@ export const askQuestions = async (
             message:
               "Please provide your LlamaIndex Cloud API key (leave blank to skip):",
           },
-          handlers,
+          questionHandlers,
         );
         program.llamaCloudKey = llamaCloudKey;
       }
@@ -746,7 +636,7 @@ export const askQuestions = async (
           choices: getVectorDbChoices(program.framework),
           initial: 0,
         },
-        handlers,
+        questionHandlers,
       );
       program.vectorDb = vectorDb;
       preferences.vectorDb = vectorDb;
@@ -781,3 +671,7 @@ export const askQuestions = async (
 
   await askPostInstallAction();
 };
+
+export const toChoice = (value: string) => {
+  return { title: value, value };
+};

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

@@ -1,35 +0,0 @@
-import os
-import yaml
-import importlib
-
-from llama_index.core.tools.tool_spec.base import BaseToolSpec
-from llama_index.core.tools.function_tool import FunctionTool
-
-
-class ToolFactory:
-
-    @staticmethod
-    def create_tool(tool_name: str, **kwargs) -> list[FunctionTool]:
-        try:
-            tool_package, tool_cls_name = tool_name.split(".")
-            module_name = f"llama_index.tools.{tool_package}"
-            module = importlib.import_module(module_name)
-            tool_class = getattr(module, tool_cls_name)
-            tool_spec: BaseToolSpec = tool_class(**kwargs)
-            return tool_spec.to_tool_list()
-        except (ImportError, AttributeError) as e:
-            raise ValueError(f"Unsupported tool: {tool_name}") from e
-        except TypeError as e:
-            raise ValueError(
-                f"Could not create tool: {tool_name}. With config: {kwargs}"
-            ) from e
-
-    @staticmethod
-    def from_env() -> list[FunctionTool]:
-        tools = []
-        if os.path.exists("config/tools.yaml"):
-            with open("config/tools.yaml", "r") as f:
-                tool_configs = yaml.safe_load(f)
-                for name, config in tool_configs.items():
-                    tools += ToolFactory.create_tool(name, **config)
-        return tools

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

@@ -0,0 +1,56 @@
+import os
+import yaml
+import importlib
+
+from llama_index.core.tools.tool_spec.base import BaseToolSpec
+from llama_index.core.tools.function_tool import FunctionTool
+
+
+class ToolType:
+    LLAMAHUB = "llamahub"
+    LOCAL = "local"
+
+
+class ToolFactory:
+
+    TOOL_SOURCE_PACKAGE_MAP = {
+        ToolType.LLAMAHUB: "llama_index.tools",
+        ToolType.LOCAL: "app.engine.tools",
+    }
+
+    @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:
+                tool_package, tool_cls_name = tool_name.split(".")
+                module_name = f"{source_package}.{tool_package}"
+                module = importlib.import_module(module_name)
+                tool_class = getattr(module, tool_cls_name)
+                tool_spec: BaseToolSpec = tool_class(**config)
+                return tool_spec.to_tool_list()
+            else:
+                module = importlib.import_module(f"{source_package}.{tool_name}")
+                tools = getattr(module, "tools")
+                if not all(isinstance(tool, FunctionTool) for tool in tools):
+                    raise ValueError(
+                        f"The module {module} does not contain valid tools"
+                    )
+                return tools
+        except ImportError as e:
+            raise ValueError(f"Failed to import tool {tool_name}: {e}")
+        except AttributeError as e:
+            raise ValueError(f"Failed to load tool {tool_name}: {e}")
+
+    @staticmethod
+    def from_env() -> list[FunctionTool]:
+        tools = []
+        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)
+                        )
+        return tools

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

@@ -0,0 +1,72 @@
+"""Open Meteo weather map tool spec."""
+
+import logging
+import requests
+import pytz
+from llama_index.core.tools import FunctionTool
+
+logger = logging.getLogger(__name__)
+
+
+class OpenMeteoWeather:
+    geo_api = "https://geocoding-api.open-meteo.com/v1"
+    weather_api = "https://api.open-meteo.com/v1"
+
+    @classmethod
+    def _get_geo_location(cls, location: str) -> dict:
+        """Get geo location from location name."""
+        params = {"name": location, "count": 10, "language": "en", "format": "json"}
+        response = requests.get(f"{cls.geo_api}/search", params=params)
+        if response.status_code != 200:
+            raise Exception(f"Failed to fetch geo location: {response.status_code}")
+        else:
+            data = response.json()
+            result = data["results"][0]
+            geo_location = {
+                "id": result["id"],
+                "name": result["name"],
+                "latitude": result["latitude"],
+                "longitude": result["longitude"],
+            }
+            return geo_location
+
+    @classmethod
+    def get_weather_information(cls, location: str) -> dict:
+        """Use this function to get the weather of any given location.
+        Note that the weather code should follow WMO Weather interpretation codes (WW):
+        0: Clear sky
+        1, 2, 3: Mainly clear, partly cloudy, and overcast
+        45, 48: Fog and depositing rime fog
+        51, 53, 55: Drizzle: Light, moderate, and dense intensity
+        56, 57: Freezing Drizzle: Light and dense intensity
+        61, 63, 65: Rain: Slight, moderate and heavy intensity
+        66, 67: Freezing Rain: Light and heavy intensity
+        71, 73, 75: Snow fall: Slight, moderate, and heavy intensity
+        77: Snow grains
+        80, 81, 82: Rain showers: Slight, moderate, and violent
+        85, 86: Snow showers slight and heavy
+        95: Thunderstorm: Slight or moderate
+        96, 99: Thunderstorm with slight and heavy hail
+        """
+        logger.info(
+            f"Calling open-meteo api to get weather information of location: {location}"
+        )
+        geo_location = cls._get_geo_location(location)
+        timezone = pytz.timezone("UTC").zone
+        params = {
+            "latitude": geo_location["latitude"],
+            "longitude": geo_location["longitude"],
+            "current": "temperature_2m,weather_code",
+            "hourly": "temperature_2m,weather_code",
+            "daily": "weather_code",
+            "timezone": timezone,
+        }
+        response = requests.get(f"{cls.weather_api}/forecast", params=params)
+        if response.status_code != 200:
+            raise Exception(
+                f"Failed to fetch weather information: {response.status_code}"
+            )
+        return response.json()
+
+
+tools = [FunctionTool.from_defaults(OpenMeteoWeather.get_weather_information)]

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

@@ -1,5 +1,6 @@
 import os
 from app.engine.index import get_index
+from fastapi import HTTPException
 
 
 def get_chat_engine():
@@ -8,8 +9,11 @@ def get_chat_engine():
 
     index = get_index()
     if index is None:
-        raise Exception(
-            "StorageContext is empty - call 'poetry run generate' to generate the storage first"
+        raise HTTPException(
+            status_code=500,
+            detail=str(
+                "StorageContext is empty - call 'poetry run generate' to generate the storage first"
+            ),
         )
 
     return index.as_chat_engine(

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

@@ -1,17 +1,13 @@
-import {
-  BaseTool,
-  OpenAIAgent,
-  QueryEngineTool,
-  Settings,
-  ToolFactory,
-} from "llamaindex";
+import { BaseToolWithCall, OpenAIAgent, QueryEngineTool } from "llamaindex";
+import { ToolsFactory } from "llamaindex/tools/ToolsFactory";
 import fs from "node:fs/promises";
 import path from "node:path";
 import { getDataSource } from "./index";
 import { STORAGE_CACHE_DIR } from "./shared";
+import { createLocalTools } from "./tools";
 
 export async function createChatEngine() {
-  let tools: BaseTool[] = [];
+  const tools: BaseToolWithCall[] = [];
 
   // Add a query engine tool if we have a data source
   // Delete this code if you don't have a data source
@@ -33,12 +29,17 @@ export async function createChatEngine() {
     const config = JSON.parse(
       await fs.readFile(path.join("config", "tools.json"), "utf8"),
     );
-    tools = tools.concat(await ToolFactory.createTools(config));
+
+    // add local tools from the 'tools' folder (if configured)
+    const localTools = createLocalTools(config.local);
+    tools.push(...localTools);
+
+    // add tools from LlamaIndexTS (if configured)
+    const llamaTools = await ToolsFactory.createTools(config.llamahub);
+    tools.push(...llamaTools);
   } catch {}
 
   return new OpenAIAgent({
     tools,
-    llm: Settings.llm,
-    verbose: true,
   });
 }

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

@@ -0,0 +1,26 @@
+import { BaseToolWithCall } from "llamaindex";
+import { WeatherTool, WeatherToolParams } from "./weather";
+
+type ToolCreator = (config: unknown) => BaseToolWithCall;
+
+const toolFactory: Record<string, ToolCreator> = {
+  weather: (config: unknown) => {
+    return new WeatherTool(config as WeatherToolParams);
+  },
+};
+
+export function createLocalTools(
+  localConfig: Record<string, unknown>,
+): BaseToolWithCall[] {
+  const tools: BaseToolWithCall[] = [];
+
+  Object.keys(localConfig).forEach((key) => {
+    if (key in toolFactory) {
+      const toolConfig = localConfig[key];
+      const tool = toolFactory[key](toolConfig);
+      tools.push(tool);
+    }
+  });
+
+  return tools;
+}

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

@@ -0,0 +1,81 @@
+import type { JSONSchemaType } from "ajv";
+import { BaseTool, ToolMetadata } from "llamaindex";
+
+interface GeoLocation {
+  id: string;
+  name: string;
+  latitude: number;
+  longitude: number;
+}
+
+export type WeatherParameter = {
+  location: string;
+};
+
+export type WeatherToolParams = {
+  metadata?: ToolMetadata<JSONSchemaType<WeatherParameter>>;
+};
+
+const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<WeatherParameter>> = {
+  name: "get_weather_information",
+  description: `
+    Use this function to get the weather of any given location.
+    Note that the weather code should follow WMO Weather interpretation codes (WW):
+    0: Clear sky
+    1, 2, 3: Mainly clear, partly cloudy, and overcast
+    45, 48: Fog and depositing rime fog
+    51, 53, 55: Drizzle: Light, moderate, and dense intensity
+    56, 57: Freezing Drizzle: Light and dense intensity
+    61, 63, 65: Rain: Slight, moderate and heavy intensity
+    66, 67: Freezing Rain: Light and heavy intensity
+    71, 73, 75: Snow fall: Slight, moderate, and heavy intensity
+    77: Snow grains
+    80, 81, 82: Rain showers: Slight, moderate, and violent
+    85, 86: Snow showers slight and heavy
+    95: Thunderstorm: Slight or moderate
+    96, 99: Thunderstorm with slight and heavy hail
+  `,
+  parameters: {
+    type: "object",
+    properties: {
+      location: {
+        type: "string",
+        description: "The location to get the weather information",
+      },
+    },
+    required: ["location"],
+  },
+};
+
+export class WeatherTool implements BaseTool<WeatherParameter> {
+  metadata: ToolMetadata<JSONSchemaType<WeatherParameter>>;
+
+  private getGeoLocation = async (location: string): Promise<GeoLocation> => {
+    const apiUrl = `https://geocoding-api.open-meteo.com/v1/search?name=${location}&count=10&language=en&format=json`;
+    const response = await fetch(apiUrl);
+    const data = await response.json();
+    const { id, name, latitude, longitude } = data.results[0];
+    return { id, name, latitude, longitude };
+  };
+
+  private getWeatherByLocation = async (location: string) => {
+    console.log(
+      "Calling open-meteo api to get weather information of location:",
+      location,
+    );
+    const { latitude, longitude } = await this.getGeoLocation(location);
+    const timezone = Intl.DateTimeFormat().resolvedOptions().timeZone;
+    const apiUrl = `https://api.open-meteo.com/v1/forecast?latitude=${latitude}&longitude=${longitude}&current=temperature_2m,weather_code&hourly=temperature_2m,weather_code&daily=weather_code&timezone=${timezone}`;
+    const response = await fetch(apiUrl);
+    const data = await response.json();
+    return data;
+  };
+
+  constructor(params?: WeatherToolParams) {
+    this.metadata = params?.metadata || DEFAULT_META_DATA;
+  }
+
+  async call(input: WeatherParameter) {
+    return await this.getWeatherByLocation(input.location);
+  }
+}

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

@@ -9,7 +9,9 @@ export async function createChatEngine() {
     );
   }
   const retriever = index.asRetriever();
-  retriever.similarityTopK = 3;
+  retriever.similarityTopK = process.env.TOP_K
+    ? parseInt(process.env.TOP_K)
+    : 3;
 
   return new ContextChatEngine({
     chatModel: Settings.llm,

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  AstraDBVectorStore,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { AstraDBVectorStore } from "llamaindex/storage/vectorStore/AstraDBVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars } from "./shared";

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

@@ -1,5 +1,6 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
-import { AstraDBVectorStore, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { AstraDBVectorStore } from "llamaindex/storage/vectorStore/AstraDBVectorStore";
 import { checkRequiredEnvVars } from "./shared";
 
 export async function getDataSource() {

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  MilvusVectorStore,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { MilvusVectorStore } from "llamaindex/storage/vectorStore/MilvusVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars, getMilvusClient } from "./shared";

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

@@ -1,4 +1,5 @@
-import { MilvusVectorStore, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { MilvusVectorStore } from "llamaindex/storage/vectorStore/MilvusVectorStore";
 import { checkRequiredEnvVars, getMilvusClient } from "./shared";
 
 export async function getDataSource() {

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  MongoDBAtlasVectorSearch,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { MongoDBAtlasVectorSearch } from "llamaindex/storage/vectorStore/MongoDBAtlasVectorSearch";
 import { MongoClient } from "mongodb";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";

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

@@ -1,5 +1,6 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
-import { MongoDBAtlasVectorSearch, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { MongoDBAtlasVectorSearch } from "llamaindex/storage/vectorStore/MongoDBAtlasVectorSearch";
 import { MongoClient } from "mongodb";
 import { checkRequiredEnvVars } from "./shared";
 

templates/components/vectordbs/typescript/none/generate.ts

@@ -1,4 +1,5 @@
-import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { storageContextFromDefaults } from "llamaindex/storage/StorageContext";
 
 import * as dotenv from "dotenv";
 

templates/components/vectordbs/typescript/none/index.ts

@@ -1,8 +1,5 @@
-import {
-  SimpleDocumentStore,
-  storageContextFromDefaults,
-  VectorStoreIndex,
-} from "llamaindex";
+import { SimpleDocumentStore, VectorStoreIndex } from "llamaindex";
+import { storageContextFromDefaults } from "llamaindex/storage/StorageContext";
 import { STORAGE_CACHE_DIR } from "./shared";
 
 export async function getDataSource() {

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  PGVectorStore,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { PGVectorStore } from "llamaindex/storage/vectorStore/PGVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import {

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

@@ -1,5 +1,6 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
-import { PGVectorStore, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { PGVectorStore } from "llamaindex/storage/vectorStore/PGVectorStore";
 import {
   PGVECTOR_SCHEMA,
   PGVECTOR_TABLE,

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

@@ -2,7 +2,7 @@ export const PGVECTOR_COLLECTION = "data";
 export const PGVECTOR_SCHEMA = "public";
 export const PGVECTOR_TABLE = "llamaindex_embedding";
 
-const REQUIRED_ENV_VARS = ["PG_CONNECTION_STRING", "OPENAI_API_KEY"];
+const REQUIRED_ENV_VARS = ["PG_CONNECTION_STRING"];
 
 export function checkRequiredEnvVars() {
   const missingEnvVars = REQUIRED_ENV_VARS.filter((envVar) => {

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  PineconeVectorStore,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { PineconeVectorStore } from "llamaindex/storage/vectorStore/PineconeVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars } from "./shared";

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

@@ -1,5 +1,6 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
-import { PineconeVectorStore, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { PineconeVectorStore } from "llamaindex/storage/vectorStore/PineconeVectorStore";
 import { checkRequiredEnvVars } from "./shared";
 
 export async function getDataSource() {

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

@@ -1,10 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
-import {
-  QdrantVectorStore,
-  VectorStoreIndex,
-  storageContextFromDefaults,
-} from "llamaindex";
+import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
+import { QdrantVectorStore } from "llamaindex/storage/vectorStore/QdrantVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars, getQdrantClient } from "./shared";
@@ -18,7 +15,10 @@ async function loadAndIndex() {
   const documents = await getDocuments();
 
   // Connect to Qdrant
-  const vectorStore = new QdrantVectorStore(collectionName, getQdrantClient());
+  const vectorStore = new QdrantVectorStore({
+    collectionName,
+    client: getQdrantClient(),
+  });
 
   const storageContext = await storageContextFromDefaults({ vectorStore });
   await VectorStoreIndex.fromDocuments(documents, {

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

@@ -1,5 +1,6 @@
 import * as dotenv from "dotenv";
-import { QdrantVectorStore, VectorStoreIndex } from "llamaindex";
+import { VectorStoreIndex } from "llamaindex";
+import { QdrantVectorStore } from "llamaindex/storage/vectorStore/QdrantVectorStore";
 import { checkRequiredEnvVars, getQdrantClient } from "./shared";
 
 dotenv.config();
@@ -7,7 +8,10 @@ dotenv.config();
 export async function getDataSource() {
   checkRequiredEnvVars();
   const collectionName = process.env.QDRANT_COLLECTION;
-  const store = new QdrantVectorStore(collectionName, getQdrantClient());
+  const store = new QdrantVectorStore({
+    collectionName,
+    client: getQdrantClient(),
+  });
 
   return await VectorStoreIndex.fromVectorStore(store);
 }

templates/types/streaming/express/npmrc

@@ -0,0 +1 @@
+node-linker=hoisted

templates/types/streaming/express/package.json

@@ -5,16 +5,18 @@
   "scripts": {
     "format": "prettier --ignore-unknown --cache --check .",
     "format:write": "prettier --ignore-unknown --write .",
-    "build": "tsup index.ts --format esm --dts",
-    "start": "node dist/index.mjs",
-    "dev": "concurrently \"tsup index.ts --format esm --dts --watch\" \"nodemon -q dist/index.mjs\""
+    "build": "tsup index.ts --format cjs --dts",
+    "start": "node dist/index.js",
+    "dev": "concurrently \"tsup index.ts --format cjs --dts --watch\" \"nodemon -q dist/index.mjs\""
   },
   "dependencies": {
-    "ai": "^2.2.25",
+    "ai": "^3.0.21",
     "cors": "^2.8.5",
     "dotenv": "^16.3.1",
     "express": "^4.18.2",
-    "llamaindex": "latest"
+    "llamaindex": "0.3.9",
+    "pdf2json": "3.0.5",
+    "ajv": "^8.12.0"
   },
   "devDependencies": {
     "@types/cors": "^2.8.16",
@@ -22,12 +24,12 @@
     "@types/node": "^20.9.5",
     "concurrently": "^8.2.2",
     "eslint": "^8.54.0",
+    "eslint-config-prettier": "^8.10.0",
     "nodemon": "^3.0.1",
-    "tsup": "^8.0.1",
-    "typescript": "^5.3.2",
     "prettier": "^3.2.5",
     "prettier-plugin-organize-imports": "^3.2.4",
-    "eslint-config-prettier": "^8.10.0",
-    "ts-node": "^10.9.2"
+    "tsx": "^4.7.2",
+    "tsup": "^8.0.1",
+    "typescript": "^5.3.2"
   }
 }

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

@@ -57,7 +57,7 @@ export const chatRequest = async (req: Request, res: Response) => {
   } catch (error) {
     console.error("[LlamaIndex]", error);
     return res.status(500).json({
-      error: (error as Error).message,
+      detail: (error as Error).message,
     });
   }
 };

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

@@ -1,8 +1,9 @@
-import { streamToResponse } from "ai";
+import { Message, StreamData, streamToResponse } from "ai";
 import { Request, Response } from "express";
-import { ChatMessage, MessageContent } from "llamaindex";
+import { ChatMessage, MessageContent, Settings } from "llamaindex";
 import { createChatEngine } from "./engine/chat";
 import { LlamaIndexStream } from "./llamaindex-stream";
+import { createCallbackManager } from "./stream-helper";
 
 const convertMessageContent = (
   textMessage: string,
@@ -25,7 +26,7 @@ const convertMessageContent = (
 
 export const chat = async (req: Request, res: Response) => {
   try {
-    const { messages, data }: { messages: ChatMessage[]; data: any } = 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({
@@ -42,36 +43,34 @@ export const chat = async (req: Request, res: Response) => {
       data?.imageUrl,
     );
 
+    // Init Vercel AI StreamData
+    const vercelStreamData = new StreamData();
+
+    // Setup callbacks
+    const callbackManager = createCallbackManager(vercelStreamData);
+
     // Calling LlamaIndex's ChatEngine to get a streamed response
-    const response = await chatEngine.chat({
-      message: userMessageContent,
-      chatHistory: messages,
-      stream: true,
+    const response = await Settings.withCallbackManager(callbackManager, () => {
+      return chatEngine.chat({
+        message: userMessageContent,
+        chatHistory: messages as ChatMessage[],
+        stream: true,
+      });
     });
 
     // Return a stream, which can be consumed by the Vercel/AI client
-    const { stream, data: streamData } = LlamaIndexStream(response, {
+    const stream = LlamaIndexStream(response, vercelStreamData, {
       parserOptions: {
         image_url: data?.imageUrl,
       },
     });
+    const processedStream = stream.pipeThrough(vercelStreamData.stream);
 
-    // Pipe LlamaIndexStream to response
-    const processedStream = stream.pipeThrough(streamData.stream);
-    return streamToResponse(processedStream, res, {
-      headers: {
-        // response MUST have the `X-Experimental-Stream-Data: 'true'` header
-        // so that the client uses the correct parsing logic, see
-        // https://sdk.vercel.ai/docs/api-reference/stream-data#on-the-server
-        "X-Experimental-Stream-Data": "true",
-        "Content-Type": "text/plain; charset=utf-8",
-        "Access-Control-Expose-Headers": "X-Experimental-Stream-Data",
-      },
-    });
+    return streamToResponse(processedStream, res);
   } catch (error) {
     console.error("[LlamaIndex]", error);
     return res.status(500).json({
-      error: (error as Error).message,
+      detail: (error as Error).message,
     });
   }
 };

templates/types/streaming/express/src/controllers/engine/settings.ts

@@ -1,19 +1,87 @@
-import { OpenAI, OpenAIEmbedding, Settings } from "llamaindex";
+import {
+  Anthropic,
+  GEMINI_EMBEDDING_MODEL,
+  GEMINI_MODEL,
+  Gemini,
+  GeminiEmbedding,
+  OpenAI,
+  OpenAIEmbedding,
+  Settings,
+} from "llamaindex";
+import { HuggingFaceEmbedding } from "llamaindex/embeddings/HuggingFaceEmbedding";
+import { OllamaEmbedding } from "llamaindex/embeddings/OllamaEmbedding";
+import { ALL_AVAILABLE_ANTHROPIC_MODELS } from "llamaindex/llm/anthropic";
+import { Ollama } from "llamaindex/llm/ollama";
 
 const CHUNK_SIZE = 512;
 const CHUNK_OVERLAP = 20;
 
 export const initSettings = async () => {
+  // HINT: you can delete the initialization code for unused model providers
+  console.log(`Using '${process.env.MODEL_PROVIDER}' model provider`);
+
+  if (!process.env.MODEL || !process.env.EMBEDDING_MODEL) {
+    throw new Error("'MODEL' and 'EMBEDDING_MODEL' env variables must be set.");
+  }
+
+  switch (process.env.MODEL_PROVIDER) {
+    case "ollama":
+      initOllama();
+      break;
+    case "anthropic":
+      initAnthropic();
+      break;
+    case "gemini":
+      initGemini();
+      break;
+    default:
+      initOpenAI();
+      break;
+  }
+  Settings.chunkSize = CHUNK_SIZE;
+  Settings.chunkOverlap = CHUNK_OVERLAP;
+};
+
+function initOpenAI() {
   Settings.llm = new OpenAI({
     model: process.env.MODEL ?? "gpt-3.5-turbo",
     maxTokens: 512,
   });
-  Settings.chunkSize = CHUNK_SIZE;
-  Settings.chunkOverlap = CHUNK_OVERLAP;
   Settings.embedModel = new OpenAIEmbedding({
     model: process.env.EMBEDDING_MODEL,
     dimensions: process.env.EMBEDDING_DIM
       ? parseInt(process.env.EMBEDDING_DIM)
       : undefined,
   });
-};
+}
+
+function initOllama() {
+  Settings.llm = new Ollama({
+    model: process.env.MODEL ?? "",
+  });
+  Settings.embedModel = new OllamaEmbedding({
+    model: process.env.EMBEDDING_MODEL ?? "",
+  });
+}
+
+function initAnthropic() {
+  const embedModelMap: Record<string, string> = {
+    "all-MiniLM-L6-v2": "Xenova/all-MiniLM-L6-v2",
+    "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
+  };
+  Settings.llm = new Anthropic({
+    model: process.env.MODEL as keyof typeof ALL_AVAILABLE_ANTHROPIC_MODELS,
+  });
+  Settings.embedModel = new HuggingFaceEmbedding({
+    modelType: embedModelMap[process.env.EMBEDDING_MODEL!],
+  });
+}
+
+function initGemini() {
+  Settings.llm = new Gemini({
+    model: process.env.MODEL as GEMINI_MODEL,
+  });
+  Settings.embedModel = new GeminiEmbedding({
+    model: process.env.EMBEDDING_MODEL as GEMINI_EMBEDDING_MODEL,
+  });
+}

templates/types/streaming/express/src/controllers/llamaindex-stream.ts

@@ -1,49 +1,65 @@
 import {
-  JSONValue,
+  StreamData,
   createCallbacksTransformer,
   createStreamDataTransformer,
-  experimental_StreamData,
   trimStartOfStreamHelper,
   type AIStreamCallbacksAndOptions,
 } from "ai";
-import { Response, StreamingAgentChatResponse } from "llamaindex";
+import {
+  Metadata,
+  NodeWithScore,
+  Response,
+  ToolCallLLMMessageOptions,
+} from "llamaindex";
+
+import { AgentStreamChatResponse } from "llamaindex/agent/base";
+import { appendImageData, appendSourceData } from "./stream-helper";
+
+type LlamaIndexResponse =
+  | AgentStreamChatResponse<ToolCallLLMMessageOptions>
+  | Response;
 
 type ParserOptions = {
   image_url?: string;
 };
 
 function createParser(
-  res: AsyncIterable<Response>,
-  data: experimental_StreamData,
+  res: AsyncIterable<LlamaIndexResponse>,
+  data: StreamData,
   opts?: ParserOptions,
 ) {
   const it = res[Symbol.asyncIterator]();
   const trimStartOfStream = trimStartOfStreamHelper();
+
+  let sourceNodes: NodeWithScore<Metadata>[] | undefined;
   return new ReadableStream<string>({
     start() {
-      // if image_url is provided, send it via the data stream
-      if (opts?.image_url) {
-        const message: JSONValue = {
-          type: "image_url",
-          image_url: {
-            url: opts.image_url,
-          },
-        };
-        data.append(message);
-      } else {
-        data.append({}); // send an empty image response for the user's message
-      }
+      appendImageData(data, opts?.image_url);
     },
     async pull(controller): Promise<void> {
       const { value, done } = await it.next();
       if (done) {
+        if (sourceNodes) {
+          appendSourceData(data, sourceNodes);
+        }
         controller.close();
-        data.append({}); // send an empty image response for the assistant's message
         data.close();
         return;
       }
 
-      const text = trimStartOfStream(value.response ?? "");
+      let delta;
+      if (value instanceof Response) {
+        // handle Response type
+        if (value.sourceNodes) {
+          // get source nodes from the first response
+          sourceNodes = value.sourceNodes;
+        }
+        delta = value.response ?? "";
+      } else {
+        // handle other types
+        delta = value.response.delta;
+      }
+      const text = trimStartOfStream(delta ?? "");
       if (text) {
         controller.enqueue(text);
       }
@@ -52,21 +68,14 @@ function createParser(
 }
 
 export function LlamaIndexStream(
-  response: StreamingAgentChatResponse | AsyncIterable<Response>,
+  response: AsyncIterable<LlamaIndexResponse>,
+  data: StreamData,
   opts?: {
     callbacks?: AIStreamCallbacksAndOptions;
     parserOptions?: ParserOptions;
   },
-): { stream: ReadableStream; data: experimental_StreamData } {
-  const data = new experimental_StreamData();
-  const res =
-    response instanceof StreamingAgentChatResponse
-      ? response.response
-      : response;
-  return {
-    stream: createParser(res, data, opts?.parserOptions)
-      .pipeThrough(createCallbacksTransformer(opts?.callbacks))
-      .pipeThrough(createStreamDataTransformer(true)),
-    data,
-  };
+): ReadableStream<Uint8Array> {
+  return createParser(response, data, opts?.parserOptions)
+    .pipeThrough(createCallbacksTransformer(opts?.callbacks))
+    .pipeThrough(createStreamDataTransformer());
 }

templates/types/streaming/express/src/controllers/stream-helper.ts

@@ -0,0 +1,97 @@
+import { StreamData } from "ai";
+import {
+  CallbackManager,
+  Metadata,
+  NodeWithScore,
+  ToolCall,
+  ToolOutput,
+} from "llamaindex";
+
+export function appendImageData(data: StreamData, imageUrl?: string) {
+  if (!imageUrl) return;
+  data.appendMessageAnnotation({
+    type: "image",
+    data: {
+      url: imageUrl,
+    },
+  });
+}
+
+export function appendSourceData(
+  data: StreamData,
+  sourceNodes?: NodeWithScore<Metadata>[],
+) {
+  if (!sourceNodes?.length) return;
+  data.appendMessageAnnotation({
+    type: "sources",
+    data: {
+      nodes: sourceNodes.map((node) => ({
+        ...node.node.toMutableJSON(),
+        id: node.node.id_,
+        score: node.score ?? null,
+      })),
+    },
+  });
+}
+
+export function appendEventData(data: StreamData, title?: string) {
+  if (!title) return;
+  data.appendMessageAnnotation({
+    type: "events",
+    data: {
+      title,
+    },
+  });
+}
+
+export function appendToolData(
+  data: StreamData,
+  toolCall: ToolCall,
+  toolOutput: ToolOutput,
+) {
+  data.appendMessageAnnotation({
+    type: "tools",
+    data: {
+      toolCall: {
+        id: toolCall.id,
+        name: toolCall.name,
+        input: toolCall.input,
+      },
+      toolOutput: {
+        output: toolOutput.output,
+        isError: toolOutput.isError,
+      },
+    },
+  });
+}
+
+export function createCallbackManager(stream: StreamData) {
+  const callbackManager = new CallbackManager();
+
+  callbackManager.on("retrieve", (data) => {
+    const { nodes, query } = data.detail;
+    appendEventData(stream, `Retrieving context for query: '${query}'`);
+    appendEventData(
+      stream,
+      `Retrieved ${nodes.length} sources to use as context for the query`,
+    );
+  });
+
+  callbackManager.on("llm-tool-call", (event) => {
+    const { name, input } = event.detail.payload.toolCall;
+    const inputString = Object.entries(input)
+      .map(([key, value]) => `${key}: ${value}`)
+      .join(", ");
+    appendEventData(
+      stream,
+      `Using tool: '${name}' with inputs: '${inputString}'`,
+    );
+  });
+
+  callbackManager.on("llm-tool-result", (event) => {
+    const { toolCall, toolResult } = event.detail.payload;
+    appendToolData(stream, toolCall, toolResult);
+  });
+
+  return callbackManager;
+}

templates/types/streaming/express/tsconfig.json

@@ -9,10 +9,5 @@
     "paths": {
       "@/*": ["./*"]
     }
-  },
-  "ts-node": {
-    "compilerOptions": {
-      "module": "commonjs"
-    }
   }
 }

templates/types/streaming/fastapi/README-template.md

@@ -11,13 +11,7 @@ poetry install
 poetry shell
 ```
 
-By default, we use the OpenAI LLM (though you can customize, see `app/settings.py`). As a result, you need to specify an `OPENAI_API_KEY` in an .env file in this directory.
-
-Example `.env` file:
-
-```
-OPENAI_API_KEY=<openai_api_key>
-```
+Then check the parameters that have been pre-configured in the `.env` file in this directory. (E.g. you might need to configure an `OPENAI_API_KEY` if you're using OpenAI as model provider).
 
 If you are using any tools or data sources, you can update their config files in the `config` folder.
 

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

@@ -1,13 +1,13 @@
 from pydantic import BaseModel
 from typing import List, Any, Optional, Dict, Tuple
 from fastapi import APIRouter, Depends, HTTPException, Request, status
-from fastapi.responses import StreamingResponse
-from llama_index.core.chat_engine.types import (
-    BaseChatEngine,
-)
+from llama_index.core.chat_engine.types import BaseChatEngine
 from llama_index.core.schema import NodeWithScore
 from llama_index.core.llms import ChatMessage, MessageRole
 from app.engine import get_chat_engine
+from app.api.routers.vercel_response import VercelStreamResponse
+from app.api.routers.messaging import EventCallbackHandler
+from aiostream import stream
 
 chat_router = r = APIRouter()
 
@@ -37,6 +37,7 @@ class _SourceNodes(BaseModel):
     id: str
     metadata: Dict[str, Any]
     score: Optional[float]
+    text: str
 
     @classmethod
     def from_source_node(cls, source_node: NodeWithScore):
@@ -44,6 +45,7 @@ class _SourceNodes(BaseModel):
             id=source_node.node.node_id,
             metadata=source_node.node.metadata,
             score=source_node.score,
+            text=source_node.node.text,  # type: ignore
         )
 
     @classmethod
@@ -89,15 +91,46 @@ async def chat(
 ):
     last_message_content, messages = await parse_chat_data(data)
 
+    event_handler = EventCallbackHandler()
+    chat_engine.callback_manager.handlers.append(event_handler)  # type: ignore
     response = await chat_engine.astream_chat(last_message_content, messages)
 
-    async def event_generator():
-        async for token in response.async_response_gen():
-            if await request.is_disconnected():
-                break
-            yield token
+    async def content_generator():
+        # Yield the text response
+        async def _text_generator():
+            async for token in response.async_response_gen():
+                yield VercelStreamResponse.convert_text(token)
+            # the text_generator is the leading stream, once it's finished, also finish the event stream
+            event_handler.is_done = True
 
-    return StreamingResponse(event_generator(), media_type="text/plain")
+        # Yield the events from the event handler
+        async def _event_generator():
+            async for event in event_handler.async_event_gen():
+                event_response = event.to_response()
+                if event_response is not None:
+                    yield VercelStreamResponse.convert_data(event_response)
+
+        combine = stream.merge(_text_generator(), _event_generator())
+        async with combine.stream() as streamer:
+            async for item in streamer:
+                if await request.is_disconnected():
+                    break
+                yield item
+
+        # Yield the source nodes
+        yield VercelStreamResponse.convert_data(
+            {
+                "type": "sources",
+                "data": {
+                    "nodes": [
+                        _SourceNodes.from_source_node(node).dict()
+                        for node in response.source_nodes
+                    ]
+                },
+            }
+        )
+
+    return VercelStreamResponse(content=content_generator())
 
 
 # non-streaming endpoint - delete if not needed

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

@@ -0,0 +1,141 @@
+import json
+import asyncio
+from typing import AsyncGenerator, Dict, Any, 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
+
+
+class CallbackEvent(BaseModel):
+    event_type: CBEventType
+    payload: Optional[Dict[str, Any]] = None
+    event_id: str = ""
+
+    def get_retrieval_message(self) -> dict | None:
+        if self.payload:
+            nodes = self.payload.get("nodes")
+            if nodes:
+                msg = f"Retrieved {len(nodes)} sources to use as context for the query"
+            else:
+                msg = f"Retrieving context for query: '{self.payload.get('query_str')}'"
+            return {
+                "type": "events",
+                "data": {"title": msg},
+            }
+        else:
+            return None
+
+    def get_tool_message(self) -> dict | 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")
+            return {
+                "type": "events",
+                "data": {
+                    "title": f"Calling tool: {tool.name} with inputs: {func_call_args}",
+                },
+            }
+
+    def _is_output_serializable(self, output: Any) -> bool:
+        try:
+            json.dumps(output)
+            return True
+        except TypeError:
+            return False
+
+    def get_agent_tool_response(self) -> dict | None:
+        response = self.payload.get("response")
+        if response is not None:
+            sources = response.sources
+            for source in sources:
+                # Return the tool response here to include the toolCall information
+                if isinstance(source, ToolOutput):
+                    if self._is_output_serializable(source.raw_output):
+                        output = source.raw_output
+                    else:
+                        output = source.content
+
+                    return {
+                        "type": "tools",
+                        "data": {
+                            "toolOutput": {
+                                "output": output,
+                                "isError": source.is_error,
+                            },
+                            "toolCall": {
+                                "id": None,  # There is no tool id in the ToolOutput
+                                "name": source.tool_name,
+                                "input": source.raw_input,
+                            },
+                        },
+                    }
+
+    def to_response(self):
+        match self.event_type:
+            case "retrieve":
+                return self.get_retrieval_message()
+            case "function_call":
+                return self.get_tool_message()
+            case "agent_step":
+                return self.get_agent_tool_response()
+            case _:
+                return None
+
+
+class EventCallbackHandler(BaseCallbackHandler):
+    _aqueue: asyncio.Queue
+    is_done: bool = False
+
+    def __init__(
+        self,
+    ):
+        """Initialize the base callback handler."""
+        ignored_events = [
+            CBEventType.CHUNKING,
+            CBEventType.NODE_PARSING,
+            CBEventType.EMBEDDING,
+            CBEventType.LLM,
+            CBEventType.TEMPLATING,
+        ]
+        super().__init__(ignored_events, ignored_events)
+        self._aqueue = asyncio.Queue()
+
+    def on_event_start(
+        self,
+        event_type: CBEventType,
+        payload: Optional[Dict[str, Any]] = None,
+        event_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)
+
+    def on_event_end(
+        self,
+        event_type: CBEventType,
+        payload: Optional[Dict[str, Any]] = None,
+        event_id: str = "",
+        **kwargs: Any,
+    ) -> None:
+        event = CallbackEvent(event_id=event_id, event_type=event_type, payload=payload)
+        if event.to_response() is not None:
+            self._aqueue.put_nowait(event)
+
+    def start_trace(self, trace_id: Optional[str] = None) -> None:
+        """No-op."""
+
+    def end_trace(
+        self,
+        trace_id: Optional[str] = None,
+        trace_map: Optional[Dict[str, List[str]]] = None,
+    ) -> None:
+        """No-op."""
+
+    async def async_event_gen(self) -> AsyncGenerator[CallbackEvent, None]:
+        while not self._aqueue.empty() or not self.is_done:
+            try:
+                yield await asyncio.wait_for(self._aqueue.get(), timeout=0.1)
+            except asyncio.TimeoutError:
+                pass

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

@@ -0,0 +1,29 @@
+import json
+from typing import Any
+from fastapi.responses import StreamingResponse
+
+
+class VercelStreamResponse(StreamingResponse):
+    """
+    Class to convert the response from the chat engine to the streaming format expected by Vercel
+    """
+
+    TEXT_PREFIX = "0:"
+    DATA_PREFIX = "8:"
+
+    @classmethod
+    def convert_text(cls, token: str):
+        # Escape newlines and double quotes to avoid breaking the stream
+        token = json.dumps(token)
+        return f"{cls.TEXT_PREFIX}{token}\n"
+
+    @classmethod
+    def convert_data(cls, data: dict):
+        data_str = json.dumps(data)
+        return f"{cls.DATA_PREFIX}[{data_str}]\n"
+
+    def __init__(self, content: Any, **kwargs):
+        super().__init__(
+            content=content,
+            **kwargs,
+        )

templates/types/streaming/fastapi/app/settings.py

@@ -1,41 +1,92 @@
 import os
 from typing import Dict
 from llama_index.core.settings import Settings
-from llama_index.llms.openai import OpenAI
-from llama_index.embeddings.openai import OpenAIEmbedding
-
-
-def llm_config_from_env() -> Dict:
-    from llama_index.core.constants import DEFAULT_TEMPERATURE
-
-    model = os.getenv("MODEL")
-    temperature = os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)
-    max_tokens = os.getenv("LLM_MAX_TOKENS")
-
-    config = {
-        "model": model,
-        "temperature": float(temperature),
-        "max_tokens": int(max_tokens) if max_tokens is not None else None,
-    }
-    return config
-
-
-def embedding_config_from_env() -> Dict:
-    model = os.getenv("EMBEDDING_MODEL")
-    dimension = os.getenv("EMBEDDING_DIM")
-
-    config = {
-        "model": model,
-        "dimension": int(dimension) if dimension is not None else None,
-    }
-    return config
 
 
 def init_settings():
-    llm_configs = llm_config_from_env()
-    embedding_configs = embedding_config_from_env()
-
-    Settings.llm = OpenAI(**llm_configs)
-    Settings.embed_model = OpenAIEmbedding(**embedding_configs)
+    model_provider = os.getenv("MODEL_PROVIDER")
+    if model_provider == "openai":
+        init_openai()
+    elif model_provider == "ollama":
+        init_ollama()
+    elif model_provider == "anthropic":
+        init_anthropic()
+    elif model_provider == "gemini":
+        init_gemini()
+    else:
+        raise ValueError(f"Invalid model provider: {model_provider}")
     Settings.chunk_size = int(os.getenv("CHUNK_SIZE", "1024"))
     Settings.chunk_overlap = int(os.getenv("CHUNK_OVERLAP", "20"))
+
+
+def init_ollama():
+    from llama_index.llms.ollama import Ollama
+    from llama_index.embeddings.ollama import OllamaEmbedding
+
+    Settings.embed_model = OllamaEmbedding(model_name=os.getenv("EMBEDDING_MODEL"))
+    Settings.llm = Ollama(model=os.getenv("MODEL"))
+
+
+def init_openai():
+    from llama_index.llms.openai import OpenAI
+    from llama_index.embeddings.openai import OpenAIEmbedding
+    from llama_index.core.constants import DEFAULT_TEMPERATURE
+
+    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)
+
+    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)
+
+
+def init_anthropic():
+    from llama_index.llms.anthropic import Anthropic
+    from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+
+    model_map: Dict[str, str] = {
+        "claude-3-opus": "claude-3-opus-20240229",
+        "claude-3-sonnet": "claude-3-sonnet-20240229",
+        "claude-3-haiku": "claude-3-haiku-20240307",
+        "claude-2.1": "claude-2.1",
+        "claude-instant-1.2": "claude-instant-1.2",
+    }
+
+    embed_model_map: Dict[str, str] = {
+        "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
+        "all-mpnet-base-v2": "sentence-transformers/all-mpnet-base-v2",
+    }
+
+    Settings.llm = Anthropic(model=model_map[os.getenv("MODEL")])
+    Settings.embed_model = HuggingFaceEmbedding(
+        model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
+    )
+
+
+def init_gemini():
+    from llama_index.llms.gemini import Gemini
+    from llama_index.embeddings.gemini import GeminiEmbedding
+
+    model_map: Dict[str, str] = {
+        "gemini-1.5-pro-latest": "models/gemini-1.5-pro-latest",
+        "gemini-pro": "models/gemini-pro",
+        "gemini-pro-vision": "models/gemini-pro-vision",
+    }
+
+    embed_model_map: Dict[str, str] = {
+        "embedding-001": "models/embedding-001",
+        "text-embedding-004": "models/text-embedding-004",
+    }
+
+    Settings.llm = Gemini(model=model_map[os.getenv("MODEL")])
+    Settings.embed_model = GeminiEmbedding(
+        model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
+    )

templates/types/streaming/fastapi/pyproject.toml

@@ -13,9 +13,9 @@ python = "^3.11,<3.12"
 fastapi = "^0.109.1"
 uvicorn = { extras = ["standard"], version = "^0.23.2" }
 python-dotenv = "^1.0.0"
+aiostream = "^0.5.2"
 llama-index = "0.10.28"
 llama-index-core = "0.10.28"
-llama-index-agent-openai = "0.2.2"
 
 [build-system]
 requires = ["poetry-core"]

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

@@ -1,19 +1,87 @@
-import { OpenAI, OpenAIEmbedding, Settings } from "llamaindex";
+import {
+  Anthropic,
+  GEMINI_EMBEDDING_MODEL,
+  GEMINI_MODEL,
+  Gemini,
+  GeminiEmbedding,
+  OpenAI,
+  OpenAIEmbedding,
+  Settings,
+} from "llamaindex";
+import { HuggingFaceEmbedding } from "llamaindex/embeddings/HuggingFaceEmbedding";
+import { OllamaEmbedding } from "llamaindex/embeddings/OllamaEmbedding";
+import { ALL_AVAILABLE_ANTHROPIC_MODELS } from "llamaindex/llm/anthropic";
+import { Ollama } from "llamaindex/llm/ollama";
 
 const CHUNK_SIZE = 512;
 const CHUNK_OVERLAP = 20;
 
 export const initSettings = async () => {
+  // HINT: you can delete the initialization code for unused model providers
+  console.log(`Using '${process.env.MODEL_PROVIDER}' model provider`);
+
+  if (!process.env.MODEL || !process.env.EMBEDDING_MODEL) {
+    throw new Error("'MODEL' and 'EMBEDDING_MODEL' env variables must be set.");
+  }
+
+  switch (process.env.MODEL_PROVIDER) {
+    case "ollama":
+      initOllama();
+      break;
+    case "anthropic":
+      initAnthropic();
+      break;
+    case "gemini":
+      initGemini();
+      break;
+    default:
+      initOpenAI();
+      break;
+  }
+  Settings.chunkSize = CHUNK_SIZE;
+  Settings.chunkOverlap = CHUNK_OVERLAP;
+};
+
+function initOpenAI() {
   Settings.llm = new OpenAI({
     model: process.env.MODEL ?? "gpt-3.5-turbo",
     maxTokens: 512,
   });
-  Settings.chunkSize = CHUNK_SIZE;
-  Settings.chunkOverlap = CHUNK_OVERLAP;
   Settings.embedModel = new OpenAIEmbedding({
     model: process.env.EMBEDDING_MODEL,
     dimensions: process.env.EMBEDDING_DIM
       ? parseInt(process.env.EMBEDDING_DIM)
       : undefined,
   });
-};
+}
+
+function initOllama() {
+  Settings.llm = new Ollama({
+    model: process.env.MODEL ?? "",
+  });
+  Settings.embedModel = new OllamaEmbedding({
+    model: process.env.EMBEDDING_MODEL ?? "",
+  });
+}
+
+function initAnthropic() {
+  const embedModelMap: Record<string, string> = {
+    "all-MiniLM-L6-v2": "Xenova/all-MiniLM-L6-v2",
+    "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
+  };
+  Settings.llm = new Anthropic({
+    model: process.env.MODEL as keyof typeof ALL_AVAILABLE_ANTHROPIC_MODELS,
+  });
+  Settings.embedModel = new HuggingFaceEmbedding({
+    modelType: embedModelMap[process.env.EMBEDDING_MODEL!],
+  });
+}
+
+function initGemini() {
+  Settings.llm = new Gemini({
+    model: process.env.MODEL as GEMINI_MODEL,
+  });
+  Settings.embedModel = new GeminiEmbedding({
+    model: process.env.EMBEDDING_MODEL as GEMINI_EMBEDDING_MODEL,
+  });
+}

templates/types/streaming/nextjs/app/api/chat/llamaindex-stream.ts

@@ -1,49 +1,65 @@
 import {
-  JSONValue,
+  StreamData,
   createCallbacksTransformer,
   createStreamDataTransformer,
-  experimental_StreamData,
   trimStartOfStreamHelper,
   type AIStreamCallbacksAndOptions,
 } from "ai";
-import { Response, StreamingAgentChatResponse } from "llamaindex";
+import {
+  Metadata,
+  NodeWithScore,
+  Response,
+  ToolCallLLMMessageOptions,
+} from "llamaindex";
+
+import { AgentStreamChatResponse } from "llamaindex/agent/base";
+import { appendImageData, appendSourceData } from "./stream-helper";
+
+type LlamaIndexResponse =
+  | AgentStreamChatResponse<ToolCallLLMMessageOptions>
+  | Response;
 
 type ParserOptions = {
   image_url?: string;
 };
 
 function createParser(
-  res: AsyncIterable<Response>,
-  data: experimental_StreamData,
+  res: AsyncIterable<LlamaIndexResponse>,
+  data: StreamData,
   opts?: ParserOptions,
 ) {
   const it = res[Symbol.asyncIterator]();
   const trimStartOfStream = trimStartOfStreamHelper();
+
+  let sourceNodes: NodeWithScore<Metadata>[] | undefined;
   return new ReadableStream<string>({
     start() {
-      // if image_url is provided, send it via the data stream
-      if (opts?.image_url) {
-        const message: JSONValue = {
-          type: "image_url",
-          image_url: {
-            url: opts.image_url,
-          },
-        };
-        data.append(message);
-      } else {
-        data.append({}); // send an empty image response for the user's message
-      }
+      appendImageData(data, opts?.image_url);
     },
     async pull(controller): Promise<void> {
       const { value, done } = await it.next();
       if (done) {
+        if (sourceNodes) {
+          appendSourceData(data, sourceNodes);
+        }
         controller.close();
-        data.append({}); // send an empty image response for the assistant's message
         data.close();
         return;
       }
 
-      const text = trimStartOfStream(value.response ?? "");
+      let delta;
+      if (value instanceof Response) {
+        // handle Response type
+        if (value.sourceNodes) {
+          // get source nodes from the first response
+          sourceNodes = value.sourceNodes;
+        }
+        delta = value.response ?? "";
+      } else {
+        // handle other types
+        delta = value.response.delta;
+      }
+      const text = trimStartOfStream(delta ?? "");
       if (text) {
         controller.enqueue(text);
       }
@@ -52,21 +68,14 @@ function createParser(
 }
 
 export function LlamaIndexStream(
-  response: StreamingAgentChatResponse | AsyncIterable<Response>,
+  response: AsyncIterable<LlamaIndexResponse>,
+  data: StreamData,
   opts?: {
     callbacks?: AIStreamCallbacksAndOptions;
     parserOptions?: ParserOptions;
   },
-): { stream: ReadableStream; data: experimental_StreamData } {
-  const data = new experimental_StreamData();
-  const res =
-    response instanceof StreamingAgentChatResponse
-      ? response.response
-      : response;
-  return {
-    stream: createParser(res, data, opts?.parserOptions)
-      .pipeThrough(createCallbacksTransformer(opts?.callbacks))
-      .pipeThrough(createStreamDataTransformer(true)),
-    data,
-  };
+): ReadableStream<Uint8Array> {
+  return createParser(response, data, opts?.parserOptions)
+    .pipeThrough(createCallbacksTransformer(opts?.callbacks))
+    .pipeThrough(createStreamDataTransformer());
 }

templates/types/streaming/nextjs/app/api/chat/route.ts

@@ -1,10 +1,11 @@
 import { initObservability } from "@/app/observability";
-import { StreamingTextResponse } from "ai";
-import { ChatMessage, MessageContent } from "llamaindex";
+import { Message, StreamData, StreamingTextResponse } from "ai";
+import { ChatMessage, MessageContent, Settings } from "llamaindex";
 import { NextRequest, NextResponse } from "next/server";
 import { createChatEngine } from "./engine/chat";
 import { initSettings } from "./engine/settings";
 import { LlamaIndexStream } from "./llamaindex-stream";
+import { createCallbackManager } from "./stream-helper";
 
 initObservability();
 initSettings();
@@ -34,7 +35,7 @@ const convertMessageContent = (
 export async function POST(request: NextRequest) {
   try {
     const body = await request.json();
-    const { messages, data }: { messages: ChatMessage[]; data: any } = body;
+    const { messages, data }: { messages: Message[]; data: any } = body;
     const userMessage = messages.pop();
     if (!messages || !userMessage || userMessage.role !== "user") {
       return NextResponse.json(
@@ -54,27 +55,35 @@ export async function POST(request: NextRequest) {
       data?.imageUrl,
     );
 
+    // Init Vercel AI StreamData
+    const vercelStreamData = new StreamData();
+
+    // Setup callbacks
+    const callbackManager = createCallbackManager(vercelStreamData);
+
     // Calling LlamaIndex's ChatEngine to get a streamed response
-    const response = await chatEngine.chat({
-      message: userMessageContent,
-      chatHistory: messages,
-      stream: true,
+    const response = await Settings.withCallbackManager(callbackManager, () => {
+      return chatEngine.chat({
+        message: userMessageContent,
+        chatHistory: messages as ChatMessage[],
+        stream: true,
+      });
     });
 
     // Transform LlamaIndex stream to Vercel/AI format
-    const { stream, data: streamData } = LlamaIndexStream(response, {
+    const stream = LlamaIndexStream(response, vercelStreamData, {
       parserOptions: {
         image_url: data?.imageUrl,
       },
     });
 
     // Return a StreamingTextResponse, which can be consumed by the Vercel/AI client
-    return new StreamingTextResponse(stream, {}, streamData);
+    return new StreamingTextResponse(stream, {}, vercelStreamData);
   } catch (error) {
     console.error("[LlamaIndex]", error);
     return NextResponse.json(
       {
-        error: (error as Error).message,
+        detail: (error as Error).message,
       },
       {
         status: 500,

templates/types/streaming/nextjs/app/api/chat/stream-helper.ts

@@ -0,0 +1,97 @@
+import { StreamData } from "ai";
+import {
+  CallbackManager,
+  Metadata,
+  NodeWithScore,
+  ToolCall,
+  ToolOutput,
+} from "llamaindex";
+
+export function appendImageData(data: StreamData, imageUrl?: string) {
+  if (!imageUrl) return;
+  data.appendMessageAnnotation({
+    type: "image",
+    data: {
+      url: imageUrl,
+    },
+  });
+}
+
+export function appendSourceData(
+  data: StreamData,
+  sourceNodes?: NodeWithScore<Metadata>[],
+) {
+  if (!sourceNodes?.length) return;
+  data.appendMessageAnnotation({
+    type: "sources",
+    data: {
+      nodes: sourceNodes.map((node) => ({
+        ...node.node.toMutableJSON(),
+        id: node.node.id_,
+        score: node.score ?? null,
+      })),
+    },
+  });
+}
+
+export function appendEventData(data: StreamData, title?: string) {
+  if (!title) return;
+  data.appendMessageAnnotation({
+    type: "events",
+    data: {
+      title,
+    },
+  });
+}
+
+export function appendToolData(
+  data: StreamData,
+  toolCall: ToolCall,
+  toolOutput: ToolOutput,
+) {
+  data.appendMessageAnnotation({
+    type: "tools",
+    data: {
+      toolCall: {
+        id: toolCall.id,
+        name: toolCall.name,
+        input: toolCall.input,
+      },
+      toolOutput: {
+        output: toolOutput.output,
+        isError: toolOutput.isError,
+      },
+    },
+  });
+}
+
+export function createCallbackManager(stream: StreamData) {
+  const callbackManager = new CallbackManager();
+
+  callbackManager.on("retrieve", (data) => {
+    const { nodes, query } = data.detail;
+    appendEventData(stream, `Retrieving context for query: '${query}'`);
+    appendEventData(
+      stream,
+      `Retrieved ${nodes.length} sources to use as context for the query`,
+    );
+  });
+
+  callbackManager.on("llm-tool-call", (event) => {
+    const { name, input } = event.detail.payload.toolCall;
+    const inputString = Object.entries(input)
+      .map(([key, value]) => `${key}: ${value}`)
+      .join(", ");
+    appendEventData(
+      stream,
+      `Using tool: '${name}' with inputs: '${inputString}'`,
+    );
+  });
+
+  callbackManager.on("llm-tool-result", (event) => {
+    const { toolCall, toolResult } = event.detail.payload;
+    appendToolData(stream, toolCall, toolResult);
+  });
+
+  return callbackManager;
+}

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

@@ -1,8 +1,6 @@
 "use client";
 
 import { useChat } from "ai/react";
-import { useMemo } from "react";
-import { insertDataIntoMessages } from "./transform";
 import { ChatInput, ChatMessages } from "./ui/chat";
 
 export default function ChatSection() {
@@ -14,22 +12,22 @@ export default function ChatSection() {
     handleInputChange,
     reload,
     stop,
-    data,
   } = useChat({
     api: process.env.NEXT_PUBLIC_CHAT_API,
     headers: {
       "Content-Type": "application/json", // using JSON because of vercel/ai 2.2.26
     },
+    onError: (error: unknown) => {
+      if (!(error instanceof Error)) throw error;
+      const message = JSON.parse(error.message);
+      alert(message.detail);
+    },
   });
 
-  const transformedMessages = useMemo(() => {
-    return insertDataIntoMessages(messages, data);
-  }, [messages, data]);
-
   return (
     <div className="space-y-4 max-w-5xl w-full">
       <ChatMessages
-        messages={transformedMessages}
+        messages={messages}
         isLoading={isLoading}
         reload={reload}
         stop={stop}
@@ -39,7 +37,7 @@ export default function ChatSection() {
         handleSubmit={handleSubmit}
         handleInputChange={handleInputChange}
         isLoading={isLoading}
-        multiModal={process.env.NEXT_PUBLIC_MODEL === "gpt-4-turbo"}
+        multiModal={true}
       />
     </div>
   );

templates/types/streaming/nextjs/app/components/transform.ts

@@ -1,19 +0,0 @@
-import { JSONValue, Message } from "ai";
-
-export const isValidMessageData = (rawData: JSONValue | undefined) => {
-  if (!rawData || typeof rawData !== "object") return false;
-  if (Object.keys(rawData).length === 0) return false;
-  return true;
-};
-
-export const insertDataIntoMessages = (
-  messages: Message[],
-  data: JSONValue[] | undefined,
-) => {
-  if (!data) return messages;
-  messages.forEach((message, i) => {
-    const rawData = data[i];
-    if (isValidMessageData(rawData)) message.data = rawData;
-  });
-  return messages;
-};

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

@@ -0,0 +1,48 @@
+import { ChevronDown, ChevronRight, Loader2 } from "lucide-react";
+import { useState } from "react";
+import { Button } from "../button";
+import {
+  Collapsible,
+  CollapsibleContent,
+  CollapsibleTrigger,
+} from "../collapsible";
+import { EventData } from "./index";
+
+export function ChatEvents({
+  data,
+  isLoading,
+}: {
+  data: EventData[];
+  isLoading: boolean;
+}) {
+  const [isOpen, setIsOpen] = useState(false);
+
+  const buttonLabel = isOpen ? "Hide events" : "Show events";
+
+  const EventIcon = isOpen ? (
+    <ChevronDown className="h-4 w-4" />
+  ) : (
+    <ChevronRight className="h-4 w-4" />
+  );
+
+  return (
+    <div className="border-l-2 border-indigo-400 pl-2">
+      <Collapsible open={isOpen} onOpenChange={setIsOpen}>
+        <CollapsibleTrigger asChild>
+          <Button variant="secondary" className="space-x-2">
+            {isLoading ? <Loader2 className="h-4 w-4 animate-spin" /> : null}
+            <span>{buttonLabel}</span>
+            {EventIcon}
+          </Button>
+        </CollapsibleTrigger>
+        <CollapsibleContent asChild>
+          <div className="mt-4 text-sm space-y-2">
+            {data.map((eventItem, index) => (
+              <div key={index}>{eventItem.title}</div>
+            ))}
+          </div>
+        </CollapsibleContent>
+      </Collapsible>
+    </div>
+  );
+}

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

@@ -0,0 +1,17 @@
+import Image from "next/image";
+import { type ImageData } from "./index";
+
+export function ChatImage({ data }: { data: ImageData }) {
+  return (
+    <div className="rounded-md max-w-[200px] shadow-md">
+      <Image
+        src={data.url}
+        width={0}
+        height={0}
+        sizes="100vw"
+        style={{ width: "100%", height: "auto" }}
+        alt=""
+      />
+    </div>
+  );
+}

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

@@ -1,51 +1,114 @@
 import { Check, Copy } from "lucide-react";
 
-import { JSONValue, Message } from "ai";
-import Image from "next/image";
+import { Message } from "ai";
+import { Fragment } from "react";
 import { Button } from "../button";
 import ChatAvatar from "./chat-avatar";
+import { ChatEvents } from "./chat-events";
+import { ChatImage } from "./chat-image";
+import { ChatSources } from "./chat-sources";
+import ChatTools from "./chat-tools";
+import {
+  AnnotationData,
+  EventData,
+  ImageData,
+  MessageAnnotation,
+  MessageAnnotationType,
+  SourceData,
+  ToolData,
+} from "./index";
 import Markdown from "./markdown";
 import { useCopyToClipboard } from "./use-copy-to-clipboard";
 
-interface ChatMessageImageData {
-  type: "image_url";
-  image_url: {
-    url: string;
-  };
+type ContentDisplayConfig = {
+  order: number;
+  component: JSX.Element | null;
+};
+
+function getAnnotationData<T extends AnnotationData>(
+  annotations: MessageAnnotation[],
+  type: MessageAnnotationType,
+): T[] {
+  return annotations.filter((a) => a.type === type).map((a) => a.data as T);
 }
 
-// This component will parse message data and render the appropriate UI.
-function ChatMessageData({ messageData }: { messageData: JSONValue }) {
-  const { image_url, type } = messageData as unknown as ChatMessageImageData;
-  if (type === "image_url") {
-    return (
-      <div className="rounded-md max-w-[200px] shadow-md">
-        <Image
-          src={image_url.url}
-          width={0}
-          height={0}
-          sizes="100vw"
-          style={{ width: "100%", height: "auto" }}
-          alt=""
-        />
-      </div>
-    );
-  }
-  return null;
+function ChatMessageContent({
+  message,
+  isLoading,
+}: {
+  message: Message;
+  isLoading: boolean;
+}) {
+  const annotations = message.annotations as MessageAnnotation[] | undefined;
+  if (!annotations?.length) return <Markdown content={message.content} />;
+
+  const imageData = getAnnotationData<ImageData>(
+    annotations,
+    MessageAnnotationType.IMAGE,
+  );
+  const eventData = getAnnotationData<EventData>(
+    annotations,
+    MessageAnnotationType.EVENTS,
+  );
+  const sourceData = getAnnotationData<SourceData>(
+    annotations,
+    MessageAnnotationType.SOURCES,
+  );
+  const toolData = getAnnotationData<ToolData>(
+    annotations,
+    MessageAnnotationType.TOOLS,
+  );
+
+  const contents: ContentDisplayConfig[] = [
+    {
+      order: -3,
+      component: imageData[0] ? <ChatImage data={imageData[0]} /> : null,
+    },
+    {
+      order: -2,
+      component:
+        eventData.length > 0 ? (
+          <ChatEvents isLoading={isLoading} data={eventData} />
+        ) : null,
+    },
+    {
+      order: -1,
+      component: toolData[0] ? <ChatTools data={toolData[0]} /> : null,
+    },
+    {
+      order: 0,
+      component: <Markdown content={message.content} />,
+    },
+    {
+      order: 1,
+      component: sourceData[0] ? <ChatSources data={sourceData[0]} /> : null,
+    },
+  ];
+
+  return (
+    <div className="flex-1 gap-4 flex flex-col">
+      {contents
+        .sort((a, b) => a.order - b.order)
+        .map((content, index) => (
+          <Fragment key={index}>{content.component}</Fragment>
+        ))}
+    </div>
+  );
 }
 
-export default function ChatMessage(chatMessage: Message) {
+export default function ChatMessage({
+  chatMessage,
+  isLoading,
+}: {
+  chatMessage: Message;
+  isLoading: boolean;
+}) {
   const { isCopied, copyToClipboard } = useCopyToClipboard({ timeout: 2000 });
   return (
     <div className="flex items-start gap-4 pr-5 pt-5">
       <ChatAvatar role={chatMessage.role} />
       <div className="group flex flex-1 justify-between gap-2">
-        <div className="flex-1 space-y-4">
-          {chatMessage.data && (
-            <ChatMessageData messageData={chatMessage.data} />
-          )}
-          <Markdown content={chatMessage.content} />
-        </div>
+        <ChatMessageContent message={chatMessage} isLoading={isLoading} />
         <Button
           onClick={() => copyToClipboard(chatMessage.content)}
           size="icon"

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

@@ -40,9 +40,16 @@ export default function ChatMessages(
         className="flex h-[50vh] flex-col gap-5 divide-y overflow-y-auto pb-4"
         ref={scrollableChatContainerRef}
       >
-        {props.messages.map((m) => (
-          <ChatMessage key={m.id} {...m} />
-        ))}
+        {props.messages.map((m, i) => {
+          const isLoadingMessage = i === messageLength - 1 && props.isLoading;
+          return (
+            <ChatMessage
+              key={m.id}
+              chatMessage={m}
+              isLoading={isLoadingMessage}
+            />
+          );
+        })}
         {isPending && (
           <div className="flex justify-center items-center pt-10">
             <Loader2 className="h-4 w-4 animate-spin" />

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

@@ -0,0 +1,94 @@
+import { ArrowUpRightSquare, Check, Copy } from "lucide-react";
+import { useMemo } from "react";
+import { Button } from "../button";
+import { HoverCard, HoverCardContent, HoverCardTrigger } from "../hover-card";
+import { SourceData, SourceNode } from "./index";
+import { useCopyToClipboard } from "./use-copy-to-clipboard";
+
+const SCORE_THRESHOLD = 0.5;
+
+export function ChatSources({ data }: { data: SourceData }) {
+  const sources = useMemo(() => {
+    return (
+      data.nodes
+        ?.filter((node) => Object.keys(node.metadata).length > 0)
+        ?.filter((node) => (node.score ?? 1) > SCORE_THRESHOLD)
+        .sort((a, b) => (b.score ?? 1) - (a.score ?? 1)) || []
+    );
+  }, [data.nodes]);
+
+  if (sources.length === 0) return null;
+
+  return (
+    <div className="space-x-2 text-sm">
+      <span className="font-semibold">Sources:</span>
+      <div className="inline-flex gap-1 items-center">
+        {sources.map((node: SourceNode, index: number) => (
+          <div key={node.id}>
+            <HoverCard>
+              <HoverCardTrigger>
+                <div className="text-xs w-5 h-5 rounded-full bg-gray-100 mb-2 flex items-center justify-center hover:text-white hover:bg-primary hover:cursor-pointer">
+                  {index + 1}
+                </div>
+              </HoverCardTrigger>
+              <HoverCardContent>
+                <NodeInfo node={node} />
+              </HoverCardContent>
+            </HoverCard>
+          </div>
+        ))}
+      </div>
+    </div>
+  );
+}
+
+function NodeInfo({ node }: { node: SourceNode }) {
+  const { isCopied, copyToClipboard } = useCopyToClipboard({ timeout: 1000 });
+
+  if (typeof node.metadata["URL"] === "string") {
+    // this is a node generated by the web loader, it contains an external URL
+    // add a link to view this URL
+    return (
+      <a
+        className="space-x-2 flex items-center my-2 hover:text-blue-900"
+        href={node.metadata["URL"]}
+        target="_blank"
+      >
+        <span>{node.metadata["URL"]}</span>
+        <ArrowUpRightSquare className="w-4 h-4" />
+      </a>
+    );
+  }
+
+  if (typeof node.metadata["file_path"] === "string") {
+    // this is a node generated by the file loader, it contains file path
+    // add a button to copy the path to the clipboard
+    const filePath = node.metadata["file_path"];
+    return (
+      <div className="flex items-center px-2 py-1 justify-between my-2">
+        <span>{filePath}</span>
+        <Button
+          onClick={() => copyToClipboard(filePath)}
+          size="icon"
+          variant="ghost"
+          className="h-12 w-12"
+        >
+          {isCopied ? (
+            <Check className="h-4 w-4" />
+          ) : (
+            <Copy className="h-4 w-4" />
+          )}
+        </Button>
+      </div>
+    );
+  }
+
+  // node generated by unknown loader, implement renderer by analyzing logged out metadata
+  console.log("Node metadata", node.metadata);
+  return (
+    <p>
+      Sorry, unknown node type. Please add a new renderer in the NodeInfo
+      component.
+    </p>
+  );
+}

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

@@ -0,0 +1,26 @@
+import { ToolData } from "./index";
+import { WeatherCard, WeatherData } from "./widgets/WeatherCard";
+
+// TODO: If needed, add displaying more tool outputs here
+export default function ChatTools({ data }: { data: ToolData }) {
+  if (!data) return null;
+  const { toolCall, toolOutput } = data;
+
+  if (toolOutput.isError) {
+    return (
+      <div className="border-l-2 border-red-400 pl-2">
+        There was an error when calling the tool {toolCall.name} with input:{" "}
+        <br />
+        {JSON.stringify(toolCall.input)}
+      </div>
+    );
+  }
+
+  switch (toolCall.name) {
+    case "get_weather_information":
+      const weatherData = toolOutput.output as unknown as WeatherData;
+      return <WeatherCard data={weatherData} />;
+    default:
+      return null;
+  }
+}

templates/types/streaming/nextjs/app/components/ui/chat/index.ts

@@ -1,5 +1,54 @@
+import { JSONValue } from "ai";
 import ChatInput from "./chat-input";
 import ChatMessages from "./chat-messages";
 
 export { type ChatHandler } from "./chat.interface";
 export { ChatInput, ChatMessages };
+
+export enum MessageAnnotationType {
+  IMAGE = "image",
+  SOURCES = "sources",
+  EVENTS = "events",
+  TOOLS = "tools",
+}
+
+export type ImageData = {
+  url: string;
+};
+
+export type SourceNode = {
+  id: string;
+  metadata: Record<string, unknown>;
+  score?: number;
+  text: string;
+};
+
+export type SourceData = {
+  nodes: SourceNode[];
+};
+
+export type EventData = {
+  title: string;
+  isCollapsed: boolean;
+};
+
+export type ToolData = {
+  toolCall: {
+    id: string;
+    name: string;
+    input: {
+      [key: string]: JSONValue;
+    };
+  };
+  toolOutput: {
+    output: JSONValue;
+    isError: boolean;
+  };
+};
+
+export type AnnotationData = ImageData | SourceData | EventData | ToolData;
+
+export type MessageAnnotation = {
+  type: MessageAnnotationType;
+  data: AnnotationData;
+};

templates/types/streaming/nextjs/app/components/ui/chat/widgets/WeatherCard.tsx

@@ -0,0 +1,213 @@
+export interface WeatherData {
+  latitude: number;
+  longitude: number;
+  generationtime_ms: number;
+  utc_offset_seconds: number;
+  timezone: string;
+  timezone_abbreviation: string;
+  elevation: number;
+  current_units: {
+    time: string;
+    interval: string;
+    temperature_2m: string;
+    weather_code: string;
+  };
+  current: {
+    time: string;
+    interval: number;
+    temperature_2m: number;
+    weather_code: number;
+  };
+  hourly_units: {
+    time: string;
+    temperature_2m: string;
+    weather_code: string;
+  };
+  hourly: {
+    time: string[];
+    temperature_2m: number[];
+    weather_code: number[];
+  };
+  daily_units: {
+    time: string;
+    weather_code: string;
+  };
+  daily: {
+    time: string[];
+    weather_code: number[];
+  };
+}
+
+// Follow WMO Weather interpretation codes (WW)
+const weatherCodeDisplayMap: Record<
+  string,
+  {
+    icon: JSX.Element;
+    status: string;
+  }
+> = {
+  "0": {
+    icon: <span>☀️</span>,
+    status: "Clear sky",
+  },
+  "1": {
+    icon: <span>🌤️</span>,
+    status: "Mainly clear",
+  },
+  "2": {
+    icon: <span>☁️</span>,
+    status: "Partly cloudy",
+  },
+  "3": {
+    icon: <span>☁️</span>,
+    status: "Overcast",
+  },
+  "45": {
+    icon: <span>🌫️</span>,
+    status: "Fog",
+  },
+  "48": {
+    icon: <span>🌫️</span>,
+    status: "Depositing rime fog",
+  },
+  "51": {
+    icon: <span>🌧️</span>,
+    status: "Drizzle",
+  },
+  "53": {
+    icon: <span>🌧️</span>,
+    status: "Drizzle",
+  },
+  "55": {
+    icon: <span>🌧️</span>,
+    status: "Drizzle",
+  },
+  "56": {
+    icon: <span>🌧️</span>,
+    status: "Freezing Drizzle",
+  },
+  "57": {
+    icon: <span>🌧️</span>,
+    status: "Freezing Drizzle",
+  },
+  "61": {
+    icon: <span>🌧️</span>,
+    status: "Rain",
+  },
+  "63": {
+    icon: <span>🌧️</span>,
+    status: "Rain",
+  },
+  "65": {
+    icon: <span>🌧️</span>,
+    status: "Rain",
+  },
+  "66": {
+    icon: <span>🌧️</span>,
+    status: "Freezing Rain",
+  },
+  "67": {
+    icon: <span>🌧️</span>,
+    status: "Freezing Rain",
+  },
+  "71": {
+    icon: <span>❄️</span>,
+    status: "Snow fall",
+  },
+  "73": {
+    icon: <span>❄️</span>,
+    status: "Snow fall",
+  },
+  "75": {
+    icon: <span>❄️</span>,
+    status: "Snow fall",
+  },
+  "77": {
+    icon: <span>❄️</span>,
+    status: "Snow grains",
+  },
+  "80": {
+    icon: <span>🌧️</span>,
+    status: "Rain showers",
+  },
+  "81": {
+    icon: <span>🌧️</span>,
+    status: "Rain showers",
+  },
+  "82": {
+    icon: <span>🌧️</span>,
+    status: "Rain showers",
+  },
+  "85": {
+    icon: <span>❄️</span>,
+    status: "Snow showers",
+  },
+  "86": {
+    icon: <span>❄️</span>,
+    status: "Snow showers",
+  },
+  "95": {
+    icon: <span>⛈️</span>,
+    status: "Thunderstorm",
+  },
+  "96": {
+    icon: <span>⛈️</span>,
+    status: "Thunderstorm",
+  },
+  "99": {
+    icon: <span>⛈️</span>,
+    status: "Thunderstorm",
+  },
+};
+
+const displayDay = (time: string) => {
+  return new Date(time).toLocaleDateString("en-US", {
+    weekday: "long",
+  });
+};
+
+export function WeatherCard({ data }: { data: WeatherData }) {
+  const currentDayString = new Date(data.current.time).toLocaleDateString(
+    "en-US",
+    {
+      weekday: "long",
+      month: "long",
+      day: "numeric",
+    },
+  );
+
+  return (
+    <div className="bg-[#61B9F2] rounded-2xl shadow-xl p-5 space-y-4 text-white w-fit">
+      <div className="flex justify-between">
+        <div className="space-y-2">
+          <div className="text-xl">{currentDayString}</div>
+          <div className="text-5xl font-semibold flex gap-4">
+            <span>
+              {data.current.temperature_2m} {data.current_units.temperature_2m}
+            </span>
+            {weatherCodeDisplayMap[data.current.weather_code].icon}
+          </div>
+        </div>
+        <span className="text-xl">
+          {weatherCodeDisplayMap[data.current.weather_code].status}
+        </span>
+      </div>
+      <div className="gap-2 grid grid-cols-6">
+        {data.daily.time.map((time, index) => {
+          if (index === 0) return null; // skip the current day
+          return (
+            <div key={time} className="flex flex-col items-center gap-4">
+              <span>{displayDay(time)}</span>
+              <div className="text-4xl">
+                {weatherCodeDisplayMap[data.daily.weather_code[index]].icon}
+              </div>
+              <span className="text-sm">
+                {weatherCodeDisplayMap[data.daily.weather_code[index]].status}
+              </span>
+            </div>
+          );
+        })}
+      </div>
+    </div>
+  );
+}

templates/types/streaming/nextjs/app/components/ui/collapsible.tsx

@@ -0,0 +1,11 @@
+"use client";
+
+import * as CollapsiblePrimitive from "@radix-ui/react-collapsible";
+
+const Collapsible = CollapsiblePrimitive.Root;
+
+const CollapsibleTrigger = CollapsiblePrimitive.CollapsibleTrigger;
+
+const CollapsibleContent = CollapsiblePrimitive.CollapsibleContent;
+
+export { Collapsible, CollapsibleContent, CollapsibleTrigger };

templates/types/streaming/nextjs/app/components/ui/hover-card.tsx

@@ -0,0 +1,29 @@
+"use client";
+
+import * as HoverCardPrimitive from "@radix-ui/react-hover-card";
+import * as React from "react";
+
+import { cn } from "./lib/utils";
+
+const HoverCard = HoverCardPrimitive.Root;
+
+const HoverCardTrigger = HoverCardPrimitive.Trigger;
+
+const HoverCardContent = React.forwardRef<
+  React.ElementRef<typeof HoverCardPrimitive.Content>,
+  React.ComponentPropsWithoutRef<typeof HoverCardPrimitive.Content>
+>(({ className, align = "center", sideOffset = 4, ...props }, ref) => (
+  <HoverCardPrimitive.Content
+    ref={ref}
+    align={align}
+    sideOffset={sideOffset}
+    className={cn(
+      "z-50 w-64 rounded-md border bg-popover p-4 text-popover-foreground shadow-md outline-none data-[state=open]:animate-in data-[state=closed]:animate-out data-[state=closed]:fade-out-0 data-[state=open]:fade-in-0 data-[state=closed]:zoom-out-95 data-[state=open]:zoom-in-95 data-[side=bottom]:slide-in-from-top-2 data-[side=left]:slide-in-from-right-2 data-[side=right]:slide-in-from-left-2 data-[side=top]:slide-in-from-bottom-2",
+      className,
+    )}
+    {...props}
+  />
+));
+HoverCardContent.displayName = HoverCardPrimitive.Content.displayName;
+
+export { HoverCard, HoverCardContent, HoverCardTrigger };

templates/types/streaming/nextjs/next.config.json

@@ -2,6 +2,7 @@
   "experimental": {
     "outputFileTracingIncludes": {
       "/*": ["./cache/**/*"]
-    }
+    },
+    "serverComponentsExternalPackages": ["sharp", "onnxruntime-node"]
   }
 }

templates/types/streaming/nextjs/package.json

@@ -10,14 +10,18 @@
     "lint": "next lint"
   },
   "dependencies": {
+    "@radix-ui/react-collapsible": "^1.0.3",
+    "@radix-ui/react-hover-card": "^1.0.7",
     "@radix-ui/react-slot": "^1.0.2",
-    "ai": "^2.2.27",
+    "ai": "^3.0.21",
+    "ajv": "^8.12.0",
     "class-variance-authority": "^0.7.0",
-    "clsx": "^1.2.1",
+    "clsx": "^2.1.1",
     "dotenv": "^16.3.1",
-    "llamaindex": "latest",
+    "llamaindex": "0.3.9",
     "lucide-react": "^0.294.0",
     "next": "^14.0.3",
+    "pdf2json": "3.0.5",
     "react": "^18.2.0",
     "react-dom": "^18.2.0",
     "react-markdown": "^8.0.7",
@@ -33,17 +37,17 @@
     "@types/node": "^20.10.3",
     "@types/react": "^18.2.42",
     "@types/react-dom": "^18.2.17",
+    "@types/react-syntax-highlighter": "^15.5.11",
     "autoprefixer": "^10.4.16",
+    "cross-env": "^7.0.3",
     "eslint": "^8.55.0",
     "eslint-config-next": "^14.0.3",
+    "eslint-config-prettier": "^8.10.0",
     "postcss": "^8.4.32",
-    "tailwindcss": "^3.3.6",
-    "typescript": "^5.3.2",
-    "@types/react-syntax-highlighter": "^15.5.11",
-    "cross-env": "^7.0.3",
     "prettier": "^3.2.5",
     "prettier-plugin-organize-imports": "^3.2.4",
-    "eslint-config-prettier": "^8.10.0",
-    "ts-node": "^10.9.2"
+    "tailwindcss": "^3.3.6",
+    "tsx": "^4.7.2",
+    "typescript": "^5.3.2"
   }
 }

templates/types/streaming/nextjs/tsconfig.json

@@ -24,10 +24,5 @@
     "forceConsistentCasingInFileNames": true
   },
   "include": ["next-env.d.ts", "**/*.ts", "**/*.tsx", ".next/types/**/*.ts"],
-  "exclude": ["node_modules"],
-  "ts-node": {
-    "compilerOptions": {
-      "module": "commonjs"
-    }
-  }
+  "exclude": ["node_modules"]
 }

templates/types/streaming/nextjs/webpack.config.mjs

@@ -1,10 +1,15 @@
 // webpack config must be a function in NextJS that is used to patch the default webpack config provided by NextJS, see https://nextjs.org/docs/pages/api-reference/next-config-js/webpack
 export default function webpack(config) {
-  // See https://webpack.js.org/configuration/resolve/#resolvealias
-  config.resolve.alias = {
-    ...config.resolve.alias,
-    sharp$: false,
-    "onnxruntime-node$": false,
+  config.resolve.fallback = {
+    aws4: false,
   };
+
+  // Following lines will fix issues with onnxruntime-node when using pnpm
+  // See: https://github.com/vercel/next.js/issues/43433
+  config.externals.push({
+    "onnxruntime-node": "commonjs onnxruntime-node",
+    sharp: "commonjs sharp",
+  });
+
   return config;
 }

templates/types/streaming/nextjs/webpack.config.o11y.mjs

@@ -1,9 +1,6 @@
 export default function webpack(config, isServer) {
-  // See https://webpack.js.org/configuration/resolve/#resolvealias
-  config.resolve.alias = {
-    ...config.resolve.alias,
-    sharp$: false,
-    "onnxruntime-node$": false,
+  config.resolve.fallback = {
+    aws4: false,
   };
   config.module.rules.push({
     test: /\.node$/,