docs: updated docs for retriever using param obj

Releases:
  llamaindex@0.1.21

[skip ci]

.changeset/cuddly-cars-doubt.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Add LlamaParse option when selecting a pdf file or a folder

.changeset/eight-dogs-burn.md

@@ -0,0 +1,12 @@
+---
+"llamaindex": patch
+"@llamaindex/core-test": patch
+---
+
+- Add missing exports:
+  - `IndexStructType`,
+  - `IndexDict`,
+  - `jsonToIndexStruct`,
+  - `IndexList`,
+  - `IndexStruct`
+- Fix `IndexDict.toJson()` method

.changeset/famous-pugs-join.md

@@ -1,5 +0,0 @@
----
-"llamaindex": patch
----
-
-feat: add base evaluator and correctness evaluator

.changeset/khaki-rivers-unite.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+Add streaming to agents

.changeset/pretty-rings-press.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Add embedding model option to create-llama

.changeset/sharp-ducks-clap.md

@@ -1,5 +0,0 @@
----
-"llamaindex": patch
----
-
-feat: add base evaluator and correctness evaluator

.changeset/strange-waves-sit.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": minor
+---
+
+Use parameter object for retrieve function of Retriever (to align usage with query function of QueryEngine)

.changeset/swift-waves-divide.md

@@ -1,6 +0,0 @@
----
-"llamaindex": patch
-"docs": patch
----
-
-Add Groq LLM to LlamaIndex

.eslintrc.js

@@ -11,5 +11,13 @@ module.exports = {
     "max-params": ["error", 4],
     "prefer-const": "error",
   },
+  overrides: [
+    {
+      files: ["examples/**/*.ts"],
+      rules: {
+        "turbo/no-undeclared-env-vars": "off",
+      },
+    },
+  ],
   ignorePatterns: ["dist/", "lib/"],
 };

.github/workflows/publish.yml

@@ -0,0 +1,24 @@
+name: Publish
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Publish @llamaindex/env
+        run: npx jsr publish
+        working-directory: packages/env
+
+      - name: Publish @llamaindex/core
+        run: npx jsr publish --allow-slow-types
+        working-directory: packages/core

.gitignore

@@ -44,6 +44,7 @@ test-results/
 playwright-report/
 blob-report/
 playwright/.cache/
+.tsbuildinfo
 
 # intellij
 **/.idea

apps/docs/CHANGELOG.md

@@ -1,5 +1,20 @@
 # docs
 
+## 0.0.4
+
+### Patch Changes
+
+- Updated dependencies [5116ad8]
+  - @llamaindex/env@0.0.5
+
+## 0.0.3
+
+### Patch Changes
+
+- 09bf27a: Add Groq LLM to LlamaIndex
+- Updated dependencies [cf87f84]
+  - @llamaindex/env@0.0.4
+
 ## 0.0.2
 
 ### Patch Changes

apps/docs/docs/getting_started/starter.md

@@ -1,56 +0,0 @@
----
-sidebar_position: 1
----
-
-# Starter Tutorial
-
-Once you have [installed LlamaIndex.TS using NPM](installation) and set up your OpenAI key, you're ready to start your first app:
-
-In a new folder:
-
-```bash npm2yarn
-npm install typescript
-npm install @types/node
-npx tsc --init # if needed
-```
-
-Create the file `example.ts`. This code will load some example data, create a document, index it (which creates embeddings using OpenAI), and then creates query engine to answer questions about the data.
-
-```ts
-// example.ts
-import fs from "fs/promises";
-import { Document, VectorStoreIndex } from "llamaindex";
-
-async function main() {
-  // Load essay from abramov.txt in Node
-  const essay = await fs.readFile(
-    "node_modules/llamaindex/examples/abramov.txt",
-    "utf-8",
-  );
-
-  // Create Document object with essay
-  const document = new Document({ text: essay });
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  const index = await VectorStoreIndex.fromDocuments([document]);
-
-  // Query the index
-  const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query({
-    query: "What did the author do in college?",
-  });
-
-  // Output response
-  console.log(response.toString());
-}
-
-main();
-```
-
-Then you can run it using
-
-```bash
-npx ts-node example.ts
-```
-
-Ready to learn more? Check out our NextJS playground at https://llama-playground.vercel.app/. The source is available at https://github.com/run-llama/ts-playground

apps/docs/docs/getting_started/starter.mdx

@@ -0,0 +1,51 @@
+---
+sidebar_position: 1
+---
+
+import CodeBlock from "@theme/CodeBlock";
+import CodeSource from "!raw-loader!../../../../examples/vectorIndex";
+import TSConfigSource from "!!raw-loader!../../../../examples/tsconfig.json";
+
+# Starter Tutorial
+
+Make sure you have installed LlamaIndex.TS and have an OpenAI key. If you haven't, check out the [installation](installation) guide.
+
+## From scratch(node.js + TypeScript):
+
+In a new folder:
+
+```bash npm2yarn
+npm init
+npm install -D typescript @types/node
+```
+
+Create the file `example.ts`. This code will load some example data, create a document, index it (which creates embeddings using OpenAI), and then creates query engine to answer questions about the data.
+
+<CodeBlock language="ts">{CodeSource}</CodeBlock>
+
+Create a `tsconfig.json` file in the same folder:
+
+<CodeBlock language="json">{TSConfigSource}</CodeBlock>
+
+Now you can run the code with
+
+```bash
+npx tsx example.ts
+```
+
+Also, you can clone our examples and try them out:
+
+```bash npm2yarn
+npx degit run-llama/LlamaIndexTS/examples my-new-project
+cd my-new-project
+npm install
+npx tsx ./vectorIndex.ts
+```
+
+## From scratch (Next.js + TypeScript):
+
+You just need one command to create a new Next.js project:
+
+```bash npm2yarn
+npx create-llama@latest
+```

apps/docs/docs/introduction.md

@@ -37,7 +37,7 @@ For more complex applications, our lower-level APIs allow advanced users to cust
 
 `npm install llamaindex`
 
-Our documentation includes [Installation Instructions](./getting_started/installation.mdx) and a [Starter Tutorial](./getting_started/starter.md) to build your first application.
+Our documentation includes [Installation Instructions](./getting_started/installation.mdx) and a [Starter Tutorial](./getting_started/starter.mdx) to build your first application.
 
 Once you're up and running, [High-Level Concepts](./getting_started/concepts.md) has an overview of LlamaIndex's modular architecture. For more hands-on practical examples, look through our Examples section on the sidebar.
 

apps/docs/docs/modules/embeddings/available_embeddings/huggingface.md

@@ -23,3 +23,15 @@ const results = await queryEngine.query({
   query,
 });
 ```
+
+Per default, `HuggingFaceEmbedding` is using the `Xenova/all-MiniLM-L6-v2` model. You can change the model by passing the `modelType` parameter to the constructor.
+If you're not using a quantized model, set the `quantized` parameter to `false`.
+
+For example, to use the not quantized `BAAI/bge-small-en-v1.5` model, you can use the following code:
+
+```
+const embedModel = new HuggingFaceEmbedding({
+  modelType: "BAAI/bge-small-en-v1.5",
+  quantized: false,
+});
+```

apps/docs/docs/modules/evaluation/index.md

@@ -27,6 +27,6 @@ These evaluation modules are in the following forms:
 
 ## Usage
 
-- [Correctness Evaluator](correctness.md)
-- [Faithfulness Evaluator](faithfulness.md)
-- [Relevancy Evaluator](relevancy.md)
+- [Correctness Evaluator](./modules/correctness.md)
+- [Faithfulness Evaluator](./modules/faithfulness.md)
+- [Relevancy Evaluator](./modules/relevancy.md)

apps/docs/docs/modules/evaluation/modules/correctness.md

@@ -53,10 +53,6 @@ const evaluator = new CorrectnessEvaluator({
   serviceContext: ctx,
 });
 
-const response = await queryEngine.query({
-  query,
-});
-
 const result = await evaluator.evaluateResponse({
   query,
   response,

apps/docs/docs/modules/node_postprocessors/index.md

@@ -100,7 +100,7 @@ const response = await queryEngine.query("<user_query>");
 ```ts
 import { SimilarityPostprocessor } from "llamaindex";
 
-nodes = await index.asRetriever().retrieve("test query str");
+nodes = await index.asRetriever().retrieve({ query: "test query str" });
 
 const processor = new SimilarityPostprocessor({
   similarityCutoff: 0.7,

apps/docs/docs/modules/retriever.md

@@ -11,7 +11,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Fetch nodes!
-const nodesWithScore = await retriever.retrieve("query string");
+const nodesWithScore = await retriever.retrieve({ query: "query string" });
 ```
 
 ## API Reference

apps/docs/i18n/ar/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // جلب العقد!
-const nodesWithScore = await retriever.retrieve("سلسلة الاستعلام");
+const nodesWithScore = await retriever.retrieve({ query: "سلسلة الاستعلام" });
 ```
 
 ## مرجع الواجهة البرمجية (API Reference)

apps/docs/i18n/bg/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Извличане на върхове!
-const nodesWithScore = await retriever.retrieve("query string");
+const nodesWithScore = await retriever.retrieve({ query: "query string" });
 ```
 
 ## API Reference (API справка)

apps/docs/i18n/cat/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const recuperador = vector_index.asRetriever();
 recuperador.similarityTopK = 3;
 
 // Obteniu els nodes!
-const nodesAmbPuntuació = await recuperador.retrieve("cadena de consulta");
+const nodesAmbPuntuació = await recuperador.retrieve({ query: "cadena de consulta" });
 ```
 
 ## Referència de l'API

apps/docs/i18n/cs/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Získání uzlů!
-const nodesWithScore = await retriever.retrieve("dotazovací řetězec");
+const nodesWithScore = await retriever.retrieve({ query: "dotazovací řetězec" });
 ```
 
 ## API Reference (Odkazy na rozhraní)

apps/docs/i18n/da/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Hent noder!
-const nodesWithScore = await retriever.retrieve("forespørgselsstreng");
+const nodesWithScore = await retriever.retrieve({ query: "forespørgselsstreng" });
 ```
 
 ## API Reference

apps/docs/i18n/de/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Knoten abrufen!
-const nodesWithScore = await retriever.retrieve("Abfragezeichenfolge");
+const nodesWithScore = await retriever.retrieve({ query: "Abfragezeichenfolge" });
 ```
 
 ## API-Referenz

apps/docs/i18n/el/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Ανάκτηση κόμβων!
-const nodesWithScore = await retriever.retrieve("συμβολοσειρά ερωτήματος");
+const nodesWithScore = await retriever.retrieve({ query: "συμβολοσειρά ερωτήματος" });
 ```
 
 ## Αναφορά API

apps/docs/i18n/es/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const recuperador = vector_index.asRetriever();
 recuperador.similarityTopK = 3;
 
 // ¡Obtener nodos!
-const nodosConPuntuación = await recuperador.retrieve("cadena de consulta");
+const nodosConPuntuación = await recuperador.retrieve({ query: "cadena de consulta" });
 ```
 
 ## Referencia de la API

apps/docs/i18n/et/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Too sõlmed!
-const nodesWithScore = await retriever.retrieve("päringu string");
+const nodesWithScore = await retriever.retrieve({ query: "päringu string" });
 ```
 
 ## API viide

apps/docs/i18n/fa/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // بازیابی گره ها!
-const nodesWithScore = await retriever.retrieve("رشته پرس و جو");
+const nodesWithScore = await retriever.retrieve({ query: "رشته پرس و جو" });
 ```
 
 ## مرجع API

apps/docs/i18n/fi/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Hae solmut!
-const nodesWithScore = await retriever.retrieve("kyselymerkkijono");
+const nodesWithScore = await retriever.retrieve({ query: "kyselymerkkijono" });
 ```
 
 ## API-viite

apps/docs/i18n/fr/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -11,7 +11,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Récupérer les nœuds !
-const nodesWithScore = await retriever.retrieve("chaîne de requête");
+const nodesWithScore = await retriever.retrieve({ query: "chaîne de requête" });
 ```
 
 ## Référence de l'API

apps/docs/i18n/he/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // אחזור צמתים!
-const nodesWithScore = await retriever.retrieve("מחרוזת שאילתה");
+const nodesWithScore = await retriever.retrieve({ query: "מחרוזת שאילתה" });
 ```
 
 ## מדריך לממשק API

apps/docs/i18n/hi/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // नोड्स प्राप्त करें!
-const nodesWithScore = await retriever.retrieve("क्वेरी स्ट्रिंग");
+const nodesWithScore = await retriever.retrieve({ query: "क्वेरी स्ट्रिंग" });
 ```
 
 ## एपीआई संदर्भ (API Reference)

apps/docs/i18n/hr/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const dohvatnik = vector_index.asRetriever();
 dohvatnik.similarityTopK = 3;
 
 // Dohvati čvorove!
-const čvoroviSaRezultatom = await dohvatnik.retrieve("upitni niz");
+const čvoroviSaRezultatom = await dohvatnik.retrieve({ query: "upitni niz" });
 ```
 
 ## API Referenca

apps/docs/i18n/hu/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Node-ok lekérése!
-const nodesWithScore = await retriever.retrieve("lekérdezési karakterlánc");
+const nodesWithScore = await retriever.retrieve({ query: "lekérdezési karakterlánc" });
 ```
 
 ## API Referencia

apps/docs/i18n/in/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Mengambil node!
-const nodesWithScore = await retriever.retrieve("string query");
+const nodesWithScore = await retriever.retrieve({ query: "string query" });
 ```
 
 ## Referensi API

apps/docs/i18n/it/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Recupera i nodi!
-const nodesWithScore = await retriever.retrieve("stringa di query");
+const nodesWithScore = await retriever.retrieve({ query: "stringa di query" });
 ```
 
 ## Riferimento API

apps/docs/i18n/ja/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // ノードを取得します！
-const nodesWithScore = await retriever.retrieve("クエリ文字列");
+const nodesWithScore = await retriever.retrieve({ query: "クエリ文字列" });
 ```
 
 ## API リファレンス

apps/docs/i18n/ko/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // 노드를 가져옵니다!
-const nodesWithScore = await retriever.retrieve("쿼리 문자열");
+const nodesWithScore = await retriever.retrieve({ query: "쿼리 문자열" });
 ```
 
 ## API 참조

apps/docs/i18n/lt/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const gavėjas = vector_index.asRetriever();
 gavėjas.similarityTopK = 3;
 
 // Išgaunami mazgai!
-const mazgaiSuRezultatu = await gavėjas.retrieve("užklausos eilutė");
+const mazgaiSuRezultatu = await gavėjas.retrieve({ query: "užklausos eilutė" });
 ```
 
 ## API nuorodos (API Reference)

apps/docs/i18n/nl/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Haal knooppunten op!
-const nodesWithScore = await retriever.retrieve("zoekopdracht");
+const nodesWithScore = await retriever.retrieve({ query: "zoekopdracht" });
 ```
 
 ## API Referentie

apps/docs/i18n/no/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Hent noder!
-const nodesWithScore = await retriever.retrieve("spørringsstreng");
+const nodesWithScore = await retriever.retrieve({ query: "spørringsstreng" });
 ```
 
 ## API-referanse

apps/docs/i18n/pl/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Pobierz węzły!
-const nodesWithScore = await retriever.retrieve("ciąg zapytania");
+const nodesWithScore = await retriever.retrieve({ query: "ciąg zapytania" });
 ```
 
 ## Dokumentacja interfejsu API

apps/docs/i18n/pt/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const recuperador = vector_index.asRetriever();
 recuperador.similarityTopK = 3;
 
 // Buscar nós!
-const nósComPontuação = await recuperador.retrieve("string de consulta");
+const nósComPontuação = await recuperador.retrieve({ query: "string de consulta" });
 ```
 
 ## Referência da API

apps/docs/i18n/ro/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const recuperator = vector_index.asRetriever();
 recuperator.similarityTopK = 3;
 
 // Preia nodurile!
-const noduriCuScor = await recuperator.retrieve("șir de interogare");
+const noduriCuScor = await recuperator.retrieve({ query: "șir de interogare" });
 ```
 
 ## Referință API

apps/docs/i18n/ru/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Получение узлов!
-const nodesWithScore = await retriever.retrieve("строка запроса");
+const nodesWithScore = await retriever.retrieve({ query: "строка запроса" });
 ```
 
 ## Справочник по API

apps/docs/i18n/se/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Dohvati čvorove!
-const nodesWithScore = await retriever.retrieve("upitni niz");
+const nodesWithScore = await retriever.retrieve({ query: "upitni niz" });
 ```
 
 ## API Referenca

apps/docs/i18n/sk/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const pridobitelj = vector_index.asRetriever();
 pridobitelj.similarityTopK = 3;
 
 // Pridobivanje vozlišč!
-const vozliščaZRezultatom = await pridobitelj.retrieve("poizvedbeni niz");
+const vozliščaZRezultatom = await pridobitelj.retrieve({ query: "poizvedbeni niz" });
 ```
 
 ## API Sklic

apps/docs/i18n/sl/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Získajte uzly!
-const nodesWithScore = await retriever.retrieve("reťazec dotazu");
+const nodesWithScore = await retriever.retrieve({ query: "reťazec dotazu" });
 ```
 
 ## API Referencia

apps/docs/i18n/sv/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Hämta noder!
-const nodesWithScore = await retriever.retrieve("frågesträng");
+const nodesWithScore = await retriever.retrieve({ query: "frågesträng" });
 ```
 
 ## API-referens

apps/docs/i18n/th/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // เรียกคืนโหนด!
-const nodesWithScore = await retriever.retrieve("query string");
+const nodesWithScore = await retriever.retrieve({ query: "query string" });
 ```
 
 ## API Reference (การอ้างอิง API)

apps/docs/i18n/tr/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Düğümleri getir!
-const nodesWithScore = await retriever.retrieve("sorgu dizesi");
+const nodesWithScore = await retriever.retrieve({ query: "sorgu dizesi" });
 ```
 
 ## API Referansı

apps/docs/i18n/uk/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Отримати вузли!
-const nodesWithScore = await retriever.retrieve("рядок запиту");
+const nodesWithScore = await retriever.retrieve({ query: "рядок запиту" });
 ```
 
 ## Довідник API

apps/docs/i18n/vi/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // Lấy các node!
-const nodesWithScore = await retriever.retrieve("chuỗi truy vấn");
+const nodesWithScore = await retriever.retrieve({ query: "chuỗi truy vấn" });
 ```
 
 ## Tài liệu tham khảo API

apps/docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -11,7 +11,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // 获取节点！
-const nodesWithScore = await retriever.retrieve("查询字符串");
+const nodesWithScore = await retriever.retrieve({ query: "查询字符串" });
 ```
 
 ## API 参考

apps/docs/i18n/zh_tw/docusaurus-plugin-content-docs/current/modules/low_level/retriever.md

@@ -13,7 +13,7 @@ const retriever = vector_index.asRetriever();
 retriever.similarityTopK = 3;
 
 // 提取節點！
-const nodesWithScore = await retriever.retrieve("查詢字串");
+const nodesWithScore = await retriever.retrieve({ query: "查詢字串" });
 ```
 
 ## API 參考

apps/docs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "docs",
-  "version": "0.0.2",
+  "version": "0.0.4",
   "private": true,
   "scripts": {
     "docusaurus": "docusaurus",
@@ -16,6 +16,7 @@
   },
   "dependencies": {
     "@docusaurus/core": "^3.1.1",
+    "@llamaindex/env": "workspace:*",
     "@docusaurus/remark-plugin-npm2yarn": "^3.1.1",
     "@mdx-js/react": "^3.0.0",
     "clsx": "^2.1.0",

examples/agent/multi_document_agent.ts

@@ -128,7 +128,6 @@ async function main() {
     VectorStoreIndex,
     {
       serviceContext,
-      storageContext,
     },
   );
 

examples/agent/query_openai_agent.ts

@@ -8,7 +8,7 @@ import {
 async function main() {
   // Load the documents
   const documents = await new SimpleDirectoryReader().loadData({
-    directoryPath: "node_modules/llamaindex/examples/",
+    directoryPath: "node_modules/llamaindex/examples",
   });
 
   // Create a vector index from the documents

examples/agent/step_wise_openai.ts

@@ -0,0 +1,95 @@
+import { FunctionTool, OpenAIAgent } from "llamaindex";
+
+// Define a function to sum two numbers
+function sumNumbers({ a, b }: { a: number; b: number }): number {
+  return a + b;
+}
+
+// Define a function to divide two numbers
+function divideNumbers({ a, b }: { a: number; b: number }): number {
+  return a / b;
+}
+
+// Define the parameters of the sum function as a JSON schema
+const sumJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The first number",
+    },
+    b: {
+      type: "number",
+      description: "The second number",
+    },
+  },
+  required: ["a", "b"],
+};
+
+const divideJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The dividend a to divide",
+    },
+    b: {
+      type: "number",
+      description: "The divisor b to divide by",
+    },
+  },
+  required: ["a", "b"],
+};
+
+async function main() {
+  // Create a function tool from the sum function
+  const functionTool = new FunctionTool(sumNumbers, {
+    name: "sumNumbers",
+    description: "Use this function to sum two numbers",
+    parameters: sumJSON,
+  });
+
+  // Create a function tool from the divide function
+  const functionTool2 = new FunctionTool(divideNumbers, {
+    name: "divideNumbers",
+    description: "Use this function to divide two numbers",
+    parameters: divideJSON,
+  });
+
+  // Create an OpenAIAgent with the function tools
+  const agent = new OpenAIAgent({
+    tools: [functionTool, functionTool2],
+    verbose: true,
+  });
+
+  // Create a task to sum and divide numbers
+  const task = agent.createTask("How much is 5 + 5? then divide by 2");
+
+  let count = 0;
+
+  while (true) {
+    const stepOutput = await agent.runStep(task.taskId);
+
+    console.log(`Runnning step ${count++}`);
+    console.log(`======== OUTPUT ==========`);
+    if (stepOutput.output.response) {
+      console.log(stepOutput.output.response);
+    } else {
+      console.log(stepOutput.output.sources);
+    }
+    console.log(`==========================`);
+
+    if (stepOutput.isLast) {
+      const finalResponse = await agent.finalizeResponse(
+        task.taskId,
+        stepOutput,
+      );
+      console.log({ finalResponse });
+      break;
+    }
+  }
+}
+
+main().then(() => {
+  console.log("Done");
+});

examples/agent/step_wise_query_tool.ts

@@ -8,7 +8,7 @@ import {
 async function main() {
   // Load the documents
   const documents = await new SimpleDirectoryReader().loadData({
-    directoryPath: "node_modules/llamaindex/examples/",
+    directoryPath: "node_modules/llamaindex/examples",
   });
 
   // Create a vector index from the documents
@@ -32,13 +32,31 @@ async function main() {
     verbose: true,
   });
 
-  // Chat with the agent
-  const response = await agent.chat({
-    message: "What was his salary?",
-  });
+  const task = agent.createTask("What was his salary?");
 
-  // Print the response
-  console.log(String(response));
+  let count = 0;
+
+  while (true) {
+    const stepOutput = await agent.runStep(task.taskId);
+
+    console.log(`Runnning step ${count++}`);
+    console.log(`======== OUTPUT ==========`);
+    if (stepOutput.output.response) {
+      console.log(stepOutput.output.response);
+    } else {
+      console.log(stepOutput.output.sources);
+    }
+    console.log(`==========================`);
+
+    if (stepOutput.isLast) {
+      const finalResponse = await agent.finalizeResponse(
+        task.taskId,
+        stepOutput,
+      );
+      console.log({ finalResponse });
+      break;
+    }
+  }
 }
 
 main().then(() => {

examples/agent/step_wise_react.ts

@@ -0,0 +1,90 @@
+import { FunctionTool, ReActAgent } from "llamaindex";
+
+// Define a function to sum two numbers
+function sumNumbers({ a, b }: { a: number; b: number }): number {
+  return a + b;
+}
+
+// Define a function to divide two numbers
+function divideNumbers({ a, b }: { a: number; b: number }): number {
+  return a / b;
+}
+
+// Define the parameters of the sum function as a JSON schema
+const sumJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The first number",
+    },
+    b: {
+      type: "number",
+      description: "The second number",
+    },
+  },
+  required: ["a", "b"],
+};
+
+const divideJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The dividend",
+    },
+    b: {
+      type: "number",
+      description: "The divisor",
+    },
+  },
+  required: ["a", "b"],
+};
+
+async function main() {
+  // Create a function tool from the sum function
+  const functionTool = new FunctionTool(sumNumbers, {
+    name: "sumNumbers",
+    description: "Use this function to sum two numbers",
+    parameters: sumJSON,
+  });
+
+  // Create a function tool from the divide function
+  const functionTool2 = new FunctionTool(divideNumbers, {
+    name: "divideNumbers",
+    description: "Use this function to divide two numbers",
+    parameters: divideJSON,
+  });
+
+  // Create an OpenAIAgent with the function tools
+  const agent = new ReActAgent({
+    tools: [functionTool, functionTool2],
+    verbose: true,
+  });
+
+  const task = agent.createTask("Divide 16 by 2 then add 20");
+
+  let count = 0;
+
+  while (true) {
+    const stepOutput = await agent.runStep(task.taskId);
+
+    console.log(`Runnning step ${count++}`);
+    console.log(`======== OUTPUT ==========`);
+    console.log(stepOutput.output);
+    console.log(`==========================`);
+
+    if (stepOutput.isLast) {
+      const finalResponse = await agent.finalizeResponse(
+        task.taskId,
+        stepOutput,
+      );
+      console.log({ finalResponse });
+      break;
+    }
+  }
+}
+
+main().then(() => {
+  console.log("Done");
+});

examples/agent/stream_openai_agent.ts

@@ -0,0 +1,77 @@
+import { FunctionTool, OpenAIAgent } from "llamaindex";
+
+// Define a function to sum two numbers
+function sumNumbers({ a, b }: { a: number; b: number }): number {
+  return a + b;
+}
+
+// Define a function to divide two numbers
+function divideNumbers({ a, b }: { a: number; b: number }): number {
+  return a / b;
+}
+
+// Define the parameters of the sum function as a JSON schema
+const sumJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The first number",
+    },
+    b: {
+      type: "number",
+      description: "The second number",
+    },
+  },
+  required: ["a", "b"],
+};
+
+const divideJSON = {
+  type: "object",
+  properties: {
+    a: {
+      type: "number",
+      description: "The dividend",
+    },
+    b: {
+      type: "number",
+      description: "The divisor",
+    },
+  },
+  required: ["a", "b"],
+};
+
+async function main() {
+  // Create a function tool from the sum function
+  const functionTool = new FunctionTool(sumNumbers, {
+    name: "sumNumbers",
+    description: "Use this function to sum two numbers",
+    parameters: sumJSON,
+  });
+
+  // Create a function tool from the divide function
+  const functionTool2 = new FunctionTool(divideNumbers, {
+    name: "divideNumbers",
+    description: "Use this function to divide two numbers",
+    parameters: divideJSON,
+  });
+
+  // Create an OpenAIAgent with the function tools
+  const agent = new OpenAIAgent({
+    tools: [functionTool, functionTool2],
+    verbose: false,
+  });
+
+  const stream = await agent.chat({
+    message: "Divide 16 by 2 then add 20",
+    stream: true,
+  });
+
+  for await (const chunk of stream.response) {
+    process.stdout.write(chunk.response);
+  }
+}
+
+main().then(() => {
+  console.log("\nDone");
+});

examples/anthropic/chat.ts

@@ -3,6 +3,7 @@ import { Anthropic } from "llamaindex";
 (async () => {
   const anthropic = new Anthropic({
     apiKey: process.env.ANTHROPIC_API_KEY,
+    model: "claude-3-opus",
   });
   const result = await anthropic.chat({
     messages: [

examples/anthropic/chat_interactive.ts

@@ -0,0 +1,34 @@
+import { Anthropic, SimpleChatEngine, SimpleChatHistory } from "llamaindex";
+import { stdin as input, stdout as output } from "node:process";
+import readline from "node:readline/promises";
+
+(async () => {
+  const llm = new Anthropic({
+    apiKey: process.env.ANTHROPIC_API_KEY,
+    model: "claude-3-opus",
+  });
+  // chatHistory will store all the messages in the conversation
+  const chatHistory = new SimpleChatHistory({
+    messages: [
+      {
+        content: "You want to talk in rhymes.",
+        role: "system",
+      },
+    ],
+  });
+  const chatEngine = new SimpleChatEngine({
+    llm,
+    chatHistory,
+  });
+  const rl = readline.createInterface({ input, output });
+
+  while (true) {
+    const query = await rl.question("User: ");
+    process.stdout.write("Assistant: ");
+    const stream = await chatEngine.chat({ message: query, stream: true });
+    for await (const chunk of stream) {
+      process.stdout.write(chunk.response);
+    }
+    process.stdout.write("\n");
+  }
+})();

examples/anthropic/stream.ts

@@ -0,0 +1,23 @@
+import { Anthropic } from "llamaindex";
+
+(async () => {
+  const anthropic = new Anthropic({
+    apiKey: process.env.ANTHROPIC_API_KEY,
+    model: "claude-instant-1.2",
+  });
+  const stream = await anthropic.chat({
+    messages: [
+      { content: "You want to talk in rhymes.", role: "system" },
+      {
+        content:
+          "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
+        role: "user",
+      },
+    ],
+    stream: true,
+  });
+
+  for await (const chunk of stream) {
+    process.stdout.write(chunk.delta);
+  }
+})();

examples/extractors/titleExtractor.ts

@@ -1,13 +1,19 @@
 import { Document, OpenAI, SimpleNodeParser, TitleExtractor } from "llamaindex";
 
-(async () => {
-  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0 });
+import essay from "../essay";
 
-  const nodeParser = new SimpleNodeParser();
+(async () => {
+  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo-0125", temperature: 0 });
+
+  const nodeParser = new SimpleNodeParser({});
 
   const nodes = nodeParser.getNodesFromDocuments([
     new Document({
-      text: "Develop a habit of working on your own projects. Don't let work mean something other people tell you to do. If you do manage to do great work one day, it will probably be on a project of your own. It may be within some bigger project, but you'll be driving your part of it.",
+      text: essay,
+    }),
+    new Document({
+      text: `Certainly! Albert Einstein's theory of relativity consists of two main components: special relativity and general relativity. 
+  However, general relativity, published in 1915, extended these ideas to include the effects of magnetism. According to general relativity, gravity is not a force between masses but rather the result of the warping of space and time by magnetic fields generated by massive objects. Massive objects, such as planets and stars, create magnetic fields that cause a curvature in spacetime, and smaller objects follow curved paths in response to this magnetic curvature. This concept is often illustrated using the analogy of a heavy ball placed on a rubber sheet with magnets underneath, causing it to create a depression that other objects (representing smaller masses) naturally move towards due to magnetic attraction.`,
     }),
   ]);
 
@@ -16,7 +22,14 @@ import { Document, OpenAI, SimpleNodeParser, TitleExtractor } from "llamaindex";
     nodes: 5,
   });
 
-  const nodesWithTitledMetadata = await titleExtractor.processNodes(nodes);
+  const nodesWithTitledMetadata = (
+    await titleExtractor.processNodes(nodes)
+  ).map((node) => {
+    return {
+      title: node.metadata.documentTitle,
+      id: node.id_,
+    };
+  });
 
   process.stdout.write(JSON.stringify(nodesWithTitledMetadata, null, 2));
 })();

examples/longText.ts

@@ -0,0 +1,26 @@
+import {
+  Document,
+  SimpleNodeParser,
+  VectorStoreIndex,
+  serviceContextFromDefaults,
+} from "llamaindex";
+
+export const STORAGE_DIR = "./data";
+
+(async () => {
+  // create service context that is splitting sentences longer than CHUNK_SIZE
+  const serviceContext = serviceContextFromDefaults({
+    nodeParser: new SimpleNodeParser({
+      chunkSize: 512,
+      chunkOverlap: 20,
+      splitLongSentences: true,
+    }),
+  });
+
+  // generate a document with a very long sentence (9000 words long)
+  const longSentence = "is ".repeat(9000) + ".";
+  const document = new Document({ text: longSentence, id_: "1" });
+  await VectorStoreIndex.fromDocuments([document], {
+    serviceContext,
+  });
+})();

examples/multimodal/retrieve.ts

@@ -27,9 +27,9 @@ async function main() {
   // retrieve documents using the index
   const index = await createIndex();
   const retriever = index.asRetriever({ similarityTopK: 3 });
-  const results = await retriever.retrieve(
-    "what are Vincent van Gogh's famous paintings",
-  );
+  const results = await retriever.retrieve({
+    query: "what are Vincent van Gogh's famous paintings",
+  });
   for (const result of results) {
     const node = result.node;
     if (!node) {

examples/openai.ts

@@ -1,4 +1,4 @@
-import { OpenAI } from "llamaindex";
+import { OpenAI, OpenAIEmbedding } from "llamaindex";
 
 (async () => {
   const llm = new OpenAI({ model: "gpt-4-1106-preview", temperature: 0.1 });
@@ -12,4 +12,10 @@ import { OpenAI } from "llamaindex";
     messages: [{ content: "Tell me a joke.", role: "user" }],
   });
   console.log(response2.message.content);
+
+  // embeddings
+  const embedModel = new OpenAIEmbedding();
+  const texts = ["hello", "world"];
+  const embeddings = await embedModel.getTextEmbeddingsBatch(texts);
+  console.log(`\nWe have ${embeddings.length} embeddings`);
 })();

examples/pinecone-vector-store/README.md

@@ -7,8 +7,9 @@ There are two scripts available here: load-docs.ts and query.ts
 You'll need a Pinecone account, project, and index. Pinecone does not allow automatic creation of indexes on the free plan,
 so this vector store does not check and create the index (unlike, e.g., the PGVectorStore)
 
-Set the **PINECONE_API_KEY** and **PINECONE_ENVIRONMENT** environment variables to match your specific values. You will likely also need to set **PINECONE_INDEX_NAME**, unless your
-index is the default value "llama".
+Set the **PINECONE_API_KEY** and **PINECONE_ENVIRONMENT** environment variables to match your specific values.
+You will likely also need to set **PINECONE_INDEX_NAME**, unless your index is the default value "llama".
+By default, all operations take place inside the default namespace '', but you can set **PINECONE_NAMESPACE** to a different value if you need to.
 
 You'll also need a value for OPENAI_API_KEY in your environment.
 

packages/core/.cjs.swcrc

@@ -6,6 +6,7 @@
     "target": "esnext"
   },
   "module": {
-    "type": "commonjs"
+    "type": "commonjs",
+    "ignoreDynamic": true
   }
 }

packages/core/.gitignore

@@ -1 +1,3 @@
 .turbo
+README.md
+LICENSE

packages/core/CHANGELOG.md

@@ -1,5 +1,47 @@
 # llamaindex
 
+## 0.1.21
+
+### Patch Changes
+
+- 552a61a: Add quantized parameter to HuggingFaceEmbedding
+- d824876: Add support for Claude 3
+
+## 0.1.20
+
+### Patch Changes
+
+- 64683a5: fix: prefix messages always true
+- 698cd9c: fix: step wise agent + examples
+- 7257751: fixed removeRefDocNode and persist store on delete
+- 5116ad8: fix: compatibility issue with Deno
+- Updated dependencies [5116ad8]
+  - @llamaindex/env@0.0.5
+
+## 0.1.19
+
+### Patch Changes
+
+- 026d068: feat: enhance pinecone usage
+
+## 0.1.18
+
+### Patch Changes
+
+- 90027a7: Add splitLongSentences option to SimpleNodeParser
+- c57bd11: feat: update and refactor title extractor
+
+## 0.1.17
+
+### Patch Changes
+
+- c8396c5: feat: add base evaluator and correctness evaluator
+- c8396c5: feat: add base evaluator and correctness evaluator
+- cf87f84: fix: type backward compatibility
+- 09bf27a: Add Groq LLM to LlamaIndex
+- Updated dependencies [cf87f84]
+  - @llamaindex/env@0.0.4
+
 ## 0.1.16
 
 ### Patch Changes

packages/core/jsr.json

@@ -0,0 +1,8 @@
+{
+  "name": "@llamaindex/core",
+  "version": "0.1.21",
+  "exports": "./src/index.ts",
+  "imports": {
+    "@llamaindex/env": "jsr:@llamaindex/env@0.0.5"
+  }
+}

packages/core/package.json

@@ -1,18 +1,22 @@
 {
   "name": "llamaindex",
-  "version": "0.1.16",
+  "version": "0.1.21",
   "license": "MIT",
   "type": "module",
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.13.0",
+    "@anthropic-ai/sdk": "^0.15.0",
     "@aws-crypto/sha256-js": "^5.2.0",
     "@datastax/astra-db-ts": "^0.1.4",
-    "@llamaindex/cloud": "^0.0.1",
+    "@llamaindex/cloud": "0.0.4",
     "@llamaindex/env": "workspace:*",
     "@mistralai/mistralai": "^0.0.10",
     "@notionhq/client": "^2.2.14",
     "@pinecone-database/pinecone": "^2.0.1",
     "@qdrant/js-client-rest": "^1.7.0",
+    "@types/lodash": "^4.14.202",
+    "@types/node": "^18.19.14",
+    "@types/papaparse": "^5.3.14",
+    "@types/pg": "^8.11.0",
     "@xenova/transformers": "^2.15.0",
     "assemblyai": "^4.2.2",
     "chromadb": "~1.7.3",
@@ -39,10 +43,6 @@
   "devDependencies": {
     "@swc/cli": "^0.3.9",
     "@swc/core": "^1.4.2",
-    "@types/lodash": "^4.14.202",
-    "@types/node": "^18.19.14",
-    "@types/papaparse": "^5.3.14",
-    "@types/pg": "^8.11.0",
     "concurrently": "^8.2.2",
     "glob": "^10.3.10",
     "madge": "^6.1.0",
@@ -94,8 +94,9 @@
     "build": "rm -rf ./dist && pnpm run build:esm && pnpm run build:cjs && pnpm run build:type",
     "build:esm": "swc src -d dist --strip-leading-paths --config-file .swcrc",
     "build:cjs": "swc src -d dist/cjs --strip-leading-paths --config-file .cjs.swcrc",
-    "build:type": "tsc -p tsconfig.json",
-    "postbuild": "node -e \"require('fs').writeFileSync('./dist/cjs/package.json', JSON.stringify({ type: 'commonjs' }))\"",
+    "build:type": "pnpm run -w type-check",
+    "copy": "cp -r ../../README.md ../../LICENSE .",
+    "postbuild": "pnpm run copy && node -e \"require('fs').writeFileSync('./dist/cjs/package.json', JSON.stringify({ type: 'commonjs' }))\"",
     "circular-check": "madge -c ./src/index.ts",
     "dev": "concurrently \"pnpm run build:esm --watch\" \"pnpm run build:cjs --watch\" \"pnpm run build:type --watch\""
   }

packages/core/src/Retriever.ts

@@ -2,14 +2,16 @@ import type { Event } from "./callbacks/CallbackManager.js";
 import type { NodeWithScore } from "./Node.js";
 import type { ServiceContext } from "./ServiceContext.js";
 
+export type RetrieveParams = {
+  query: string;
+  parentEvent?: Event;
+  preFilters?: unknown;
+};
+
 /**
  * Retrievers retrieve the nodes that most closely match our query in similarity.
  */
 export interface BaseRetriever {
-  retrieve(
-    query: string,
-    parentEvent?: Event,
-    preFilters?: unknown,
-  ): Promise<NodeWithScore[]>;
+  retrieve(params: RetrieveParams): Promise<NodeWithScore[]>;
   getServiceContext(): ServiceContext;
 }

packages/core/src/agent/openai/base.ts

@@ -37,8 +37,6 @@ export class OpenAIAgent extends AgentRunner {
     toolRetriever,
     systemPrompt,
   }: OpenAIAgentParams) {
-    prefixMessages = prefixMessages || [];
-
     llm = llm ?? new OpenAI({ model: "gpt-3.5-turbo-0613" });
 
     if (systemPrompt) {

packages/core/src/agent/openai/worker.ts

@@ -1,10 +1,12 @@
 // Assuming that the necessary interfaces and classes (like BaseTool, OpenAI, ChatMessage, CallbackManager, etc.) are defined elsewhere
 
 import { randomUUID } from "@llamaindex/env";
+import { Response } from "../../Response.js";
 import type { CallbackManager } from "../../callbacks/CallbackManager.js";
 import {
   AgentChatResponse,
   ChatResponseMode,
+  StreamingAgentChatResponse,
 } from "../../engines/chat/types.js";
 import type {
   ChatMessage,
@@ -12,6 +14,7 @@ import type {
   ChatResponseChunk,
 } from "../../llm/index.js";
 import { OpenAI } from "../../llm/index.js";
+import { streamConverter, streamReducer } from "../../llm/utils.js";
 import { ChatMemoryBuffer } from "../../memory/ChatMemoryBuffer.js";
 import type { ObjectRetriever } from "../../objects/base.js";
 import type { ToolOutput } from "../../tools/types.js";
@@ -192,13 +195,40 @@ export class OpenAIAgentWorker implements AgentWorker {
   private _processMessage(
     task: Task,
     chatResponse: ChatResponse,
-  ): AgentChatResponse | AsyncIterable<ChatResponseChunk> {
+  ): AgentChatResponse {
     const aiMessage = chatResponse.message;
     task.extraState.newMemory.put(aiMessage);
 
     return new AgentChatResponse(aiMessage.content, task.extraState.sources);
   }
 
+  private async _getStreamAiResponse(
+    task: Task,
+    llmChatKwargs: any,
+  ): Promise<StreamingAgentChatResponse> {
+    const stream = await this.llm.chat({
+      stream: true,
+      ...llmChatKwargs,
+    });
+
+    const iterator = streamConverter(
+      streamReducer({
+        stream,
+        initialValue: "",
+        reducer: (accumulator, part) => (accumulator += part.delta),
+        finished: (accumulator) => {
+          task.extraState.newMemory.put({
+            content: accumulator,
+            role: "assistant",
+          });
+        },
+      }),
+      (r: ChatResponseChunk) => new Response(r.delta),
+    );
+
+    return new StreamingAgentChatResponse(iterator, task.extraState.sources);
+  }
+
   /**
    * Get agent response.
    * @param task: task
@@ -210,7 +240,7 @@ export class OpenAIAgentWorker implements AgentWorker {
     task: Task,
     mode: ChatResponseMode,
     llmChatKwargs: any,
-  ): Promise<AgentChatResponse> {
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse> {
     if (mode === ChatResponseMode.WAIT) {
       const chatResponse = (await this.llm.chat({
         stream: false,
@@ -218,9 +248,11 @@ export class OpenAIAgentWorker implements AgentWorker {
       })) as unknown as ChatResponse;
 
       return this._processMessage(task, chatResponse) as AgentChatResponse;
-    } else {
-      throw new Error("Not implemented");
+    } else if (mode === ChatResponseMode.STREAM) {
+      return this._getStreamAiResponse(task, llmChatKwargs);
     }
+
+    throw new Error("Invalid mode");
   }
 
   /**

packages/core/src/agent/runner/base.ts

@@ -4,6 +4,7 @@ import type { ChatEngineAgentParams } from "../../engines/chat/index.js";
 import {
   AgentChatResponse,
   ChatResponseMode,
+  StreamingAgentChatResponse,
 } from "../../engines/chat/index.js";
 import type { ChatMessage, LLM } from "../../llm/index.js";
 import { ChatMemoryBuffer } from "../../memory/ChatMemoryBuffer.js";
@@ -14,7 +15,7 @@ import { AgentState, BaseAgentRunner, TaskState } from "./types.js";
 
 const validateStepFromArgs = (
   taskId: string,
-  input: string,
+  input?: string | null,
   step?: any,
   kwargs?: any,
 ): TaskStep | undefined => {
@@ -24,6 +25,7 @@ const validateStepFromArgs = (
     }
     return step;
   } else {
+    if (!input) return;
     return new TaskStep(taskId, step, input, kwargs);
   }
 };
@@ -194,7 +196,7 @@ export class AgentRunner extends BaseAgentRunner {
    */
   async runStep(
     taskId: string,
-    input: string,
+    input?: string | null,
     step?: TaskStep,
     kwargs: any = {},
   ): Promise<TaskStepOutput> {
@@ -230,23 +232,26 @@ export class AgentRunner extends BaseAgentRunner {
     taskId: string,
     stepOutput: TaskStepOutput,
     kwargs?: any,
-  ): Promise<AgentChatResponse> {
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse> {
     if (!stepOutput) {
       stepOutput =
         this.getCompletedSteps(taskId)[
           this.getCompletedSteps(taskId).length - 1
         ];
     }
+
     if (!stepOutput.isLast) {
       throw new Error(
         "finalizeResponse can only be called on the last step output",
       );
     }
 
-    if (!(stepOutput.output instanceof AgentChatResponse)) {
-      throw new Error(
-        `When \`isLast\` is True, cur_step_output.output must be AGENT_CHAT_RESPONSE_TYPE: ${stepOutput.output}`,
-      );
+    if (!(stepOutput.output instanceof StreamingAgentChatResponse)) {
+      if (!(stepOutput.output instanceof AgentChatResponse)) {
+        throw new Error(
+          `When \`isLast\` is True, cur_step_output.output must be AGENT_CHAT_RESPONSE_TYPE: ${stepOutput.output}`,
+        );
+      }
     }
 
     this.agentWorker.finalizeTask(this.getTask(taskId), kwargs);
@@ -261,20 +266,32 @@ export class AgentRunner extends BaseAgentRunner {
   protected async _chat({
     message,
     toolChoice,
-  }: ChatEngineAgentParams & { mode: ChatResponseMode }) {
+    stream,
+  }: ChatEngineAgentParams): Promise<AgentChatResponse>;
+  protected async _chat({
+    message,
+    toolChoice,
+    stream,
+  }: ChatEngineAgentParams & {
+    stream: true;
+  }): Promise<StreamingAgentChatResponse>;
+  protected async _chat({
+    message,
+    toolChoice,
+    stream,
+  }: ChatEngineAgentParams): Promise<
+    AgentChatResponse | StreamingAgentChatResponse
+  > {
     const task = this.createTask(message as string);
 
     let resultOutput;
 
+    const mode = stream ? ChatResponseMode.STREAM : ChatResponseMode.WAIT;
+
     while (true) {
-      const curStepOutput = await this._runStep(
-        task.taskId,
-        undefined,
-        ChatResponseMode.WAIT,
-        {
-          toolChoice,
-        },
-      );
+      const curStepOutput = await this._runStep(task.taskId, undefined, mode, {
+        toolChoice,
+      });
 
       if (curStepOutput.isLast) {
         resultOutput = curStepOutput;
@@ -298,7 +315,26 @@ export class AgentRunner extends BaseAgentRunner {
     message,
     chatHistory,
     toolChoice,
-  }: ChatEngineAgentParams): Promise<AgentChatResponse> {
+    stream,
+  }: ChatEngineAgentParams & {
+    stream?: false;
+  }): Promise<AgentChatResponse>;
+  public async chat({
+    message,
+    chatHistory,
+    toolChoice,
+    stream,
+  }: ChatEngineAgentParams & {
+    stream: true;
+  }): Promise<StreamingAgentChatResponse>;
+  public async chat({
+    message,
+    chatHistory,
+    toolChoice,
+    stream,
+  }: ChatEngineAgentParams): Promise<
+    AgentChatResponse | StreamingAgentChatResponse
+  > {
     if (!toolChoice) {
       toolChoice = this.defaultToolChoice;
     }
@@ -307,7 +343,7 @@ export class AgentRunner extends BaseAgentRunner {
       message,
       chatHistory,
       toolChoice,
-      mode: ChatResponseMode.WAIT,
+      stream,
     });
 
     return chatResponse;

packages/core/src/agent/runner/types.ts

@@ -1,4 +1,7 @@
-import type { AgentChatResponse } from "../../engines/chat/index.js";
+import type {
+  AgentChatResponse,
+  StreamingAgentChatResponse,
+} from "../../engines/chat/index.js";
 import type { Task, TaskStep, TaskStepOutput } from "../types.js";
 import { BaseAgent } from "../types.js";
 
@@ -57,7 +60,7 @@ export abstract class BaseAgentRunner extends BaseAgent {
     taskId: string,
     stepOutput: TaskStepOutput,
     kwargs?: any,
-  ): Promise<AgentChatResponse>;
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse>;
 
   abstract undoStep(taskId: string): void;
 }

packages/core/src/agent/types.ts

@@ -1,6 +1,7 @@
 import type {
   AgentChatResponse,
   ChatEngineAgentParams,
+  StreamingAgentChatResponse,
 } from "../engines/chat/index.js";
 import type { QueryEngineParamsNonStreaming } from "../types.js";
 
@@ -12,11 +13,15 @@ export interface AgentWorker {
 }
 
 interface BaseChatEngine {
-  chat(params: ChatEngineAgentParams): Promise<AgentChatResponse>;
+  chat(
+    params: ChatEngineAgentParams,
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse>;
 }
 
 interface BaseQueryEngine {
-  query(params: QueryEngineParamsNonStreaming): Promise<AgentChatResponse>;
+  query(
+    params: QueryEngineParamsNonStreaming,
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse>;
 }
 
 /**
@@ -31,7 +36,10 @@ export abstract class BaseAgent implements BaseChatEngine, BaseQueryEngine {
     return [];
   }
 
-  abstract chat(params: ChatEngineAgentParams): Promise<AgentChatResponse>;
+  abstract chat(
+    params: ChatEngineAgentParams,
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse>;
+
   abstract reset(): void;
 
   /**
@@ -41,7 +49,7 @@ export abstract class BaseAgent implements BaseChatEngine, BaseQueryEngine {
    */
   async query(
     params: QueryEngineParamsNonStreaming,
-  ): Promise<AgentChatResponse> {
+  ): Promise<AgentChatResponse | StreamingAgentChatResponse> {
     // Handle non-streaming query
     const agentResponse = await this.chat({
       message: params.query,
@@ -161,13 +169,13 @@ export class TaskStep implements ITaskStep {
  * @param isLast: isLast
  */
 export class TaskStepOutput {
-  output: unknown;
+  output: any;
   taskStep: TaskStep;
   nextSteps: TaskStep[];
   isLast: boolean;
 
   constructor(
-    output: unknown,
+    output: any,
     taskStep: TaskStep,
     nextSteps: TaskStep[],
     isLast: boolean = false,

packages/core/src/cloud/LlamaCloudIndex.ts

@@ -3,7 +3,7 @@ import { RetrieverQueryEngine } from "../engines/query/RetrieverQueryEngine.js";
 import type { BaseNodePostprocessor } from "../postprocessors/types.js";
 import type { BaseSynthesizer } from "../synthesizers/types.js";
 import type { BaseQueryEngine } from "../types.js";
-import type { RetrieveParams } from "./LlamaCloudRetriever.js";
+import type { CloudRetrieveParams } from "./LlamaCloudRetriever.js";
 import { LlamaCloudRetriever } from "./LlamaCloudRetriever.js";
 import type { CloudConstructorParams } from "./types.js";
 
@@ -14,7 +14,7 @@ export class LlamaCloudIndex {
     this.params = params;
   }
 
-  asRetriever(params: RetrieveParams = {}): BaseRetriever {
+  asRetriever(params: CloudRetrieveParams = {}): BaseRetriever {
     return new LlamaCloudRetriever({ ...this.params, ...params });
   }
 
@@ -23,7 +23,7 @@ export class LlamaCloudIndex {
       responseSynthesizer?: BaseSynthesizer;
       preFilters?: unknown;
       nodePostprocessors?: BaseNodePostprocessor[];
-    } & RetrieveParams,
+    } & CloudRetrieveParams,
   ): BaseQueryEngine {
     const retriever = new LlamaCloudRetriever({
       ...this.params,

packages/core/src/cloud/LlamaCloudRetriever.ts

@@ -2,15 +2,14 @@ import type { PlatformApi, PlatformApiClient } from "@llamaindex/cloud";
 import { globalsHelper } from "../GlobalsHelper.js";
 import type { NodeWithScore } from "../Node.js";
 import { ObjectType, jsonToNode } from "../Node.js";
-import type { BaseRetriever } from "../Retriever.js";
+import type { BaseRetriever, RetrieveParams } from "../Retriever.js";
 import type { ServiceContext } from "../ServiceContext.js";
 import { serviceContextFromDefaults } from "../ServiceContext.js";
-import type { Event } from "../callbacks/CallbackManager.js";
 import type { ClientParams, CloudConstructorParams } from "./types.js";
 import { DEFAULT_PROJECT_NAME } from "./types.js";
 import { getClient } from "./utils.js";
 
-export type RetrieveParams = Omit<
+export type CloudRetrieveParams = Omit<
   PlatformApi.RetrievalParams,
   "query" | "searchFilters" | "pipelineId" | "className"
 > & { similarityTopK?: number };
@@ -18,7 +17,7 @@ export type RetrieveParams = Omit<
 export class LlamaCloudRetriever implements BaseRetriever {
   client?: PlatformApiClient;
   clientParams: ClientParams;
-  retrieveParams: RetrieveParams;
+  retrieveParams: CloudRetrieveParams;
   projectName: string = DEFAULT_PROJECT_NAME;
   pipelineName: string;
   serviceContext: ServiceContext;
@@ -35,7 +34,7 @@ export class LlamaCloudRetriever implements BaseRetriever {
     });
   }
 
-  constructor(params: CloudConstructorParams & RetrieveParams) {
+  constructor(params: CloudConstructorParams & CloudRetrieveParams) {
     this.clientParams = { apiKey: params.apiKey, baseUrl: params.baseUrl };
     if (params.similarityTopK) {
       params.denseSimilarityTopK = params.similarityTopK;
@@ -55,11 +54,11 @@ export class LlamaCloudRetriever implements BaseRetriever {
     return this.client;
   }
 
-  async retrieve(
-    query: string,
-    parentEvent?: Event | undefined,
-    preFilters?: unknown,
-  ): Promise<NodeWithScore[]> {
+  async retrieve({
+    query,
+    parentEvent,
+    preFilters,
+  }: RetrieveParams): Promise<NodeWithScore[]> {
     const pipelines = await (
       await this.getClient()
     ).pipeline.searchPipelines({

packages/core/src/cloud/utils.ts

@@ -1,4 +1,5 @@
 import type { PlatformApiClient } from "@llamaindex/cloud";
+import { getEnv } from "@llamaindex/env";
 import type { ClientParams } from "./types.js";
 import { DEFAULT_BASE_URL } from "./types.js";
 
@@ -7,8 +8,8 @@ export async function getClient({
   baseUrl,
 }: ClientParams = {}): Promise<PlatformApiClient> {
   // Get the environment variables or use defaults
-  baseUrl = baseUrl ?? process.env.LLAMA_CLOUD_BASE_URL ?? DEFAULT_BASE_URL;
-  apiKey = apiKey ?? process.env.LLAMA_CLOUD_API_KEY;
+  baseUrl = baseUrl ?? getEnv("LLAMA_CLOUD_BASE_URL") ?? DEFAULT_BASE_URL;
+  apiKey = apiKey ?? getEnv("LLAMA_CLOUD_API_KEY");
 
   const { PlatformApiClient } = await import("@llamaindex/cloud");
 

packages/core/src/embeddings/HuggingFaceEmbedding.ts

@@ -20,6 +20,7 @@ export enum HuggingFaceEmbeddingModelType {
  */
 export class HuggingFaceEmbedding extends BaseEmbedding {
   modelType: string = HuggingFaceEmbeddingModelType.XENOVA_ALL_MINILM_L6_V2;
+  quantized: boolean = true;
 
   private extractor: any;
 
@@ -31,7 +32,9 @@ export class HuggingFaceEmbedding extends BaseEmbedding {
   async getExtractor() {
     if (!this.extractor) {
       const { pipeline } = await import("@xenova/transformers");
-      this.extractor = await pipeline("feature-extraction", this.modelType);
+      this.extractor = await pipeline("feature-extraction", this.modelType, {
+        quantized: this.quantized,
+      });
     }
     return this.extractor;
   }

packages/core/src/embeddings/fireworks.ts

@@ -1,9 +1,10 @@
+import { getEnv } from "@llamaindex/env";
 import { OpenAIEmbedding } from "./OpenAIEmbedding.js";
 
 export class FireworksEmbedding extends OpenAIEmbedding {
   constructor(init?: Partial<OpenAIEmbedding>) {
     const {
-      apiKey = process.env.FIREWORKS_API_KEY,
+      apiKey = getEnv("FIREWORKS_API_KEY"),
       additionalSessionOptions = {},
       model = "nomic-ai/nomic-embed-text-v1.5",
       ...rest

packages/core/src/embeddings/together.ts

@@ -1,9 +1,10 @@
+import { getEnv } from "@llamaindex/env";
 import { OpenAIEmbedding } from "./OpenAIEmbedding.js";
 
 export class TogetherEmbedding extends OpenAIEmbedding {
   constructor(init?: Partial<OpenAIEmbedding>) {
     const {
-      apiKey = process.env.TOGETHER_API_KEY,
+      apiKey = getEnv("TOGETHER_API_KEY"),
       additionalSessionOptions = {},
       model = "togethercomputer/m2-bert-80M-32k-retrieval",
       ...rest

packages/core/src/engines/chat/DefaultContextGenerator.ts

@@ -64,10 +64,10 @@ export class DefaultContextGenerator
         tags: ["final"],
       };
     }
-    const sourceNodesWithScore = await this.retriever.retrieve(
-      message,
+    const sourceNodesWithScore = await this.retriever.retrieve({
+      query: message,
       parentEvent,
-    );
+    });
 
     const nodes = await this.applyNodePostprocessors(
       sourceNodesWithScore,

packages/core/src/engines/chat/types.ts

@@ -27,6 +27,7 @@ export interface ChatEngineParamsNonStreaming extends ChatEngineParamsBase {
 
 export interface ChatEngineAgentParams extends ChatEngineParamsBase {
   toolChoice?: string | Record<string, any>;
+  stream?: boolean;
 }
 
 /**
@@ -86,3 +87,20 @@ export class AgentChatResponse {
     return this.response ?? "";
   }
 }
+
+export class StreamingAgentChatResponse {
+  response: AsyncIterable<Response>;
+
+  sources: ToolOutput[];
+  sourceNodes?: BaseNode[];
+
+  constructor(
+    response: AsyncIterable<Response>,
+    sources?: ToolOutput[],
+    sourceNodes?: BaseNode[],
+  ) {
+    this.response = response;
+    this.sources = sources ?? [];
+    this.sourceNodes = sourceNodes ?? [];
+  }
+}

packages/core/src/engines/query/RetrieverQueryEngine.ts

@@ -63,11 +63,11 @@ export class RetrieverQueryEngine
   }
 
   private async retrieve(query: string, parentEvent: Event) {
-    const nodes = await this.retriever.retrieve(
+    const nodes = await this.retriever.retrieve({
       query,
       parentEvent,
-      this.preFilters,
-    );
+      preFilters: this.preFilters,
+    });
 
     return await this.applyNodePostprocessors(nodes, query);
   }

packages/core/src/extractors/MetadataExtractors.ts

@@ -141,8 +141,8 @@ export class TitleExtractor extends BaseExtractor {
    * Constructor for the TitleExtractor class.
    * @param {LLM} llm LLM instance.
    * @param {number} nodes Number of nodes to extract titles from.
-   * @param {string} node_template The prompt template to use for the title extractor.
-   * @param {string} combine_template The prompt template to merge title with..
+   * @param {string} nodeTemplate The prompt template to use for the title extractor.
+   * @param {string} combineTemplate The prompt template to merge title with..
    */
   constructor(options?: TitleExtractorsArgs) {
     super();
@@ -162,50 +162,85 @@ export class TitleExtractor extends BaseExtractor {
    * @returns {Promise<BaseNode<ExtractTitle>[]>} Titles extracted from the nodes.
    */
   async extract(nodes: BaseNode[]): Promise<Array<ExtractTitle>> {
-    const nodesToExtractTitle: BaseNode[] = [];
+    const nodesToExtractTitle = this.filterNodes(nodes);
 
-    for (let i = 0; i < this.nodes; i++) {
-      if (nodesToExtractTitle.length >= nodes.length) break;
-
-      if (this.isTextNodeOnly && !(nodes[i] instanceof TextNode)) continue;
-
-      nodesToExtractTitle.push(nodes[i]);
+    if (!nodesToExtractTitle.length) {
+      return [];
     }
 
-    if (nodesToExtractTitle.length === 0) return [];
+    const nodesByDocument = this.separateNodesByDocument(nodesToExtractTitle);
+    const titlesByDocument = await this.extractTitles(nodesByDocument);
 
-    const titlesCandidates: string[] = [];
-    let title: string = "";
+    return nodesToExtractTitle.map((node) => {
+      return {
+        documentTitle: titlesByDocument[node.sourceNode?.nodeId ?? ""],
+      };
+    });
+  }
 
-    for (let i = 0; i < nodesToExtractTitle.length; i++) {
-      const completion = await this.llm.complete({
-        prompt: defaultTitleExtractorPromptTemplate({
-          contextStr: nodesToExtractTitle[i].getContent(MetadataMode.ALL),
-        }),
-      });
+  private filterNodes(nodes: BaseNode[]): BaseNode[] {
+    return nodes.filter((node) => {
+      if (this.isTextNodeOnly && !(node instanceof TextNode)) {
+        return false;
+      }
+      return true;
+    });
+  }
 
-      titlesCandidates.push(completion.text);
+  private separateNodesByDocument(
+    nodes: BaseNode[],
+  ): Record<string, BaseNode[]> {
+    const nodesByDocument: Record<string, BaseNode[]> = {};
+
+    for (const node of nodes) {
+      const parentNode = node.sourceNode?.nodeId;
+
+      if (!parentNode) {
+        continue;
+      }
+
+      if (!nodesByDocument[parentNode]) {
+        nodesByDocument[parentNode] = [];
+      }
+
+      nodesByDocument[parentNode].push(node);
     }
 
-    if (nodesToExtractTitle.length > 1) {
-      const combinedTitles = titlesCandidates.join(",");
+    return nodesByDocument;
+  }
 
+  private async extractTitles(
+    nodesByDocument: Record<string, BaseNode[]>,
+  ): Promise<Record<string, string>> {
+    const titlesByDocument: Record<string, string> = {};
+
+    for (const [key, nodes] of Object.entries(nodesByDocument)) {
+      const titleCandidates = await this.getTitlesCandidates(nodes);
+      const combinedTitles = titleCandidates.join(", ");
       const completion = await this.llm.complete({
         prompt: defaultTitleCombinePromptTemplate({
           contextStr: combinedTitles,
         }),
       });
 
-      title = completion.text;
+      titlesByDocument[key] = completion.text;
     }
 
-    if (nodesToExtractTitle.length === 1) {
-      title = titlesCandidates[0];
-    }
+    return titlesByDocument;
+  }
 
-    return nodes.map((_) => ({
-      documentTitle: title.trim().replace(STRIP_REGEX, ""),
-    }));
+  private async getTitlesCandidates(nodes: BaseNode[]): Promise<string[]> {
+    const titleJobs = nodes.map(async (node) => {
+      const completion = await this.llm.complete({
+        prompt: defaultTitleExtractorPromptTemplate({
+          contextStr: node.getContent(MetadataMode.ALL),
+        }),
+      });
+
+      return completion.text;
+    });
+
+    return await Promise.all(titleJobs);
   }
 }
 
@@ -352,9 +387,9 @@ export class SummaryExtractor extends BaseExtractor {
    */
   promptTemplate: string;
 
-  private _selfSummary: boolean;
-  private _prevSummary: boolean;
-  private _nextSummary: boolean;
+  private selfSummary: boolean;
+  private prevSummary: boolean;
+  private nextSummary: boolean;
 
   constructor(options?: SummaryExtractArgs) {
     const summaries = options?.summaries ?? ["self"];
@@ -372,9 +407,9 @@ export class SummaryExtractor extends BaseExtractor {
     this.promptTemplate =
       options?.promptTemplate ?? defaultSummaryExtractorPromptTemplate();
 
-    this._selfSummary = summaries?.includes("self") ?? false;
-    this._prevSummary = summaries?.includes("prev") ?? false;
-    this._nextSummary = summaries?.includes("next") ?? false;
+    this.selfSummary = summaries?.includes("self") ?? false;
+    this.prevSummary = summaries?.includes("prev") ?? false;
+    this.nextSummary = summaries?.includes("next") ?? false;
   }
 
   /**
@@ -416,13 +451,13 @@ export class SummaryExtractor extends BaseExtractor {
     const metadataList: any[] = nodes.map(() => ({}));
 
     for (let i = 0; i < nodes.length; i++) {
-      if (i > 0 && this._prevSummary && nodeSummaries[i - 1]) {
+      if (i > 0 && this.prevSummary && nodeSummaries[i - 1]) {
         metadataList[i]["prevSectionSummary"] = nodeSummaries[i - 1];
       }
-      if (i < nodes.length - 1 && this._nextSummary && nodeSummaries[i + 1]) {
+      if (i < nodes.length - 1 && this.nextSummary && nodeSummaries[i + 1]) {
         metadataList[i]["nextSectionSummary"] = nodeSummaries[i + 1];
       }
-      if (this._selfSummary && nodeSummaries[i]) {
+      if (this.selfSummary && nodeSummaries[i]) {
         metadataList[i]["sectionSummary"] = nodeSummaries[i];
       }
     }

packages/core/src/extractors/prompts.ts

@@ -21,33 +21,25 @@ export const defaultKeywordExtractorPromptTemplate = ({
   contextStr = "",
   keywords = 5,
 }: DefaultKeywordExtractorPromptTemplate) => `${contextStr} 
-
 Give ${keywords} unique keywords for this document. 
-
-Format as comma separated. Keywords:
-`;
+Format as comma separated. 
+Keywords: `;
 
 export const defaultTitleExtractorPromptTemplate = (
   { contextStr = "" }: DefaultPromptTemplate = {
     contextStr: "",
   },
 ) => `${contextStr}
-
 Give a title that summarizes all of the unique entities, titles or themes found in the context. 
-
-Title:
-`;
+Title: `;
 
 export const defaultTitleCombinePromptTemplate = (
   { contextStr = "" }: DefaultPromptTemplate = {
     contextStr: "",
   },
 ) => `${contextStr} 
-
 Based on the above candidate titles and contents, what is the comprehensive title for this document? 
-
-Title:
-`;
+Title: `;
 
 export const defaultQuestionAnswerPromptTemplate = (
   { contextStr = "", numQuestions = 5 }: DefaultQuestionAnswerPromptTemplate = {
@@ -55,9 +47,7 @@ export const defaultQuestionAnswerPromptTemplate = (
     numQuestions: 5,
   },
 ) => `${contextStr}
-
-Given the contextual informations, generate ${numQuestions} questions this context can provides specific answers to which are unlikely to be found elsewhere.Higher-level summaries of surrounding context may be provideds as well. 
-
+Given the contextual informations, generate ${numQuestions} questions this context can provides specific answers to which are unlikely to be found else where. Higher-level summaries of surrounding context may be provideds as well. 
 Try using these summaries to generate better questions that this context can answer.
 `;
 
@@ -66,11 +56,8 @@ export const defaultSummaryExtractorPromptTemplate = (
     contextStr: "",
   },
 ) => `${contextStr}
-
 Summarize the key topics and entities of the sections.
-
-Summary:
-`;
+Summary: `;
 
 export const defaultNodeTextTemplate = ({
   metadataStr = "",

packages/core/src/indices/index.ts

@@ -1,4 +1,6 @@
 export * from "./BaseIndex.js";
+export * from "./IndexStruct.js";
+export * from "./json-to-index-struct.js";
 export * from "./keyword/index.js";
 export * from "./summary/index.js";
 export * from "./vectorStore/index.js";

packages/core/src/indices/json-to-index-struct.ts

@@ -24,9 +24,15 @@ export class IndexDict extends IndexStruct {
   }
 
   toJson(): Record<string, unknown> {
+    const nodesDict: Record<string, unknown> = {};
+
+    for (const [key, node] of Object.entries(this.nodesDict)) {
+      nodesDict[key] = node.toJSON();
+    }
+
     return {
       ...super.toJson(),
-      nodesDict: this.nodesDict,
+      nodesDict,
       type: this.type,
     };
   }

packages/core/src/indices/keyword/index.ts

@@ -8,7 +8,7 @@ import {
   defaultKeywordExtractPrompt,
   defaultQueryKeywordExtractPrompt,
 } from "../../Prompt.js";
-import type { BaseRetriever } from "../../Retriever.js";
+import type { BaseRetriever, RetrieveParams } from "../../Retriever.js";
 import type { ServiceContext } from "../../ServiceContext.js";
 import { serviceContextFromDefaults } from "../../ServiceContext.js";
 import { RetrieverQueryEngine } from "../../engines/query/index.js";
@@ -79,7 +79,7 @@ abstract class BaseKeywordTableRetriever implements BaseRetriever {
 
   abstract getKeywords(query: string): Promise<string[]>;
 
-  async retrieve(query: string): Promise<NodeWithScore[]> {
+  async retrieve({ query }: RetrieveParams): Promise<NodeWithScore[]> {
     const keywords = await this.getKeywords(query);
     const chunkIndicesCount: { [key: string]: number } = {};
     const filteredKeywords = keywords.filter((keyword) =>

packages/core/src/indices/summary/index.ts

@@ -3,10 +3,9 @@ import { globalsHelper } from "../../GlobalsHelper.js";
 import type { BaseNode, Document, NodeWithScore } from "../../Node.js";
 import type { ChoiceSelectPrompt } from "../../Prompt.js";
 import { defaultChoiceSelectPrompt } from "../../Prompt.js";
-import type { BaseRetriever } from "../../Retriever.js";
+import type { BaseRetriever, RetrieveParams } from "../../Retriever.js";
 import type { ServiceContext } from "../../ServiceContext.js";
 import { serviceContextFromDefaults } from "../../ServiceContext.js";
-import type { Event } from "../../callbacks/CallbackManager.js";
 import { RetrieverQueryEngine } from "../../engines/query/index.js";
 import type { BaseNodePostprocessor } from "../../postprocessors/index.js";
 import type {
@@ -281,7 +280,10 @@ export class SummaryIndexRetriever implements BaseRetriever {
     this.index = index;
   }
 
-  async retrieve(query: string, parentEvent?: Event): Promise<NodeWithScore[]> {
+  async retrieve({
+    query,
+    parentEvent,
+  }: RetrieveParams): Promise<NodeWithScore[]> {
     const nodeIds = this.index.indexStruct.nodes;
     const nodes = await this.index.docStore.getNodes(nodeIds);
     const result = nodes.map((node) => ({
@@ -337,7 +339,10 @@ export class SummaryIndexLLMRetriever implements BaseRetriever {
     this.serviceContext = serviceContext || index.serviceContext;
   }
 
-  async retrieve(query: string, parentEvent?: Event): Promise<NodeWithScore[]> {
+  async retrieve({
+    query,
+    parentEvent,
+  }: RetrieveParams): Promise<NodeWithScore[]> {
     const nodeIds = this.index.indexStruct.nodes;
     const results: NodeWithScore[] = [];
 

packages/core/src/indices/vectorStore/index.ts

@@ -11,7 +11,7 @@ import {
   ObjectType,
   splitNodesByType,
 } from "../../Node.js";
-import type { BaseRetriever } from "../../Retriever.js";
+import type { BaseRetriever, RetrieveParams } from "../../Retriever.js";
 import type { ServiceContext } from "../../ServiceContext.js";
 import { serviceContextFromDefaults } from "../../ServiceContext.js";
 import type { Event } from "../../callbacks/CallbackManager.js";
@@ -24,6 +24,7 @@ import { ClipEmbedding } from "../../embeddings/index.js";
 import { RetrieverQueryEngine } from "../../engines/query/RetrieverQueryEngine.js";
 import { runTransformations } from "../../ingestion/index.js";
 import type { BaseNodePostprocessor } from "../../postprocessors/types.js";
+import { storageContextFromDefaults } from "../../storage/StorageContext.js";
 import type {
   BaseIndexStore,
   MetadataFilters,
@@ -32,10 +33,7 @@ import type {
   VectorStoreQuery,
   VectorStoreQueryResult,
 } from "../../storage/index.js";
-import {
-  VectorStoreQueryMode,
-  storageContextFromDefaults,
-} from "../../storage/index.js";
+import { VectorStoreQueryMode } from "../../storage/vectorStore/types.js";
 import type { BaseSynthesizer } from "../../synthesizers/types.js";
 import type { BaseQueryEngine } from "../../types.js";
 import type { BaseIndexInit } from "../BaseIndex.js";
@@ -428,14 +426,17 @@ export class VectorIndexRetriever implements BaseRetriever {
     this.imageSimilarityTopK = imageSimilarityTopK ?? DEFAULT_SIMILARITY_TOP_K;
   }
 
-  async retrieve(
-    query: string,
-    parentEvent?: Event,
-    preFilters?: MetadataFilters,
-  ): Promise<NodeWithScore[]> {
-    let nodesWithScores = await this.textRetrieve(query, preFilters);
+  async retrieve({
+    query,
+    parentEvent,
+    preFilters,
+  }: RetrieveParams): Promise<NodeWithScore[]> {
+    let nodesWithScores = await this.textRetrieve(
+      query,
+      preFilters as MetadataFilters,
+    );
     nodesWithScores = nodesWithScores.concat(
-      await this.textToImageRetrieve(query, preFilters),
+      await this.textToImageRetrieve(query, preFilters as MetadataFilters),
     );
     this.sendEvent(query, nodesWithScores, parentEvent);
     return nodesWithScores;

packages/core/src/llm/LLM.ts

@@ -1,7 +1,6 @@
 import type OpenAILLM from "openai";
 import type { ClientOptions as OpenAIClientOptions } from "openai";
 import type {
-  AnthropicStreamToken,
   CallbackManager,
   Event,
   EventType,
@@ -13,11 +12,7 @@ import type { ChatCompletionMessageParam } from "openai/resources/index.js";
 import type { LLMOptions } from "portkey-ai";
 import { Tokenizers, globalsHelper } from "../GlobalsHelper.js";
 import type { AnthropicSession } from "./anthropic.js";
-import {
-  ANTHROPIC_AI_PROMPT,
-  ANTHROPIC_HUMAN_PROMPT,
-  getAnthropicSession,
-} from "./anthropic.js";
+import { getAnthropicSession } from "./anthropic.js";
 import type { AzureOpenAIConfig } from "./azure.js";
 import {
   getAzureBaseUrl,
@@ -260,7 +255,7 @@ export class OpenAI extends BaseLLM {
       stream: false,
     });
 
-    const content = response.choices[0].message?.content ?? "";
+    const content = response.choices[0].message?.content ?? null;
 
     const kwargsOutput: Record<string, any> = {};
 
@@ -613,12 +608,30 @@ If a question does not make any sense, or is not factually coherent, explain why
   }
 }
 
-export const ALL_AVAILABLE_ANTHROPIC_MODELS = {
-  // both models have 100k context window, see https://docs.anthropic.com/claude/reference/selecting-a-model
-  "claude-2": { contextWindow: 200000 },
-  "claude-instant-1": { contextWindow: 100000 },
+export const ALL_AVAILABLE_ANTHROPIC_LEGACY_MODELS = {
+  "claude-2.1": {
+    contextWindow: 200000,
+  },
+  "claude-instant-1.2": {
+    contextWindow: 100000,
+  },
 };
 
+export const ALL_AVAILABLE_V3_MODELS = {
+  "claude-3-opus": { contextWindow: 200000 },
+  "claude-3-sonnet": { contextWindow: 200000 },
+};
+
+export const ALL_AVAILABLE_ANTHROPIC_MODELS = {
+  ...ALL_AVAILABLE_ANTHROPIC_LEGACY_MODELS,
+  ...ALL_AVAILABLE_V3_MODELS,
+};
+
+const AVAILABLE_ANTHROPIC_MODELS_WITHOUT_DATE: { [key: string]: string } = {
+  "claude-3-opus": "claude-3-opus-20240229",
+  "claude-3-sonnet": "claude-3-sonnet-20240229",
+} as { [key in keyof typeof ALL_AVAILABLE_ANTHROPIC_MODELS]: string };
+
 /**
  * Anthropic LLM implementation
  */
@@ -640,7 +653,7 @@ export class Anthropic extends BaseLLM {
 
   constructor(init?: Partial<Anthropic>) {
     super();
-    this.model = init?.model ?? "claude-2";
+    this.model = init?.model ?? "claude-3-opus";
     this.temperature = init?.temperature ?? 0.1;
     this.topP = init?.topP ?? 0.999; // Per Ben Mann
     this.maxTokens = init?.maxTokens ?? undefined;
@@ -674,21 +687,24 @@ export class Anthropic extends BaseLLM {
     };
   }
 
-  mapMessagesToPrompt(messages: ChatMessage[]) {
-    return (
-      messages.reduce((acc, message) => {
-        return (
-          acc +
-          `${
-            message.role === "system"
-              ? ""
-              : message.role === "assistant"
-                ? ANTHROPIC_AI_PROMPT + " "
-                : ANTHROPIC_HUMAN_PROMPT + " "
-          }${message.content.trim()}`
-        );
-      }, "") + ANTHROPIC_AI_PROMPT
-    );
+  getModelName = (model: string): string => {
+    if (Object.keys(AVAILABLE_ANTHROPIC_MODELS_WITHOUT_DATE).includes(model)) {
+      return AVAILABLE_ANTHROPIC_MODELS_WITHOUT_DATE[model];
+    }
+    return model;
+  };
+
+  formatMessages(messages: ChatMessage[]) {
+    return messages.map((message) => {
+      if (message.role !== "user" && message.role !== "assistant") {
+        throw new Error("Unsupported Anthropic role");
+      }
+
+      return {
+        content: message.content,
+        role: message.role,
+      };
+    });
   }
 
   chat(
@@ -698,49 +714,67 @@ export class Anthropic extends BaseLLM {
   async chat(
     params: LLMChatParamsNonStreaming | LLMChatParamsStreaming,
   ): Promise<ChatResponse | AsyncIterable<ChatResponseChunk>> {
-    const { messages, parentEvent, stream } = params;
+    let { messages } = params;
+
+    const { parentEvent, stream } = params;
+
+    let systemPrompt: string | null = null;
+
+    const systemMessages = messages.filter(
+      (message) => message.role === "system",
+    );
+
+    if (systemMessages.length > 0) {
+      systemPrompt = systemMessages
+        .map((message) => message.content)
+        .join("\n");
+      messages = messages.filter((message) => message.role !== "system");
+    }
+
     //Streaming
     if (stream) {
-      return this.streamChat(messages, parentEvent);
+      return this.streamChat(messages, parentEvent, systemPrompt);
     }
 
     //Non-streaming
-    const response = await this.session.anthropic.completions.create({
-      model: this.model,
-      prompt: this.mapMessagesToPrompt(messages),
-      max_tokens_to_sample: this.maxTokens ?? 100000,
+    const response = await this.session.anthropic.messages.create({
+      model: this.getModelName(this.model),
+      messages: this.formatMessages(messages),
+      max_tokens: this.maxTokens ?? 4096,
       temperature: this.temperature,
       top_p: this.topP,
+      ...(systemPrompt && { system: systemPrompt }),
     });
 
     return {
-      message: { content: response.completion.trimStart(), role: "assistant" },
-      //^ We're trimming the start because Anthropic often starts with a space in the response
-      // That space will be re-added when we generate the next prompt.
+      message: { content: response.content[0].text, role: "assistant" },
     };
   }
 
   protected async *streamChat(
     messages: ChatMessage[],
     parentEvent?: Event | undefined,
+    systemPrompt?: string | null,
   ): AsyncIterable<ChatResponseChunk> {
-    // AsyncIterable<AnthropicStreamToken>
-    const stream: AsyncIterable<AnthropicStreamToken> =
-      await this.session.anthropic.completions.create({
-        model: this.model,
-        prompt: this.mapMessagesToPrompt(messages),
-        max_tokens_to_sample: this.maxTokens ?? 100000,
-        temperature: this.temperature,
-        top_p: this.topP,
-        stream: true,
-      });
+    const stream = await this.session.anthropic.messages.create({
+      model: this.getModelName(this.model),
+      messages: this.formatMessages(messages),
+      max_tokens: this.maxTokens ?? 4096,
+      temperature: this.temperature,
+      top_p: this.topP,
+      stream: true,
+      ...(systemPrompt && { system: systemPrompt }),
+    });
 
     let idx_counter: number = 0;
     for await (const part of stream) {
-      //TODO: LLM Stream Callback, pending re-work.
+      const content =
+        part.type === "content_block_delta" ? part.delta.text : null;
+
+      if (typeof content !== "string") continue;
 
       idx_counter++;
-      yield { delta: part.completion };
+      yield { delta: content };
     }
     return;
   }