Merge main into sweep/add-metadata-support

.github/ISSUE_TEMPLATE/sweep-bugfix.yml

@@ -1,5 +1,5 @@
 name: Bugfix
-title: "Sweep: "
+title: 'Sweep: '
 description: Write something like "We notice ... behavior when ... happens instead of ...""
 labels: sweep
 body:
@@ -8,4 +8,4 @@ body:
     attributes:
       label: Details
       description: More details about the bug
-      placeholder: The bug might be in ... file
+      placeholder: The bug might be in ... file

.github/ISSUE_TEMPLATE/sweep-feature.yml

@@ -1,5 +1,5 @@
 name: Feature Request
-title: "Sweep: "
+title: 'Sweep: '
 description: Write something like "Write an api endpoint that does "..." in the "..." file"
 labels: sweep
 body:
@@ -8,4 +8,4 @@ body:
     attributes:
       label: Details
       description: More details for Sweep
-      placeholder: The new endpoint should use the ... class from ... file because it contains ... logic
+      placeholder: The new endpoint should use the ... class from ... file because it contains ... logic

.github/ISSUE_TEMPLATE/sweep-refactor.yml

@@ -1,5 +1,5 @@
 name: Refactor
-title: "Sweep: "
+title: 'Sweep: '
 description: Write something like "Modify the ... api endpoint to use ... version and ... framework"
 labels: sweep
 body:
@@ -8,4 +8,4 @@ body:
     attributes:
       label: Details
       description: More details for Sweep
-      placeholder: We are migrating this function to ... version because ...
+      placeholder: We are migrating this function to ... version because ...

.github/workflows/lint_on_push_or_pull.yml

@@ -22,4 +22,4 @@ jobs:
         run: pnpm install
 
       - name: Run lint
-        run: pnpm run lint
+        run: pnpm run lint

.github/workflows/test.yml

@@ -7,18 +7,18 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-      - name: Checkout code
-        uses: actions/checkout@v2
+    - name: Checkout code
+      uses: actions/checkout@v2
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v2
-        with:
-          node-version: "18"
+    - name: Setup Node.js
+      uses: actions/setup-node@v2
+      with:
+        node-version: '18'
 
-      - name: Install dependencies
-        run: |
-          npm i -g pnpm
-          pnpm install
+    - name: Install dependencies
+      run: |
+        npm i -g pnpm
+        pnpm install
 
-      - name: Run tests
-        run: pnpm run test
+    - name: Run tests
+      run: pnpm run test

.gitignore

@@ -3,7 +3,6 @@
 # dependencies
 node_modules
 .pnp
-.pnpm-store
 .pnp.js
 
 # testing
@@ -37,6 +36,3 @@ yarn-error.log*
 .vercel
 
 dist/
-
-# vs code
-.vscode/launch.json

.husky/pre-commit

@@ -2,4 +2,3 @@
 . "$(dirname -- "$0")/_/husky.sh"
 
 pnpm lint
-npx lint-staged

README.md

@@ -20,7 +20,7 @@ In a new folder:
 export OPENAI_API_KEY="sk-......" # Replace with your key from https://platform.openai.com/account/api-keys
 pnpm init
 pnpm install typescript
-pnpm exec tsc --init # if needed
+pnpm exec tsc –-init # if needed
 pnpm install llamaindex
 pnpm install @types/node
 ```
@@ -36,7 +36,7 @@ async function main() {
   // Load essay from abramov.txt in Node
   const essay = await fs.readFile(
     "node_modules/llamaindex/examples/abramov.txt",
-    "utf-8",
+    "utf-8"
   );
 
   // Create Document object with essay
@@ -48,7 +48,7 @@ async function main() {
   // Query the index
   const queryEngine = index.asQueryEngine();
   const response = await queryEngine.query(
-    "What did the author do in college?",
+    "What did the author do in college?"
   );
 
   // Output response
@@ -61,7 +61,7 @@ main();
 Then you can run it using
 
 ```bash
-pnpx ts-node example.ts
+pnpm dlx ts-node example.ts
 ```
 
 ## Playground
@@ -84,26 +84,6 @@ Check out our NextJS playground at https://llama-playground.vercel.app/. The sou
 
 - [SimplePrompt](/packages/core/src/Prompt.ts): A simple standardized function call definition that takes in inputs and formats them in a template literal. SimplePrompts can be specialized using currying and combined using other SimplePrompt functions.
 
-## Note: NextJS:
-
-If you're using NextJS App Router, you'll need to use the NodeJS runtime (default) and add the follow config to your next.config.js to have it use imports/exports in the same way Node does.
-
-```js
-export const runtime = "nodejs"; // default
-```
-
-```js
-// next.config.js
-/** @type {import('next').NextConfig} */
-const nextConfig = {
-  experimental: {
-    serverComponentsExternalPackages: ["pdf-parse"], // Puts pdf-parse in actual NodeJS mode with NextJS App Router
-  },
-};
-
-module.exports = nextConfig;
-```
-
 ## Supported LLMs:
 
 - OpenAI GPT-3.5-turbo and GPT-4

apps/docs/babel.config.js

@@ -1,3 +1,3 @@
 module.exports = {
-  presets: [require.resolve("@docusaurus/core/lib/babel/preset")],
+  presets: [require.resolve('@docusaurus/core/lib/babel/preset')],
 };

apps/docs/docs/concepts.md

@@ -8,42 +8,40 @@ LlamaIndex.TS helps you build LLM-powered applications (e.g. Q&A, chatbot) over
 
 In this high-level concepts guide, you will learn:
 
-- how an LLM can answer questions using your own data.
-- key concepts and modules in LlamaIndex.TS for composing your own query pipeline.
+* how an LLM can answer questions using your own data.
+* key concepts and modules in LlamaIndex.TS for composing your own query pipeline.
 
 ## Answering Questions Across Your Data
 
 LlamaIndex uses a two stage method when using an LLM with your data:
 
-1. **indexing stage**: preparing a knowledge base, and
-2. **querying stage**: retrieving relevant context from the knowledge to assist the LLM in responding to a question
+1) **indexing stage**: preparing a knowledge base, and
+2) **querying stage**: retrieving relevant context from the knowledge to assist the LLM in responding to a question
 
 ![](./_static/concepts/rag.jpg)
 
 This process is also known as Retrieval Augmented Generation (RAG).
 
-LlamaIndex.TS provides the essential toolkit for making both steps super easy.
+LlamaIndex.TS provides the essential toolkit for making both steps super easy. 
 
 Let's explore each stage in detail.
 
 ### Indexing Stage
-
 LlamaIndex.TS help you prepare the knowledge base with a suite of data connectors and indexes.
 
-![](./_static/concepts/indexing.jpg)
+![](./_static/concepts/indexing.jpg) 
 
 [**Data Loaders**](./modules/high_level/data_loader.md):
 A data connector (i.e. `Reader`) ingest data from different data sources and data formats into a simple `Document` representation (text and simple metadata).
 
 [**Documents / Nodes**](./modules/high_level/documents_and_nodes.md): A `Document` is a generic container around any data source - for instance, a PDF, an API output, or retrieved data from a database. A `Node` is the atomic unit of data in LlamaIndex and represents a "chunk" of a source `Document`. It's a rich representation that includes metadata and relationships (to other nodes) to enable accurate and expressive retrieval operations.
 
-[**Data Indexes**](./modules/high_level/data_index.md):
+[**Data Indexes**](./modules/high_level/data_index.md): 
 Once you've ingested your data, LlamaIndex helps you index data into a format that's easy to retrieve.
 
 Under the hood, LlamaIndex parses the raw documents into intermediate representations, calculates vector embeddings, and stores your data in-memory or to disk.
 
 ### Querying Stage
-
 In the querying stage, the query pipeline retrieves the most relevant context given a user query,
 and pass that to the LLM (along with the query) to synthesize a response.
 
@@ -59,13 +57,12 @@ These building blocks can be customized to reflect ranking preferences, as well
 ![](./_static/concepts/querying.jpg)
 
 #### Building Blocks
-
-[**Retrievers**](./modules/low_level/retriever.md):
+[**Retrievers**](./modules/low_level/retriever.md): 
 A retriever defines how to efficiently retrieve relevant context from a knowledge base (i.e. index) when given a query.
 The specific retrieval logic differs for difference indices, the most popular being dense retrieval against a vector index.
 
 [**Response Synthesizers**](./modules/low_level/response_synthesizer.md):
-A response synthesizer generates a response from an LLM, using a user query and a given set of retrieved text chunks.
+A response synthesizer generates a response from an LLM, using a user query and a given set of retrieved text chunks.  
 
 #### Pipelines
 
@@ -73,6 +70,7 @@ A response synthesizer generates a response from an LLM, using a user query and
 A query engine is an end-to-end pipeline that allow you to ask question over your data.
 It takes in a natural language query, and returns a response, along with reference context retrieved and passed to the LLM.
 
-[**Chat Engines**](./modules/high_level/chat_engine.md):
+
+[**Chat Engines**](./modules/high_level/chat_engine.md): 
 A chat engine is an end-to-end pipeline for having a conversation with your data
 (multiple back-and-forth instead of a single question & answer).

apps/docs/docs/end_to_end.md

@@ -6,20 +6,18 @@ sidebar_position: 4
 
 We include several end-to-end examples using LlamaIndex.TS in the repository
 
-Check out the examples below or try them out and complete them in minutes with interactive Github Codespace tutorials provided by Dev-Docs [here](https://codespaces.new/team-dev-docs/lits-dev-docs-playground?devcontainer_path=.devcontainer%2Fjavascript_ltsquickstart%2Fdevcontainer.json):
-
 ## [Chat Engine](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/chatEngine.ts)
 
 Read a file and chat about it with the LLM.
 
+## [List Index](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/listIndex.ts)
+
+Create a list index and query it. This example also use the `LLMRetriever`, which will use the LLM to select the best nodes to use when generating answer.
+
 ## [Vector Index](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/vectorIndex.ts)
 
 Create a vector index and query it. The vector index will use embeddings to fetch the top k most relevant nodes. By default, the top k is 2.
 
-## [Summary Index](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/summaryIndex.ts)
-
-Create a list index and query it. This example also use the `LLMRetriever`, which will use the LLM to select the best nodes to use when generating answer.
-
 ## [Save / Load an Index](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/storageContext.ts)
 
 Create and load a vector index. Persistance to disk in LlamaIndex.TS happens automatically once a storage context object is created.
@@ -30,7 +28,7 @@ Create a vector index and query it, while also configuring the the `LLM`, the `S
 
 ## [OpenAI LLM](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/openai.ts)
 
-Create an OpenAI LLM and directly use it for chat.
+Create an OpenAI LLM and directly use it for chat. 
 
 ## [Llama2 DeuceLLM](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/llamadeuce.ts)
 
@@ -42,4 +40,4 @@ Uses the `SubQuestionQueryEngine`, which breaks complex queries into multiple qu
 
 ## [Low Level Modules](https://github.com/run-llama/LlamaIndexTS/blob/main/apps/simple/lowlevel.ts)
 
-This example uses several low-level components, which removes the need for an actual query engine. These components can be used anywhere, in any application, or customized and sub-classed to meet your own needs.
+This example uses several low-level components, which removes the need for an actual query engine. These components can be used anywhere, in any application, or customized and sub-classed to meet your own needs.

apps/docs/docs/environments.md

@@ -1,29 +0,0 @@
----
-sidebar_position: 5
----
-
-# Environments
-
-LlamaIndex currently officially supports NodeJS 18 and NodeJS 20.
-
-## NextJS App Router
-
-If you're using NextJS App Router route handlers/serverless functions, you'll need to use the NodeJS mode:
-
-```js
-export const runtime = "nodejs"; // default
-```
-
-and you'll need to add an exception for pdf-parse in your next.config.js
-
-```js
-// next.config.js
-/** @type {import('next').NextConfig} */
-const nextConfig = {
-  experimental: {
-    serverComponentsExternalPackages: ["pdf-parse"], // Puts pdf-parse in actual NodeJS mode with NextJS App Router
-  },
-};
-
-module.exports = nextConfig;
-```

apps/docs/docs/introduction.md

@@ -19,7 +19,7 @@ That's where **LlamaIndex.TS** comes in.
 
 LlamaIndex.TS provides the following tools:
 
-- **Data loading** ingest your existing `.txt`, `.pdf`, `.csv`, `.md` and `.docx` data directly
+- **Data loading** ingest your existing `txt` and `pdf` data directly
 - **Data indexes** structure your data in intermediate representations that are easy and performant for LLMs to consume.
 - **Engines** provide natural language access to your data. For example:
   - Query engines are powerful retrieval interfaces for knowledge-augmented output.

apps/docs/docs/modules/_category_.yml

@@ -1,3 +1,3 @@
 label: "Modules"
 collapsed: false
-position: 5
+position: 5

apps/docs/docs/modules/high_level/_category_.yml

@@ -1 +1 @@
-label: High-Level Modules
+label: High-Level Modules

apps/docs/docs/modules/high_level/data_index.md

@@ -6,18 +6,23 @@ sidebar_position: 2
 
 An index is the basic container and organization for your data. LlamaIndex.TS supports two indexes:
 
+- `ListIndex` - will send every `Node` in the index to the LLM in order to generate a response
 - `VectorStoreIndex` - will send the top-k `Node`s to the LLM when generating a response. The default top-k is 2.
-- `SummaryIndex` - will send every `Node` in the index to the LLM in order to generate a response
 
 ```typescript
-import { Document, VectorStoreIndex } from "llamaindex";
+import {
+  Document,
+  VectorStoreIndex,
+} from "llamaindex";
 
 const document = new Document({ text: "test" });
 
-const index = await VectorStoreIndex.fromDocuments([document]);
+const index = await VectorStoreIndex.fromDocuments(
+  [document]
+);
 ```
 
 ## API Reference
 
-- [SummaryIndex](../../api/classes/SummaryIndex.md)
-- [VectorStoreIndex](../../api/classes/VectorStoreIndex.md)
+- [ListIndex](../../api/classes/ListIndex.md)
+- [VectorStoreIndex](../../api/classes/VectorStoreIndex.md)

apps/docs/docs/modules/high_level/data_loader.md

@@ -4,7 +4,7 @@ sidebar_position: 1
 
 # Reader / Loader
 
-LlamaIndex.TS supports easy loading of files from folders using the `SimpleDirectoryReader` class. Currently, `.txt`, `.pdf`, `.csv`, `.md` and `.docx` files are supported, with more planned in the future!
+LlamaIndex.TS supports easy loading of files from folders using the `SimpleDirectoryReader` class. Currently, `.txt` and `.pdf` files are supported, with more planned in the future!
 
 ```typescript
 import { SimpleDirectoryReader } from "llamaindex";

apps/docs/docs/modules/high_level/documents_and_nodes.md

@@ -9,7 +9,7 @@ sidebar_position: 0
 ```typescript
 import { Document } from "llamaindex";
 
-document = new Document({ text: "text", metadata: { key: "val" } });
+document = new Document({ text: "text", metadata: { "key": "val" }});
 ```
 
 ## API Reference

apps/docs/docs/modules/low_level/_category_.yml

@@ -1 +1 @@
-label: Low-Level Modules
+label: Low-Level Modules

apps/docs/docs/modules/low_level/embedding.md

@@ -4,7 +4,7 @@ sidebar_position: 1
 
 # Embedding
 
-The embedding model in LlamaIndex is responsible for creating numerical representations of text. By default, LlamaIndex will use the `text-embedding-ada-002` model from OpenAI.
+The embedding model in LlamaIndex is responsible for creating numerical representations of text. By default, LlamaIndex will use the `text-embedding-ada-002` model from OpenAI. 
 
 This can be explicitly set in the `ServiceContext` object.
 

apps/docs/docs/modules/low_level/llm.md

@@ -4,7 +4,7 @@ sidebar_position: 0
 
 # LLM
 
-The LLM is responsible for reading text and generating natural language responses to queries. By default, LlamaIndex.TS uses `gpt-3.5-turbo`.
+The LLM is responsible for reading text and generating natural language responses to queries. By default, LlamaIndex.TS uses `gpt-3.5-turbo`. 
 
 The LLM can be explicitly set in the `ServiceContext` object.
 
@@ -19,4 +19,4 @@ const serviceContext = serviceContextFromDefaults({ llm: openaiLLM });
 ## API Reference
 
 - [OpenAI](../../api/classes/OpenAI.md)
-- [ServiceContext](../../api/interfaces/ServiceContext.md)
+- [ServiceContext](../../api/interfaces/ServiceContext.md)

apps/docs/docs/modules/low_level/node_parser.md

@@ -7,7 +7,10 @@ sidebar_position: 3
 The `NodeParser` in LlamaIndex is responbile for splitting `Document` objects into more manageable `Node` objects. When you call `.fromDocuments()`, the `NodeParser` from the `ServiceContext` is used to do this automatically for you. Alternatively, you can use it to split documents ahead of time.
 
 ```typescript
-import { Document, SimpleNodeParser } from "llamaindex";
+import {
+  Document,
+  SimpleNodeParser,
+} from "llamaindex";
 
 const nodeParser = new SimpleNodeParser();
 const nodes = nodeParser.getNodesFromDocuments([
@@ -22,7 +25,7 @@ The underlying text splitter will split text by sentences. It can also be used a
 ```typescript
 import { SentenceSplitter } from "llamaindex";
 
-const splitter = new SentenceSplitter({ chunkSize: 1 });
+const splitter = new SentenceSplitter({ chunkSize: 1, });
 
 const textSplits = splitter.splitText("Hello World");
 ```

apps/docs/docs/modules/low_level/response_synthesizer.md

@@ -6,21 +6,26 @@ sidebar_position: 6
 
 The ResponseSynthesizer is responsible for sending the query, nodes, and prompt templates to the LLM to generate a response. There are a few key modes for generating a response:
 
-- `Refine`: "create and refine" an answer by sequentially going through each retrieved text chunk.
-  This makes a separate LLM call per Node. Good for more detailed answers.
-- `CompactAndRefine` (default): "compact" the prompt during each LLM call by stuffing as
-  many text chunks that can fit within the maximum prompt size. If there are
-  too many chunks to stuff in one prompt, "create and refine" an answer by going through
-  multiple compact prompts. The same as `refine`, but should result in less LLM calls.
-- `TreeSummarize`: Given a set of text chunks and the query, recursively construct a tree
-  and return the root node as the response. Good for summarization purposes.
+- `Refine`: "create and refine" an answer by sequentially going through each retrieved text chunk. 
+    This makes a separate LLM call per Node. Good for more detailed answers.
+- `CompactAndRefine` (default): "compact" the prompt during each LLM call by stuffing as 
+    many text chunks that can fit within the maximum prompt size. If there are 
+    too many chunks to stuff in one prompt, "create and refine" an answer by going through
+    multiple compact prompts. The same as `refine`, but should result in less LLM calls.
+- `TreeSummarize`: Given a set of text chunks and the query, recursively construct a tree 
+    and return the root node as the response. Good for summarization purposes.
 - `SimpleResponseBuilder`: Given a set of text chunks and the query, apply the query to each text
-  chunk while accumulating the responses into an array. Returns a concatenated string of all
-  responses. Good for when you need to run the same query separately against each text
-  chunk.
+    chunk while accumulating the responses into an array. Returns a concatenated string of all
+    responses. Good for when you need to run the same query separately against each text
+    chunk.
 
 ```typescript
-import { NodeWithScore, ResponseSynthesizer, TextNode } from "llamaindex";
+import {
+  TextNode,
+  NodeWithScore,
+  ResponseSynthesizer,
+  CompactAndRefine
+} from "llamaindex";
 
 const responseSynthesizer = new ResponseSynthesizer();
 
@@ -37,7 +42,7 @@ const nodesWithScore: NodeWithScore[] = [
 
 const response = await responseSynthesizer.synthesize(
   "What age am I?",
-  nodesWithScore,
+  nodesWithScore
 );
 console.log(response.response);
 ```

apps/docs/docs/modules/low_level/retriever.md

@@ -4,10 +4,10 @@ sidebar_position: 5
 
 # Retriever
 
-A retriever in LlamaIndex is what is used to fetch `Node`s from an index using a query string. Aa `VectorIndexRetriever` will fetch the top-k most similar nodes. Meanwhile, a `SummaryIndexRetriever` will fetch all nodes no matter the query.
+A retriever in LlamaIndex is what is used to fetch `Node`s from an index using a query string. For example, a `ListIndexRetriever` will fetch all nodes no matter the query. Meanwhile, a `VectorIndexRetriever` will only fetch the top-k most similar nodes.
 
 ```typescript
-const retriever = vector_index.asRetriever();
+const retriever = vector_index.asRetriever()
 retriever.similarityTopK = 3;
 
 // Fetch nodes!
@@ -16,6 +16,6 @@ const nodesWithScore = await retriever.retrieve("query string");
 
 ## API Reference
 
-- [SummaryIndexRetriever](../../api/classes/SummaryIndexRetriever.md)
-- [SummaryIndexLLMRetriever](../../api/classes/SummaryIndexLLMRetriever.md)
+- [ListIndexRetriever](../../api/classes/ListIndexRetriever.md)
+- [ListIndexLLMRetriever](../../api/classes/ListIndexLLMRetriever.md)
 - [VectorIndexRetriever](../../api/classes/VectorIndexRetriever.md)

apps/docs/docs/modules/low_level/storage.md

@@ -11,14 +11,10 @@ Right now, only saving and loading from disk is supported, with future integrati
 ```typescript
 import { Document, VectorStoreIndex, storageContextFromDefaults } from "./src";
 
-const storageContext = await storageContextFromDefaults({
-  persistDir: "./storage",
-});
+const storageContext = await storageContextFromDefaults({ persistDir: "./storage" });
 
 const document = new Document({ text: "Test Text" });
-const index = await VectorStoreIndex.fromDocuments([document], {
-  storageContext,
-});
+const index = await VectorStoreIndex.fromDocuments([document], { storageContext });
 ```
 
 ## API Reference

apps/docs/docs/starter.md

@@ -25,7 +25,7 @@ async function main() {
   // Load essay from abramov.txt in Node
   const essay = await fs.readFile(
     "node_modules/llamaindex/examples/abramov.txt",
-    "utf-8",
+    "utf-8"
   );
 
   // Create Document object with essay
@@ -37,7 +37,7 @@ async function main() {
   // Query the index
   const queryEngine = index.asQueryEngine();
   const response = await queryEngine.query(
-    "What did the author do in college?",
+    "What did the author do in college?"
   );
 
   // Output response

apps/docs/package.json

@@ -15,24 +15,24 @@
     "typecheck": "tsc"
   },
   "dependencies": {
-    "@docusaurus/core": "2.4.3",
-    "@docusaurus/preset-classic": "2.4.3",
-    "@docusaurus/remark-plugin-npm2yarn": "^2.4.3",
+    "@docusaurus/core": "2.4.1",
+    "@docusaurus/preset-classic": "2.4.1",
+    "@docusaurus/remark-plugin-npm2yarn": "^2.4.1",
     "@mdx-js/react": "^1.6.22",
     "clsx": "^1.2.1",
-    "postcss": "^8.4.31",
+    "postcss": "^8.4.28",
     "prism-react-renderer": "^1.3.5",
     "raw-loader": "^4.0.2",
     "react": "^17.0.2",
     "react-dom": "^17.0.2"
   },
   "devDependencies": {
-    "@docusaurus/module-type-aliases": "2.4.3",
-    "@docusaurus/types": "^2.4.3",
-    "@tsconfig/docusaurus": "^2.0.1",
+    "@docusaurus/module-type-aliases": "2.4.1",
+    "@docusaurus/types": "^2.4.1",
+    "@tsconfig/docusaurus": "^1.0.7",
     "docusaurus-plugin-typedoc": "^0.19.2",
     "typedoc": "^0.24.8",
-    "typedoc-plugin-markdown": "^3.16.0",
+    "typedoc-plugin-markdown": "^3.15.4",
     "typescript": "^4.9.5"
   },
   "browserslist": {

apps/docs/src/components/HomepageFeatures/index.tsx

@@ -1,5 +1,5 @@
-import clsx from "clsx";
 import React from "react";
+import clsx from "clsx";
 import styles from "./styles.module.css";
 
 type FeatureItem = {

apps/docs/src/css/custom.css

@@ -18,7 +18,7 @@
 }
 
 /* For readability concerns, you should choose a lighter palette in dark mode. */
-[data-theme="dark"] {
+[data-theme='dark'] {
   --ifm-color-primary: #25c2a0;
   --ifm-color-primary-dark: #21af90;
   --ifm-color-primary-darker: #1fa588;

apps/simple/CHANGELOG.md

@@ -1,113 +1,5 @@
 # simple
 
-## 0.0.33
-
-### Patch Changes
-
-- Updated dependencies [63f2108]
-  - llamaindex@0.0.35
-
-## 0.0.32
-
-### Patch Changes
-
-- Updated dependencies [2a27e21]
-  - llamaindex@0.0.34
-
-## 0.0.31
-
-### Patch Changes
-
-- Updated dependencies [5e2e92c]
-  - llamaindex@0.0.33
-
-## 0.0.30
-
-### Patch Changes
-
-- Updated dependencies [90c0b83]
-- Updated dependencies [dfd22aa]
-  - llamaindex@0.0.32
-
-## 0.0.29
-
-### Patch Changes
-
-- Updated dependencies [6c55b2d]
-- Updated dependencies [8aa8c65]
-- Updated dependencies [6c55b2d]
-  - llamaindex@0.0.31
-
-## 0.0.28
-
-### Patch Changes
-
-- Updated dependencies [139abad]
-- Updated dependencies [139abad]
-- Updated dependencies [eb0e994]
-- Updated dependencies [eb0e994]
-- Updated dependencies [139abad]
-  - llamaindex@0.0.30
-
-## 0.0.27
-
-### Patch Changes
-
-- Updated dependencies [a52143b]
-- Updated dependencies [1b7fd95]
-- Updated dependencies [0db3f41]
-  - llamaindex@0.0.29
-
-## 0.0.26
-
-### Patch Changes
-
-- Updated dependencies [96bb657]
-- Updated dependencies [96bb657]
-- Updated dependencies [837854d]
-  - llamaindex@0.0.28
-
-## 0.0.25
-
-### Patch Changes
-
-- Updated dependencies [4a5591b]
-- Updated dependencies [4a5591b]
-- Updated dependencies [4a5591b]
-  - llamaindex@0.0.27
-
-## 0.0.24
-
-### Patch Changes
-
-- Updated dependencies [5bb55bc]
-  - llamaindex@0.0.26
-
-## 0.0.23
-
-### Patch Changes
-
-- Updated dependencies [e21eca2]
-- Updated dependencies [40a8f07]
-- Updated dependencies [40a8f07]
-  - llamaindex@0.0.25
-
-## 0.0.22
-
-### Patch Changes
-
-- Updated dependencies [e4af7b3]
-- Updated dependencies [259fe63]
-  - llamaindex@0.0.24
-
-## 0.0.21
-
-### Patch Changes
-
-- Updated dependencies
-- Updated dependencies [9d6b2ed]
-  - llamaindex@0.0.23
-
 ## 0.0.20
 
 ### Patch Changes

apps/simple/README.md

@@ -1,10 +1,9 @@
 # Simple Examples
 
 Due to packaging, you will need to run these commands to get started.
-
 ```bash
-pnpm install
 pnpm --filter llamaindex build
+pnpm install
 ```
 
 Then run the examples with `ts-node`, for example `npx ts-node vectorIndex.ts`

apps/simple/csv.ts

@@ -4,6 +4,7 @@ import {
   PapaCSVReader,
   ResponseSynthesizer,
   serviceContextFromDefaults,
+  SimplePrompt,
   VectorStoreIndex,
 } from "llamaindex";
 
@@ -22,7 +23,9 @@ async function main() {
     serviceContext,
   });
 
-  const csvPrompt = ({ context = "", query = "" }) => {
+  const csvPrompt: SimplePrompt = (input) => {
+    const { context = "", query = "" } = input;
+
     return `The following CSV file is loaded from ${path}
 \`\`\`csv
 ${context}

apps/simple/directory.ts

@@ -1,24 +0,0 @@
-import { SimpleDirectoryReader } from "llamaindex";
-
-function callback(
-  category: string,
-  name: string,
-  status: any,
-  message?: string,
-): boolean {
-  console.log(category, name, status, message);
-  if (name.endsWith(".pdf")) {
-    console.log("I DON'T WANT PDF FILES!");
-    return false;
-  }
-  return true;
-}
-
-async function main() {
-  // Load page
-  const reader = new SimpleDirectoryReader(callback);
-  const params = { directoryPath: "./data" };
-  await reader.loadData(params);
-}
-
-main().catch(console.error);

apps/simple/html.ts

@@ -1,21 +0,0 @@
-import { HTMLReader, VectorStoreIndex } from "llamaindex";
-
-async function main() {
-  // Load page
-  const reader = new HTMLReader();
-  const documents = await reader.loadData("data/18-1_Changelog.html");
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  const index = await VectorStoreIndex.fromDocuments(documents);
-
-  // Query the index
-  const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What were the notable changes in 18.1?",
-  );
-
-  // Output response
-  console.log(response.toString());
-}
-
-main().catch(console.error);

apps/simple/keywordIndex.ts

@@ -1,32 +0,0 @@
-import {
-  Document,
-  KeywordTableIndex,
-  KeywordTableRetrieverMode,
-} from "llamaindex";
-import essay from "./essay";
-
-async function main() {
-  const document = new Document({ text: essay, id_: "essay" });
-  const index = await KeywordTableIndex.fromDocuments([document]);
-
-  const allModes: KeywordTableRetrieverMode[] = [
-    KeywordTableRetrieverMode.DEFAULT,
-    KeywordTableRetrieverMode.SIMPLE,
-    KeywordTableRetrieverMode.RAKE,
-  ];
-  allModes.forEach(async (mode) => {
-    const queryEngine = index.asQueryEngine({
-      retriever: index.asRetriever({
-        mode,
-      }),
-    });
-    const response = await queryEngine.query(
-      "What did the author do growing up?",
-    );
-    console.log(response.toString());
-  });
-}
-
-main().catch((e: Error) => {
-  console.error(e, e.stack);
-});

apps/simple/listIndex.ts

@@ -1,9 +1,9 @@
 import {
   Document,
-  SimpleNodeParser,
-  SummaryIndex,
-  SummaryRetrieverMode,
+  ListIndex,
+  ListRetrieverMode,
   serviceContextFromDefaults,
+  SimpleNodeParser,
 } from "llamaindex";
 import essay from "./essay";
 
@@ -14,11 +14,9 @@ async function main() {
     }),
   });
   const document = new Document({ text: essay, id_: "essay" });
-  const index = await SummaryIndex.fromDocuments([document], {
-    serviceContext,
-  });
+  const index = await ListIndex.fromDocuments([document], { serviceContext });
   const queryEngine = index.asQueryEngine({
-    retriever: index.asRetriever({ mode: SummaryRetrieverMode.LLM }),
+    retriever: index.asRetriever({ mode: ListRetrieverMode.LLM }),
   });
   const response = await queryEngine.query(
     "What did the author do growing up?",

apps/simple/llmStream.ts

@@ -1,47 +0,0 @@
-import { ChatMessage, SimpleChatEngine } from "llamaindex";
-import { stdin as input, stdout as output } from "node:process";
-import readline from "node:readline/promises";
-import { Anthropic } from "../../packages/core/src/llm/LLM";
-
-async function main() {
-  const query: string = `
-Where is Istanbul?
-  `;
-
-  // const llm = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.1 });
-  const llm = new Anthropic();
-  const message: ChatMessage = { content: query, role: "user" };
-
-  //TODO: Add callbacks later
-
-  //Stream Complete
-  //Note: Setting streaming flag to true or false will auto-set your return type to
-  //either an AsyncGenerator or a Response.
-  // Omitting the streaming flag automatically sets streaming to false
-
-  const chatEngine: SimpleChatEngine = new SimpleChatEngine({
-    chatHistory: undefined,
-    llm: llm,
-  });
-
-  const rl = readline.createInterface({ input, output });
-  while (true) {
-    const query = await rl.question("Query: ");
-
-    if (!query) {
-      break;
-    }
-
-    //Case 1: .chat(query, undefined, true) => Stream
-    //Case 2: .chat(query, undefined, false) => Response object
-    //Case 3: .chat(query, undefined) => Response object
-    const chatStream = await chatEngine.chat(query, undefined, true);
-    var accumulated_result = "";
-    for await (const part of chatStream) {
-      accumulated_result += part;
-      process.stdout.write(part);
-    }
-  }
-}
-
-main();

apps/simple/lowlevel.ts

@@ -1,9 +1,9 @@
 import {
   Document,
+  TextNode,
   NodeWithScore,
   ResponseSynthesizer,
   SimpleNodeParser,
-  TextNode,
 } from "llamaindex";
 
 (async () => {
@@ -29,7 +29,7 @@ import {
 
   const response = await responseSynthesizer.synthesize(
     "What age am I?",
-    nodesWithScore,
+    nodesWithScore
   );
   console.log(response.response);
 })();

apps/simple/markdown.ts

@@ -1,20 +0,0 @@
-import { MarkdownReader, VectorStoreIndex } from "llamaindex";
-
-async function main() {
-  // Load Markdown file
-  const reader = new MarkdownReader();
-  const documents = await reader.loadData("node_modules/llamaindex/README.md");
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  const index = await VectorStoreIndex.fromDocuments(documents);
-
-  // Query the index
-  const queryEngine = index.asQueryEngine();
-
-  const response = await queryEngine.query("What does the example code do?");
-
-  // Output response
-  console.log(response.toString());
-}
-
-main().catch(console.error);

apps/simple/mongo.ts

@@ -1,68 +0,0 @@
-import { MongoClient } from "mongodb";
-import { Document } from "../../packages/core/src/Node";
-import { VectorStoreIndex } from "../../packages/core/src/indices";
-import { SimpleMongoReader } from "../../packages/core/src/readers/SimpleMongoReader";
-
-import { stdin as input, stdout as output } from "node:process";
-import readline from "node:readline/promises";
-
-async function main() {
-  //Dummy test code
-  const query: object = { _id: "waldo" };
-  const options: object = {};
-  const projections: object = { embedding: 0 };
-  const limit: number = Infinity;
-  const uri: string = process.env.MONGODB_URI ?? "fake_uri";
-  const client: MongoClient = new MongoClient(uri);
-
-  //Where the real code starts
-  const MR = new SimpleMongoReader(client);
-  const documents: Document[] = await MR.loadData(
-    "data",
-    "posts",
-    1,
-    {},
-    options,
-    projections,
-  );
-
-  //
-  //If you need to look at low-level details of
-  // a queryEngine (for example, needing to check each individual node)
-  //
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  // var storageContext = await storageContextFromDefaults({});
-  // var serviceContext = serviceContextFromDefaults({});
-  // const docStore = storageContext.docStore;
-
-  // for (const doc of documents) {
-  //   docStore.setDocumentHash(doc.id_, doc.hash);
-  // }
-  // const nodes = serviceContext.nodeParser.getNodesFromDocuments(documents);
-  // console.log(nodes);
-
-  //
-  //Making Vector Store from documents
-  //
-
-  const index = await VectorStoreIndex.fromDocuments(documents);
-  // Create query engine
-  const queryEngine = index.asQueryEngine();
-
-  const rl = readline.createInterface({ input, output });
-  while (true) {
-    const query = await rl.question("Query: ");
-
-    if (!query) {
-      break;
-    }
-
-    const response = await queryEngine.query(query);
-
-    // Output response
-    console.log(response.toString());
-  }
-}
-
-main();

apps/simple/notion.ts

@@ -1,89 +0,0 @@
-import { Client } from "@notionhq/client";
-import { program } from "commander";
-import { NotionReader, VectorStoreIndex } from "llamaindex";
-import { stdin as input, stdout as output } from "node:process";
-// readline/promises is still experimental so not in @types/node yet
-// @ts-ignore
-import readline from "node:readline/promises";
-
-program
-  .argument("[page]", "Notion page id (must be provided)")
-  .action(async (page, _options, command) => {
-    // Initializing a client
-
-    if (!process.env.NOTION_TOKEN) {
-      console.log(
-        "No NOTION_TOKEN found in environment variables. You will need to register an integration https://www.notion.com/my-integrations and put it in your NOTION_TOKEN environment variable.",
-      );
-      return;
-    }
-
-    const notion = new Client({
-      auth: process.env.NOTION_TOKEN,
-    });
-
-    if (!page) {
-      const response = await notion.search({
-        filter: {
-          value: "page",
-          property: "object",
-        },
-        sort: {
-          direction: "descending",
-          timestamp: "last_edited_time",
-        },
-      });
-
-      const { results } = response;
-
-      if (results.length === 0) {
-        console.log(
-          "No pages found. You will need to share it with your integration. (tap the three dots on the top right, find Add connections, and add your integration)",
-        );
-        return;
-      } else {
-        const pages = results
-          .map((result) => {
-            if (!("url" in result)) {
-              return null;
-            }
-
-            return {
-              id: result.id,
-              url: result.url,
-            };
-          })
-          .filter((page) => page !== null);
-        console.log("Found pages:");
-        console.table(pages);
-        console.log(`To run, run ts-node ${command.name()} [page id]`);
-        return;
-      }
-    }
-
-    const reader = new NotionReader({ client: notion });
-    const documents = await reader.loadData(page);
-    console.log(documents);
-
-    // Split text and create embeddings. Store them in a VectorStoreIndex
-    const index = await VectorStoreIndex.fromDocuments(documents);
-
-    // Create query engine
-    const queryEngine = index.asQueryEngine();
-
-    const rl = readline.createInterface({ input, output });
-    while (true) {
-      const query = await rl.question("Query: ");
-
-      if (!query) {
-        break;
-      }
-
-      const response = await queryEngine.query(query);
-
-      // Output response
-      console.log(response.toString());
-    }
-  });
-
-program.parse();

apps/simple/openai.ts

@@ -1,7 +1,14 @@
 import { OpenAI } from "llamaindex";
 
 (async () => {
-  const llm = new OpenAI({ model: "gpt-4-1106-preview", temperature: 0.1 });
+  const llm = new OpenAI({
+    model: "gpt-3.5-turbo",
+    temperature: 0.1,
+    additionalChatOptions: { frequency_penalty: 0.1 },
+    additionalSessionOptions: {
+      defaultHeaders: { "X-Test-Header-Please-Ignore": "true" },
+    },
+  });
 
   // complete api
   const response1 = await llm.complete("How are you?");
@@ -9,7 +16,7 @@ import { OpenAI } from "llamaindex";
 
   // chat api
   const response2 = await llm.chat([
-    { content: "Tell me a joke.", role: "user" },
+    { content: "Tell me a joke!", role: "user" },
   ]);
   console.log(response2.message.content);
 })();

apps/simple/package.json

@@ -1,16 +1,12 @@
 {
-  "version": "0.0.33",
+  "version": "0.0.20",
   "private": true,
   "name": "simple",
   "dependencies": {
-    "@notionhq/client": "^2.2.13",
-    "@pinecone-database/pinecone": "^1.1.2",
-    "commander": "^11.1.0",
     "llamaindex": "workspace:*"
   },
   "devDependencies": {
-    "@types/node": "^18.18.6",
-    "ts-node": "^10.9.1"
+    "@types/node": "^18.17.6"
   },
   "scripts": {
     "lint": "eslint ."

apps/simple/portkey.ts

@@ -1,23 +0,0 @@
-import { Portkey } from "llamaindex";
-
-(async () => {
-  const llms = [{}];
-  const portkey = new Portkey({
-    mode: "single",
-    llms: [
-      {
-        provider: "anyscale",
-        virtual_key: "anyscale-3b3c04",
-        model: "meta-llama/Llama-2-13b-chat-hf",
-        max_tokens: 2000,
-      },
-    ],
-  });
-  const result = portkey.stream_chat([
-    { role: "system", content: "You are a helpful assistant." },
-    { role: "user", content: "Tell me a joke." },
-  ]);
-  for await (const res of result) {
-    process.stdout.write(res);
-  }
-})();

apps/simple/vectorIndexCustomize.ts

@@ -3,7 +3,6 @@ import {
   OpenAI,
   RetrieverQueryEngine,
   serviceContextFromDefaults,
-  SimilarityPostprocessor,
   VectorStoreIndex,
 } from "llamaindex";
 import essay from "./essay";
@@ -13,7 +12,7 @@ async function main() {
   const document = new Document({ text: essay, id_: "essay" });
 
   const serviceContext = serviceContextFromDefaults({
-    llm: new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.1 }),
+    llm: new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.0 }),
   });
 
   const index = await VectorStoreIndex.fromDocuments([document], {
@@ -22,16 +21,8 @@ async function main() {
 
   const retriever = index.asRetriever();
   retriever.similarityTopK = 5;
-  const nodePostprocessor = new SimilarityPostprocessor({
-    similarityCutoff: 0.7,
-  });
   // TODO: cannot pass responseSynthesizer into retriever query engine
-  const queryEngine = new RetrieverQueryEngine(
-    retriever,
-    undefined,
-    undefined,
-    [nodePostprocessor],
-  );
+  const queryEngine = new RetrieverQueryEngine(retriever);
 
   const response = await queryEngine.query(
     "What did the author do growing up?",

apps/simple/vectorIndexFromVectorStore.ts

@@ -1,197 +0,0 @@
-import {
-  OpenAI,
-  ResponseSynthesizer,
-  RetrieverQueryEngine,
-  serviceContextFromDefaults,
-  TextNode,
-  TreeSummarize,
-  VectorIndexRetriever,
-  VectorStore,
-  VectorStoreIndex,
-  VectorStoreQuery,
-  VectorStoreQueryResult,
-} from "llamaindex";
-
-import { Index, Pinecone, RecordMetadata } from "@pinecone-database/pinecone";
-
-/**
- * Please do not use this class in production; it's only for demonstration purposes.
- */
-class PineconeVectorStore<T extends RecordMetadata = RecordMetadata>
-  implements VectorStore
-{
-  storesText = true;
-  isEmbeddingQuery = false;
-
-  indexName!: string;
-  pineconeClient!: Pinecone;
-  index!: Index<T>;
-
-  constructor({ indexName, client }: { indexName: string; client: Pinecone }) {
-    this.indexName = indexName;
-    this.pineconeClient = client;
-    this.index = client.index<T>(indexName);
-  }
-
-  client() {
-    return this.pineconeClient;
-  }
-
-  async query(
-    query: VectorStoreQuery,
-    kwargs?: any,
-  ): Promise<VectorStoreQueryResult> {
-    let queryEmbedding: number[] = [];
-    if (query.queryEmbedding) {
-      if (typeof query.alpha === "number") {
-        const alpha = query.alpha;
-        queryEmbedding = query.queryEmbedding.map((v) => v * alpha);
-      } else {
-        queryEmbedding = query.queryEmbedding;
-      }
-    }
-
-    // Current LlamaIndexTS implementation only support exact match filter, so we use kwargs instead.
-    const filter = kwargs?.filter || {};
-
-    const response = await this.index.query({
-      filter,
-      vector: queryEmbedding,
-      topK: query.similarityTopK,
-      includeValues: true,
-      includeMetadata: true,
-    });
-
-    console.log(
-      `Numbers of vectors returned by Pinecone after preFilters are applied: ${
-        response?.matches?.length || 0
-      }.`,
-    );
-
-    const topKIds: string[] = [];
-    const topKNodes: TextNode[] = [];
-    const topKScores: number[] = [];
-
-    const metadataToNode = (metadata?: T): Partial<TextNode> => {
-      if (!metadata) {
-        throw new Error("metadata is undefined.");
-      }
-
-      const nodeContent = metadata["_node_content"];
-      if (!nodeContent) {
-        throw new Error("nodeContent is undefined.");
-      }
-
-      if (typeof nodeContent !== "string") {
-        throw new Error("nodeContent is not a string.");
-      }
-
-      return JSON.parse(nodeContent);
-    };
-
-    if (response.matches) {
-      for (const match of response.matches) {
-        const node = new TextNode({
-          ...metadataToNode(match.metadata),
-          embedding: match.values,
-        });
-
-        topKIds.push(match.id);
-        topKNodes.push(node);
-        topKScores.push(match.score ?? 0);
-      }
-    }
-
-    const result = {
-      ids: topKIds,
-      nodes: topKNodes,
-      similarities: topKScores,
-    };
-
-    return result;
-  }
-
-  add(): Promise<string[]> {
-    return Promise.resolve([]);
-  }
-
-  delete(): Promise<void> {
-    throw new Error("Method `delete` not implemented.");
-  }
-
-  persist(): Promise<void> {
-    throw new Error("Method `persist` not implemented.");
-  }
-}
-
-/**
- * The goal of this example is to show how to use Pinecone as a vector store
- * for LlamaIndexTS with(out) preFilters.
- *
- * It should not be used in production like that,
- * as you might want to find a proper PineconeVectorStore implementation.
- */
-async function main() {
-  process.env.PINECONE_API_KEY = "Your Pinecone API Key.";
-  process.env.PINECONE_ENVIRONMENT = "Your Pinecone Environment.";
-  process.env.PINECONE_PROJECT_ID = "Your Pinecone Project ID.";
-  process.env.PINECONE_INDEX_NAME = "Your Pinecone Index Name.";
-  process.env.OPENAI_API_KEY = "Your OpenAI API Key.";
-  process.env.OPENAI_API_ORGANIZATION = "Your OpenAI API Organization.";
-
-  const getPineconeVectorStore = async () => {
-    return new PineconeVectorStore({
-      indexName: process.env.PINECONE_INDEX_NAME || "index-name",
-      client: new Pinecone(),
-    });
-  };
-
-  const getServiceContext = () => {
-    const openAI = new OpenAI({
-      model: "gpt-4",
-      apiKey: process.env.OPENAI_API_KEY,
-    });
-
-    return serviceContextFromDefaults({
-      llm: openAI,
-    });
-  };
-
-  const getQueryEngine = async (filter: unknown) => {
-    const vectorStore = await getPineconeVectorStore();
-    const serviceContext = getServiceContext();
-
-    const vectorStoreIndex = await VectorStoreIndex.fromVectorStore(
-      vectorStore,
-      serviceContext,
-    );
-
-    const retriever = new VectorIndexRetriever({
-      index: vectorStoreIndex,
-      similarityTopK: 500,
-    });
-
-    const responseSynthesizer = new ResponseSynthesizer({
-      serviceContext,
-      responseBuilder: new TreeSummarize(serviceContext),
-    });
-
-    return new RetrieverQueryEngine(retriever, responseSynthesizer, {
-      filter,
-    });
-  };
-
-  // whatever is a key from your metadata
-  const queryEngine = await getQueryEngine({
-    whatever: {
-      $gte: 1,
-      $lte: 100,
-    },
-  });
-
-  const response = await queryEngine.query("How many results do you have?");
-
-  console.log(response.toString());
-}
-
-main().catch(console.error);

apps/simple/vision.ts

@@ -1,15 +0,0 @@
-import { OpenAI } from "llamaindex";
-
-(async () => {
-  const llm = new OpenAI({ model: "gpt-4-vision-preview", temperature: 0.1 });
-
-  // complete api
-  const response1 = await llm.complete("How are you?");
-  console.log(response1.message.content);
-
-  // chat api
-  const response2 = await llm.chat([
-    { content: "Tell me a joke!", role: "user" },
-  ]);
-  console.log(response2.message.content);
-})();

examples/csv.ts

@@ -4,6 +4,7 @@ import {
   PapaCSVReader,
   ResponseSynthesizer,
   serviceContextFromDefaults,
+  SimplePrompt,
   VectorStoreIndex,
 } from "llamaindex";
 
@@ -22,7 +23,9 @@ async function main() {
     serviceContext,
   });
 
-  const csvPrompt = ({ context = "", query = "" }) => {
+  const csvPrompt: SimplePrompt = (input) => {
+    const { context = "", query = "" } = input;
+
     return `The following CSV file is loaded from ${path}
 \`\`\`csv
 ${context}

examples/directory.ts

@@ -1,24 +0,0 @@
-import { SimpleDirectoryReader } from "llamaindex";
-
-function callback(
-  category: string,
-  name: string,
-  status: any,
-  message?: string,
-): boolean {
-  console.log(category, name, status, message);
-  if (name.endsWith(".pdf")) {
-    console.log("I DON'T WANT PDF FILES!");
-    return false;
-  }
-  return true;
-}
-
-async function main() {
-  // Load page
-  const reader = new SimpleDirectoryReader(callback);
-  const params = { directoryPath: "./data" };
-  await reader.loadData(params);
-}
-
-main().catch(console.error);

examples/html.ts

@@ -1,21 +0,0 @@
-import { HTMLReader, VectorStoreIndex } from "llamaindex";
-
-async function main() {
-  // Load page
-  const reader = new HTMLReader();
-  const documents = await reader.loadData("data/18-1_Changelog.html");
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  const index = await VectorStoreIndex.fromDocuments(documents);
-
-  // Query the index
-  const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What were the notable changes in 18.1?",
-  );
-
-  // Output response
-  console.log(response.toString());
-}
-
-main().catch(console.error);

examples/keywordIndex.ts

@@ -1,32 +0,0 @@
-import {
-  Document,
-  KeywordTableIndex,
-  KeywordTableRetrieverMode,
-} from "llamaindex";
-import essay from "./essay";
-
-async function main() {
-  const document = new Document({ text: essay, id_: "essay" });
-  const index = await KeywordTableIndex.fromDocuments([document]);
-
-  const allModes: KeywordTableRetrieverMode[] = [
-    KeywordTableRetrieverMode.DEFAULT,
-    KeywordTableRetrieverMode.SIMPLE,
-    KeywordTableRetrieverMode.RAKE,
-  ];
-  allModes.forEach(async (mode) => {
-    const queryEngine = index.asQueryEngine({
-      retriever: index.asRetriever({
-        mode,
-      }),
-    });
-    const response = await queryEngine.query(
-      "What did the author do growing up?",
-    );
-    console.log(response.toString());
-  });
-}
-
-main().catch((e: Error) => {
-  console.error(e, e.stack);
-});

examples/listIndex.ts

@@ -1,9 +1,9 @@
 import {
   Document,
-  SimpleNodeParser,
-  SummaryIndex,
-  SummaryRetrieverMode,
+  ListIndex,
+  ListRetrieverMode,
   serviceContextFromDefaults,
+  SimpleNodeParser,
 } from "llamaindex";
 import essay from "./essay";
 
@@ -14,11 +14,9 @@ async function main() {
     }),
   });
   const document = new Document({ text: essay, id_: "essay" });
-  const index = await SummaryIndex.fromDocuments([document], {
-    serviceContext,
-  });
+  const index = await ListIndex.fromDocuments([document], { serviceContext });
   const queryEngine = index.asQueryEngine({
-    retriever: index.asRetriever({ mode: SummaryRetrieverMode.LLM }),
+    retriever: index.asRetriever({ mode: ListRetrieverMode.LLM }),
   });
   const response = await queryEngine.query(
     "What did the author do growing up?",

examples/llmStream.ts

@@ -1,47 +0,0 @@
-import { ChatMessage, SimpleChatEngine } from "llamaindex";
-import { stdin as input, stdout as output } from "node:process";
-import readline from "node:readline/promises";
-import { Anthropic } from "../../packages/core/src/llm/LLM";
-
-async function main() {
-  const query: string = `
-Where is Istanbul?
-  `;
-
-  // const llm = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.1 });
-  const llm = new Anthropic();
-  const message: ChatMessage = { content: query, role: "user" };
-
-  //TODO: Add callbacks later
-
-  //Stream Complete
-  //Note: Setting streaming flag to true or false will auto-set your return type to
-  //either an AsyncGenerator or a Response.
-  // Omitting the streaming flag automatically sets streaming to false
-
-  const chatEngine: SimpleChatEngine = new SimpleChatEngine({
-    chatHistory: undefined,
-    llm: llm,
-  });
-
-  const rl = readline.createInterface({ input, output });
-  while (true) {
-    const query = await rl.question("Query: ");
-
-    if (!query) {
-      break;
-    }
-
-    //Case 1: .chat(query, undefined, true) => Stream
-    //Case 2: .chat(query, undefined, false) => Response object
-    //Case 3: .chat(query, undefined) => Response object
-    const chatStream = await chatEngine.chat(query, undefined, true);
-    var accumulated_result = "";
-    for await (const part of chatStream) {
-      accumulated_result += part;
-      process.stdout.write(part);
-    }
-  }
-}
-
-main();

examples/lowlevel.ts

@@ -1,9 +1,9 @@
 import {
   Document,
+  TextNode,
   NodeWithScore,
   ResponseSynthesizer,
   SimpleNodeParser,
-  TextNode,
 } from "llamaindex";
 
 (async () => {
@@ -29,7 +29,7 @@ import {
 
   const response = await responseSynthesizer.synthesize(
     "What age am I?",
-    nodesWithScore,
+    nodesWithScore
   );
   console.log(response.response);
 })();

examples/markdown.ts

@@ -1,20 +0,0 @@
-import { MarkdownReader, VectorStoreIndex } from "llamaindex";
-
-async function main() {
-  // Load Markdown file
-  const reader = new MarkdownReader();
-  const documents = await reader.loadData("node_modules/llamaindex/README.md");
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  const index = await VectorStoreIndex.fromDocuments(documents);
-
-  // Query the index
-  const queryEngine = index.asQueryEngine();
-
-  const response = await queryEngine.query("What does the example code do?");
-
-  // Output response
-  console.log(response.toString());
-}
-
-main().catch(console.error);

examples/mongo.ts

@@ -1,68 +0,0 @@
-import { MongoClient } from "mongodb";
-import { Document } from "../../packages/core/src/Node";
-import { VectorStoreIndex } from "../../packages/core/src/indices";
-import { SimpleMongoReader } from "../../packages/core/src/readers/SimpleMongoReader";
-
-import { stdin as input, stdout as output } from "node:process";
-import readline from "node:readline/promises";
-
-async function main() {
-  //Dummy test code
-  const query: object = { _id: "waldo" };
-  const options: object = {};
-  const projections: object = { embedding: 0 };
-  const limit: number = Infinity;
-  const uri: string = process.env.MONGODB_URI ?? "fake_uri";
-  const client: MongoClient = new MongoClient(uri);
-
-  //Where the real code starts
-  const MR = new SimpleMongoReader(client);
-  const documents: Document[] = await MR.loadData(
-    "data",
-    "posts",
-    1,
-    {},
-    options,
-    projections,
-  );
-
-  //
-  //If you need to look at low-level details of
-  // a queryEngine (for example, needing to check each individual node)
-  //
-
-  // Split text and create embeddings. Store them in a VectorStoreIndex
-  // var storageContext = await storageContextFromDefaults({});
-  // var serviceContext = serviceContextFromDefaults({});
-  // const docStore = storageContext.docStore;
-
-  // for (const doc of documents) {
-  //   docStore.setDocumentHash(doc.id_, doc.hash);
-  // }
-  // const nodes = serviceContext.nodeParser.getNodesFromDocuments(documents);
-  // console.log(nodes);
-
-  //
-  //Making Vector Store from documents
-  //
-
-  const index = await VectorStoreIndex.fromDocuments(documents);
-  // Create query engine
-  const queryEngine = index.asQueryEngine();
-
-  const rl = readline.createInterface({ input, output });
-  while (true) {
-    const query = await rl.question("Query: ");
-
-    if (!query) {
-      break;
-    }
-
-    const response = await queryEngine.query(query);
-
-    // Output response
-    console.log(response.toString());
-  }
-}
-
-main();

examples/notion.ts

@@ -1,89 +0,0 @@
-import { Client } from "@notionhq/client";
-import { program } from "commander";
-import { NotionReader, VectorStoreIndex } from "llamaindex";
-import { stdin as input, stdout as output } from "node:process";
-// readline/promises is still experimental so not in @types/node yet
-// @ts-ignore
-import readline from "node:readline/promises";
-
-program
-  .argument("[page]", "Notion page id (must be provided)")
-  .action(async (page, _options, command) => {
-    // Initializing a client
-
-    if (!process.env.NOTION_TOKEN) {
-      console.log(
-        "No NOTION_TOKEN found in environment variables. You will need to register an integration https://www.notion.com/my-integrations and put it in your NOTION_TOKEN environment variable.",
-      );
-      return;
-    }
-
-    const notion = new Client({
-      auth: process.env.NOTION_TOKEN,
-    });
-
-    if (!page) {
-      const response = await notion.search({
-        filter: {
-          value: "page",
-          property: "object",
-        },
-        sort: {
-          direction: "descending",
-          timestamp: "last_edited_time",
-        },
-      });
-
-      const { results } = response;
-
-      if (results.length === 0) {
-        console.log(
-          "No pages found. You will need to share it with your integration. (tap the three dots on the top right, find Add connections, and add your integration)",
-        );
-        return;
-      } else {
-        const pages = results
-          .map((result) => {
-            if (!("url" in result)) {
-              return null;
-            }
-
-            return {
-              id: result.id,
-              url: result.url,
-            };
-          })
-          .filter((page) => page !== null);
-        console.log("Found pages:");
-        console.table(pages);
-        console.log(`To run, run ts-node ${command.name()} [page id]`);
-        return;
-      }
-    }
-
-    const reader = new NotionReader({ client: notion });
-    const documents = await reader.loadData(page);
-    console.log(documents);
-
-    // Split text and create embeddings. Store them in a VectorStoreIndex
-    const index = await VectorStoreIndex.fromDocuments(documents);
-
-    // Create query engine
-    const queryEngine = index.asQueryEngine();
-
-    const rl = readline.createInterface({ input, output });
-    while (true) {
-      const query = await rl.question("Query: ");
-
-      if (!query) {
-        break;
-      }
-
-      const response = await queryEngine.query(query);
-
-      // Output response
-      console.log(response.toString());
-    }
-  });
-
-program.parse();

examples/openai.ts

@@ -1,15 +1,13 @@
 import { OpenAI } from "llamaindex";
 
 (async () => {
-  const llm = new OpenAI({ model: "gpt-4-1106-preview", temperature: 0.1 });
-
+  const llm = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.0 });
+  
   // complete api
   const response1 = await llm.complete("How are you?");
   console.log(response1.message.content);
 
   // chat api
-  const response2 = await llm.chat([
-    { content: "Tell me a joke.", role: "user" },
-  ]);
+  const response2 = await llm.chat([{ content: "Tell me a joke!", role: "user" }]);
   console.log(response2.message.content);
 })();

examples/portkey.ts

@@ -1,23 +0,0 @@
-import { Portkey } from "llamaindex";
-
-(async () => {
-  const llms = [{}];
-  const portkey = new Portkey({
-    mode: "single",
-    llms: [
-      {
-        provider: "anyscale",
-        virtual_key: "anyscale-3b3c04",
-        model: "meta-llama/Llama-2-13b-chat-hf",
-        max_tokens: 2000,
-      },
-    ],
-  });
-  const result = portkey.stream_chat([
-    { role: "system", content: "You are a helpful assistant." },
-    { role: "user", content: "Tell me a joke." },
-  ]);
-  for await (const res of result) {
-    process.stdout.write(res);
-  }
-})();

examples/readonly.ts

@@ -1,37 +0,0 @@
-import { execSync } from "child_process";
-import {
-  PDFReader,
-  serviceContextFromDefaults,
-  storageContextFromDefaults,
-  VectorStoreIndex,
-} from "llamaindex";
-
-const STORAGE_DIR = "./cache";
-
-async function main() {
-  // write the index to disk
-  const serviceContext = serviceContextFromDefaults({});
-  const storageContext = await storageContextFromDefaults({
-    persistDir: `${STORAGE_DIR}`,
-  });
-  const reader = new PDFReader();
-  const documents = await reader.loadData("data/brk-2022.pdf");
-  await VectorStoreIndex.fromDocuments(documents, {
-    storageContext,
-    serviceContext,
-  });
-  console.log("wrote index to disk - now trying to read it");
-  // make index dir read only
-  execSync(`chmod -R 555 ${STORAGE_DIR}`);
-  // reopen index
-  const readOnlyStorageContext = await storageContextFromDefaults({
-    persistDir: `${STORAGE_DIR}`,
-  });
-  await VectorStoreIndex.init({
-    storageContext: readOnlyStorageContext,
-    serviceContext,
-  });
-  console.log("read only index successfully opened");
-}
-
-main().catch(console.error);

examples/vectorIndexCustomize.ts

@@ -3,7 +3,6 @@ import {
   OpenAI,
   RetrieverQueryEngine,
   serviceContextFromDefaults,
-  SimilarityPostprocessor,
   VectorStoreIndex,
 } from "llamaindex";
 import essay from "./essay";
@@ -13,7 +12,7 @@ async function main() {
   const document = new Document({ text: essay, id_: "essay" });
 
   const serviceContext = serviceContextFromDefaults({
-    llm: new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.1 }),
+    llm: new OpenAI({ model: "gpt-3.5-turbo", temperature: 0.0 }),
   });
 
   const index = await VectorStoreIndex.fromDocuments([document], {
@@ -22,16 +21,8 @@ async function main() {
 
   const retriever = index.asRetriever();
   retriever.similarityTopK = 5;
-  const nodePostprocessor = new SimilarityPostprocessor({
-    similarityCutoff: 0.7,
-  });
   // TODO: cannot pass responseSynthesizer into retriever query engine
-  const queryEngine = new RetrieverQueryEngine(
-    retriever,
-    undefined,
-    undefined,
-    [nodePostprocessor],
-  );
+  const queryEngine = new RetrieverQueryEngine(retriever);
 
   const response = await queryEngine.query(
     "What did the author do growing up?",

examples/vectorIndexFromVectorStore.ts

@@ -1,197 +0,0 @@
-import {
-  OpenAI,
-  ResponseSynthesizer,
-  RetrieverQueryEngine,
-  serviceContextFromDefaults,
-  TextNode,
-  TreeSummarize,
-  VectorIndexRetriever,
-  VectorStore,
-  VectorStoreIndex,
-  VectorStoreQuery,
-  VectorStoreQueryResult,
-} from "llamaindex";
-
-import { Index, Pinecone, RecordMetadata } from "@pinecone-database/pinecone";
-
-/**
- * Please do not use this class in production; it's only for demonstration purposes.
- */
-class PineconeVectorStore<T extends RecordMetadata = RecordMetadata>
-  implements VectorStore
-{
-  storesText = true;
-  isEmbeddingQuery = false;
-
-  indexName!: string;
-  pineconeClient!: Pinecone;
-  index!: Index<T>;
-
-  constructor({ indexName, client }: { indexName: string; client: Pinecone }) {
-    this.indexName = indexName;
-    this.pineconeClient = client;
-    this.index = client.index<T>(indexName);
-  }
-
-  client() {
-    return this.pineconeClient;
-  }
-
-  async query(
-    query: VectorStoreQuery,
-    kwargs?: any,
-  ): Promise<VectorStoreQueryResult> {
-    let queryEmbedding: number[] = [];
-    if (query.queryEmbedding) {
-      if (typeof query.alpha === "number") {
-        const alpha = query.alpha;
-        queryEmbedding = query.queryEmbedding.map((v) => v * alpha);
-      } else {
-        queryEmbedding = query.queryEmbedding;
-      }
-    }
-
-    // Current LlamaIndexTS implementation only support exact match filter, so we use kwargs instead.
-    const filter = kwargs?.filter || {};
-
-    const response = await this.index.query({
-      filter,
-      vector: queryEmbedding,
-      topK: query.similarityTopK,
-      includeValues: true,
-      includeMetadata: true,
-    });
-
-    console.log(
-      `Numbers of vectors returned by Pinecone after preFilters are applied: ${
-        response?.matches?.length || 0
-      }.`,
-    );
-
-    const topKIds: string[] = [];
-    const topKNodes: TextNode[] = [];
-    const topKScores: number[] = [];
-
-    const metadataToNode = (metadata?: T): Partial<TextNode> => {
-      if (!metadata) {
-        throw new Error("metadata is undefined.");
-      }
-
-      const nodeContent = metadata["_node_content"];
-      if (!nodeContent) {
-        throw new Error("nodeContent is undefined.");
-      }
-
-      if (typeof nodeContent !== "string") {
-        throw new Error("nodeContent is not a string.");
-      }
-
-      return JSON.parse(nodeContent);
-    };
-
-    if (response.matches) {
-      for (const match of response.matches) {
-        const node = new TextNode({
-          ...metadataToNode(match.metadata),
-          embedding: match.values,
-        });
-
-        topKIds.push(match.id);
-        topKNodes.push(node);
-        topKScores.push(match.score ?? 0);
-      }
-    }
-
-    const result = {
-      ids: topKIds,
-      nodes: topKNodes,
-      similarities: topKScores,
-    };
-
-    return result;
-  }
-
-  add(): Promise<string[]> {
-    return Promise.resolve([]);
-  }
-
-  delete(): Promise<void> {
-    throw new Error("Method `delete` not implemented.");
-  }
-
-  persist(): Promise<void> {
-    throw new Error("Method `persist` not implemented.");
-  }
-}
-
-/**
- * The goal of this example is to show how to use Pinecone as a vector store
- * for LlamaIndexTS with(out) preFilters.
- *
- * It should not be used in production like that,
- * as you might want to find a proper PineconeVectorStore implementation.
- */
-async function main() {
-  process.env.PINECONE_API_KEY = "Your Pinecone API Key.";
-  process.env.PINECONE_ENVIRONMENT = "Your Pinecone Environment.";
-  process.env.PINECONE_PROJECT_ID = "Your Pinecone Project ID.";
-  process.env.PINECONE_INDEX_NAME = "Your Pinecone Index Name.";
-  process.env.OPENAI_API_KEY = "Your OpenAI API Key.";
-  process.env.OPENAI_API_ORGANIZATION = "Your OpenAI API Organization.";
-
-  const getPineconeVectorStore = async () => {
-    return new PineconeVectorStore({
-      indexName: process.env.PINECONE_INDEX_NAME || "index-name",
-      client: new Pinecone(),
-    });
-  };
-
-  const getServiceContext = () => {
-    const openAI = new OpenAI({
-      model: "gpt-4",
-      apiKey: process.env.OPENAI_API_KEY,
-    });
-
-    return serviceContextFromDefaults({
-      llm: openAI,
-    });
-  };
-
-  const getQueryEngine = async (filter: unknown) => {
-    const vectorStore = await getPineconeVectorStore();
-    const serviceContext = getServiceContext();
-
-    const vectorStoreIndex = await VectorStoreIndex.fromVectorStore(
-      vectorStore,
-      serviceContext,
-    );
-
-    const retriever = new VectorIndexRetriever({
-      index: vectorStoreIndex,
-      similarityTopK: 500,
-    });
-
-    const responseSynthesizer = new ResponseSynthesizer({
-      serviceContext,
-      responseBuilder: new TreeSummarize(serviceContext),
-    });
-
-    return new RetrieverQueryEngine(retriever, responseSynthesizer, {
-      filter,
-    });
-  };
-
-  // whatever is a key from your metadata
-  const queryEngine = await getQueryEngine({
-    whatever: {
-      $gte: 1,
-      $lte: 100,
-    },
-  });
-
-  const response = await queryEngine.query("How many results do you have?");
-
-  console.log(response.toString());
-}
-
-main().catch(console.error);

examples/vision.ts

@@ -1,15 +0,0 @@
-import { OpenAI } from "llamaindex";
-
-(async () => {
-  const llm = new OpenAI({ model: "gpt-4-vision-preview", temperature: 0.1 });
-
-  // complete api
-  const response1 = await llm.complete("How are you?");
-  console.log(response1.message.content);
-
-  // chat api
-  const response2 = await llm.chat([
-    { content: "Tell me a joke!", role: "user" },
-  ]);
-  console.log(response2.message.content);
-})();

package.json

@@ -3,7 +3,7 @@
   "scripts": {
     "build": "turbo run build",
     "dev": "turbo run dev",
-    "format": "prettier --write \"**/*.{js,jsx,ts,tsx,md}\"",
+    "format": "prettier --write \"**/*.{ts,tsx,md}\"",
     "lint": "turbo run lint",
     "prepare": "husky install",
     "test": "turbo run test",
@@ -11,27 +11,24 @@
     "publish-snapshot": "turbo run build lint test && changeset version --snapshot && changeset publish"
   },
   "devDependencies": {
-    "@changesets/cli": "^2.26.2",
-    "@turbo/gen": "^1.10.16",
-    "@types/jest": "^29.5.8",
-    "eslint": "^8.53.0",
+    "@turbo/gen": "^1.10.13",
+    "@types/jest": "^29.5.4",
+    "eslint": "^7.32.0",
     "eslint-config-custom": "workspace:*",
     "husky": "^8.0.3",
-    "jest": "^29.7.0",
-    "lint-staged": "^15.1.0",
-    "prettier": "^3.1.0",
-    "prettier-plugin-organize-imports": "^3.2.4",
+    "jest": "^29.6.3",
+    "prettier": "^3.0.2",
+    "prettier-plugin-organize-imports": "^3.2.3",
     "ts-jest": "^29.1.1",
-    "turbo": "^1.10.16"
+    "turbo": "^1.10.13"
+  },
+  "packageManager": "pnpm@7.15.0",
+  "dependencies": {
+    "@changesets/cli": "^2.26.2"
   },
-  "packageManager": "pnpm@8.10.5+sha256.a4bd9bb7b48214bbfcd95f264bd75bb70d100e5d4b58808f5cd6ab40c6ac21c5",
   "pnpm": {
     "overrides": {
-      "trim": "1.0.1",
-      "@babel/traverse": "7.23.2"
+      "trim": "1.0.1"
     }
-  },
-  "lint-staged": {
-    "*.{js,jsx,ts,tsx,md}": "prettier --write"
   }
 }

packages/core/CHANGELOG.md

@@ -1,100 +1,5 @@
 # llamaindex
 
-## 0.0.35
-
-### Patch Changes
-
-- 63f2108: Add multimodal support (thanks @marcusschiesser)
-
-## 0.0.34
-
-### Patch Changes
-
-- 2a27e21: Add support for gpt-3.5-turbo-1106
-
-## 0.0.33
-
-### Patch Changes
-
-- 5e2e92c: gpt-4-1106-preview and gpt-4-vision-preview from OpenAI dev day
-
-## 0.0.32
-
-### Patch Changes
-
-- 90c0b83: Add HTMLReader (thanks @mtutty)
-- dfd22aa: Add observer/filter to the SimpleDirectoryReader (thanks @mtutty)
-
-## 0.0.31
-
-### Patch Changes
-
-- 6c55b2d: Give HistoryChatEngine pluggable options (thanks @marcusschiesser)
-- 8aa8c65: Add SimilarityPostProcessor (thanks @TomPenguin)
-- 6c55b2d: Added LLMMetadata (thanks @marcusschiesser)
-
-## 0.0.30
-
-### Patch Changes
-
-- 139abad: Streaming improvements including Anthropic (thanks @kkang2097)
-- 139abad: Portkey integration (Thank you @noble-varghese)
-- eb0e994: Add export for PromptHelper (thanks @zigamall)
-- eb0e994: Publish ESM module again
-- 139abad: Pinecone demo (thanks @Einsenhorn)
-
-## 0.0.29
-
-### Patch Changes
-
-- a52143b: Added DocxReader for Word documents (thanks @jayantasamaddar)
-- 1b7fd95: Updated OpenAI streaming (thanks @kkang2097)
-- 0db3f41: Migrated to Tiktoken lite, which hopefully fixes the Windows issue
-
-## 0.0.28
-
-### Patch Changes
-
-- 96bb657: Typesafe metadata (thanks @TomPenguin)
-- 96bb657: MongoReader (thanks @kkang2097)
-- 837854d: Make OutputParser less strict and add tests (Thanks @kkang2097)
-
-## 0.0.27
-
-### Patch Changes
-
-- 4a5591b: Chat History summarization (thanks @marcusschiesser)
-- 4a5591b: Notion database support (thanks @TomPenguin)
-- 4a5591b: KeywordIndex (thanks @swk777)
-
-## 0.0.26
-
-### Patch Changes
-
-- 5bb55bc: Add notion loader (thank you @TomPenguin!)
-
-## 0.0.25
-
-### Patch Changes
-
-- e21eca2: OpenAI 4.3.1 and Anthropic 0.6.2
-- 40a8f07: Update READMEs (thanks @andfk)
-- 40a8f07: Bug: missing exports from storage (thanks @aashutoshrathi)
-
-## 0.0.24
-
-### Patch Changes
-
-- e4af7b3: Renamed ListIndex to SummaryIndex to better indicate its use.
-- 259fe63: Strong types for prompts.
-
-## 0.0.23
-
-### Patch Changes
-
-- Added MetadataMode to ResponseSynthesizer (thanks @TomPenguin)
-- 9d6b2ed: Added Markdown Reader (huge shoutout to @swk777)
-
 ## 0.0.22
 
 ### Patch Changes

packages/core/README.md

@@ -20,7 +20,7 @@ In a new folder:
 export OPENAI_API_KEY="sk-......" # Replace with your key from https://platform.openai.com/account/api-keys
 pnpm init
 pnpm install typescript
-pnpm exec tsc --init # if needed
+pnpm exec tsc –-init # if needed
 pnpm install llamaindex
 pnpm install @types/node
 ```
@@ -36,7 +36,7 @@ async function main() {
   // Load essay from abramov.txt in Node
   const essay = await fs.readFile(
     "node_modules/llamaindex/examples/abramov.txt",
-    "utf-8",
+    "utf-8"
   );
 
   // Create Document object with essay
@@ -48,7 +48,7 @@ async function main() {
   // Query the index
   const queryEngine = index.asQueryEngine();
   const response = await queryEngine.query(
-    "What did the author do in college?",
+    "What did the author do in college?"
   );
 
   // Output response
@@ -61,7 +61,7 @@ main();
 Then you can run it using
 
 ```bash
-pnpx ts-node example.ts
+pnpm dlx ts-node example.ts
 ```
 
 ## Playground

packages/core/package.json

@@ -1,47 +1,34 @@
 {
   "name": "llamaindex",
-  "version": "0.0.35",
-  "license": "MIT",
+  "version": "0.0.22",
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.9.0",
-    "@notionhq/client": "^2.2.13",
-    "js-tiktoken": "^1.0.7",
+    "@anthropic-ai/sdk": "^0.6.1",
     "lodash": "^4.17.21",
-    "mammoth": "^1.6.0",
-    "md-utils-ts": "^2.0.0",
-    "mongodb": "^6.2.0",
-    "notion-md-crawler": "^0.0.2",
-    "openai": "^4.16.1",
+    "openai": "^4.2.0",
     "papaparse": "^5.4.1",
     "pdf-parse": "^1.1.1",
-    "portkey-ai": "^0.1.16",
-    "rake-modified": "^1.0.8",
-    "replicate": "^0.21.1",
-    "string-strip-html": "^13.4.3",
-    "uuid": "^9.0.1",
+    "replicate": "^0.16.1",
+    "tiktoken-node": "^0.0.6",
+    "uuid": "^9.0.0",
     "wink-nlp": "^1.14.3"
   },
   "devDependencies": {
-    "@types/lodash": "^4.14.200",
-    "@types/node": "^18.18.8",
-    "@types/papaparse": "^5.3.10",
-    "@types/pdf-parse": "^1.1.3",
-    "@types/uuid": "^9.0.6",
+    "@types/lodash": "^4.14.197",
+    "@types/node": "^18.17.9",
+    "@types/papaparse": "^5.3.8",
+    "@types/pdf-parse": "^1.1.1",
+    "@types/uuid": "^9.0.2",
     "node-stdlib-browser": "^1.2.0",
-    "tsup": "^7.2.0",
-    "typescript": "^5.2.2"
+    "tsup": "^7.2.0"
   },
   "engines": {
     "node": ">=18.0.0"
   },
   "types": "./dist/index.d.ts",
   "main": "./dist/index.js",
-  "module": "./dist/index.mjs",
-  "repository": "run-llama/LlamaIndexTS",
   "scripts": {
     "lint": "eslint .",
     "test": "jest",
-    "build": "tsup src/index.ts --format esm,cjs --dts",
-    "dev": "tsup src/index.ts --format esm,cjs --dts --watch"
+    "build": "tsup src/index.ts --format esm,cjs --dts"
   }
 }

packages/core/src/ChatEngine.ts

@@ -1,20 +1,17 @@
-import { v4 as uuidv4 } from "uuid";
-import { ChatHistory } from "./ChatHistory";
-import { NodeWithScore, TextNode } from "./Node";
+import { ChatMessage, OpenAI, ChatResponse, LLM } from "./llm/LLM";
+import { TextNode } from "./Node";
 import {
-  CondenseQuestionPrompt,
-  ContextSystemPrompt,
+  SimplePrompt,
+  contextSystemPrompt,
   defaultCondenseQuestionPrompt,
-  defaultContextSystemPrompt,
   messagesToHistoryStr,
 } from "./Prompt";
 import { BaseQueryEngine } from "./QueryEngine";
 import { Response } from "./Response";
 import { BaseRetriever } from "./Retriever";
 import { ServiceContext, serviceContextFromDefaults } from "./ServiceContext";
+import { v4 as uuidv4 } from "uuid";
 import { Event } from "./callbacks/CallbackManager";
-import { BaseNodePostprocessor } from "./indices/BaseNodePostprocessor";
-import { ChatMessage, LLM, OpenAI } from "./llm/LLM";
 
 /**
  * A ChatEngine is used to handle back and forth chats between the application and the LLM.
@@ -24,16 +21,8 @@ export interface ChatEngine {
    * Send message along with the class's current chat history to the LLM.
    * @param message
    * @param chatHistory optional chat history if you want to customize the chat history
-   * @param streaming optional streaming flag, which auto-sets the return value if True.
    */
-  chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
-    message: string,
-    chatHistory?: ChatMessage[],
-    streaming?: T,
-  ): Promise<R>;
+  chat(message: string, chatHistory?: ChatMessage[]): Promise<Response>;
 
   /**
    * Resets the chat history so that it's empty.
@@ -53,45 +42,13 @@ export class SimpleChatEngine implements ChatEngine {
     this.llm = init?.llm ?? new OpenAI();
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(message: string, chatHistory?: ChatMessage[], streaming?: T): Promise<R> {
-    //Streaming option
-    if (streaming) {
-      return this.streamChat(message, chatHistory) as R;
-    }
-
-    //Non-streaming option
+  async chat(message: string, chatHistory?: ChatMessage[]): Promise<Response> {
     chatHistory = chatHistory ?? this.chatHistory;
     chatHistory.push({ content: message, role: "user" });
-    const response = await this.llm.chat(chatHistory, undefined);
+    const response = await this.llm.chat(chatHistory);
     chatHistory.push(response.message);
     this.chatHistory = chatHistory;
-    return new Response(response.message.content) as R;
-  }
-
-  protected async *streamChat(
-    message: string,
-    chatHistory?: ChatMessage[],
-  ): AsyncGenerator<string, void, unknown> {
-    chatHistory = chatHistory ?? this.chatHistory;
-    chatHistory.push({ content: message, role: "user" });
-    const response_generator = await this.llm.chat(
-      chatHistory,
-      undefined,
-      true,
-    );
-
-    var accumulator: string = "";
-    for await (const part of response_generator) {
-      accumulator += part;
-      yield part;
-    }
-
-    chatHistory.push({ content: accumulator, role: "assistant" });
-    this.chatHistory = chatHistory;
-    return;
+    return new Response(response.message.content);
   }
 
   reset() {
@@ -113,13 +70,13 @@ export class CondenseQuestionChatEngine implements ChatEngine {
   queryEngine: BaseQueryEngine;
   chatHistory: ChatMessage[];
   serviceContext: ServiceContext;
-  condenseMessagePrompt: CondenseQuestionPrompt;
+  condenseMessagePrompt: SimplePrompt;
 
   constructor(init: {
     queryEngine: BaseQueryEngine;
     chatHistory: ChatMessage[];
     serviceContext?: ServiceContext;
-    condenseMessagePrompt?: CondenseQuestionPrompt;
+    condenseMessagePrompt?: SimplePrompt;
   }) {
     this.queryEngine = init.queryEngine;
     this.chatHistory = init?.chatHistory ?? [];
@@ -135,19 +92,15 @@ export class CondenseQuestionChatEngine implements ChatEngine {
     return this.serviceContext.llm.complete(
       defaultCondenseQuestionPrompt({
         question: question,
-        chatHistory: chatHistoryStr,
-      }),
+        chat_history: chatHistoryStr,
+      })
     );
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
+  async chat(
     message: string,
-    chatHistory?: ChatMessage[] | undefined,
-    streaming?: T,
-  ): Promise<R> {
+    chatHistory?: ChatMessage[] | undefined
+  ): Promise<Response> {
     chatHistory = chatHistory ?? this.chatHistory;
 
     const condensedQuestion = (
@@ -159,7 +112,7 @@ export class CondenseQuestionChatEngine implements ChatEngine {
     chatHistory.push({ content: message, role: "user" });
     chatHistory.push({ content: response.response, role: "assistant" });
 
-    return response as R;
+    return response;
   }
 
   reset() {
@@ -167,118 +120,54 @@ export class CondenseQuestionChatEngine implements ChatEngine {
   }
 }
 
-export interface Context {
-  message: ChatMessage;
-  nodes: NodeWithScore[];
-}
-
-export interface ContextGenerator {
-  generate(message: string, parentEvent?: Event): Promise<Context>;
-}
-
-export class DefaultContextGenerator implements ContextGenerator {
-  retriever: BaseRetriever;
-  contextSystemPrompt: ContextSystemPrompt;
-  nodePostprocessors: BaseNodePostprocessor[];
-
-  constructor(init: {
-    retriever: BaseRetriever;
-    contextSystemPrompt?: ContextSystemPrompt;
-    nodePostprocessors?: BaseNodePostprocessor[];
-  }) {
-    this.retriever = init.retriever;
-    this.contextSystemPrompt =
-      init?.contextSystemPrompt ?? defaultContextSystemPrompt;
-    this.nodePostprocessors = init.nodePostprocessors || [];
-  }
-
-  private applyNodePostprocessors(nodes: NodeWithScore[]) {
-    return this.nodePostprocessors.reduce(
-      (nodes, nodePostprocessor) => nodePostprocessor.postprocessNodes(nodes),
-      nodes,
-    );
-  }
-
-  async generate(message: string, parentEvent?: Event): Promise<Context> {
-    if (!parentEvent) {
-      parentEvent = {
-        id: uuidv4(),
-        type: "wrapper",
-        tags: ["final"],
-      };
-    }
-    const sourceNodesWithScore = await this.retriever.retrieve(
-      message,
-      parentEvent,
-    );
-
-    const nodes = this.applyNodePostprocessors(sourceNodesWithScore);
-
-    return {
-      message: {
-        content: this.contextSystemPrompt({
-          context: nodes.map((r) => (r.node as TextNode).text).join("\n\n"),
-        }),
-        role: "system",
-      },
-      nodes,
-    };
-  }
-}
-
 /**
  * ContextChatEngine uses the Index to get the appropriate context for each query.
  * The context is stored in the system prompt, and the chat history is preserved,
  * ideally allowing the appropriate context to be surfaced for each query.
  */
 export class ContextChatEngine implements ChatEngine {
-  chatModel: LLM;
+  retriever: BaseRetriever;
+  chatModel: OpenAI;
   chatHistory: ChatMessage[];
-  contextGenerator: ContextGenerator;
 
   constructor(init: {
     retriever: BaseRetriever;
-    chatModel?: LLM;
+    chatModel?: OpenAI;
     chatHistory?: ChatMessage[];
-    contextSystemPrompt?: ContextSystemPrompt;
-    nodePostprocessors?: BaseNodePostprocessor[];
   }) {
+    this.retriever = init.retriever;
     this.chatModel =
       init.chatModel ?? new OpenAI({ model: "gpt-3.5-turbo-16k" });
     this.chatHistory = init?.chatHistory ?? [];
-    this.contextGenerator = new DefaultContextGenerator({
-      retriever: init.retriever,
-      contextSystemPrompt: init?.contextSystemPrompt,
-    });
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
-    message: string,
-    chatHistory?: ChatMessage[] | undefined,
-    streaming?: T,
-  ): Promise<R> {
+  async chat(message: string, chatHistory?: ChatMessage[] | undefined) {
     chatHistory = chatHistory ?? this.chatHistory;
 
-    //Streaming option
-    if (streaming) {
-      return this.streamChat(message, chatHistory) as R;
-    }
-
     const parentEvent: Event = {
       id: uuidv4(),
       type: "wrapper",
       tags: ["final"],
     };
-    const context = await this.contextGenerator.generate(message, parentEvent);
+    const sourceNodesWithScore = await this.retriever.retrieve(
+      message,
+      parentEvent
+    );
+
+    const systemMessage: ChatMessage = {
+      content: contextSystemPrompt({
+        context: sourceNodesWithScore
+          .map((r) => (r.node as TextNode).text)
+          .join("\n\n"),
+      }),
+      role: "system",
+    };
 
     chatHistory.push({ content: message, role: "user" });
 
     const response = await this.chatModel.chat(
-      [context.message, ...chatHistory],
-      parentEvent,
+      [systemMessage, ...chatHistory],
+      parentEvent
     );
     chatHistory.push(response.message);
 
@@ -286,146 +175,11 @@ export class ContextChatEngine implements ChatEngine {
 
     return new Response(
       response.message.content,
-      context.nodes.map((r) => r.node),
-    ) as R;
-  }
-
-  protected async *streamChat(
-    message: string,
-    chatHistory?: ChatMessage[] | undefined,
-  ): AsyncGenerator<string, void, unknown> {
-    chatHistory = chatHistory ?? this.chatHistory;
-
-    const parentEvent: Event = {
-      id: uuidv4(),
-      type: "wrapper",
-      tags: ["final"],
-    };
-    const context = await this.contextGenerator.generate(message, parentEvent);
-
-    chatHistory.push({ content: message, role: "user" });
-
-    const response_stream = await this.chatModel.chat(
-      [context.message, ...chatHistory],
-      parentEvent,
-      true,
+      sourceNodesWithScore.map((r) => r.node)
     );
-    var accumulator: string = "";
-    for await (const part of response_stream) {
-      accumulator += part;
-      yield part;
-    }
-
-    chatHistory.push({ content: accumulator, role: "assistant" });
-
-    this.chatHistory = chatHistory;
-
-    return;
   }
 
   reset() {
     this.chatHistory = [];
   }
 }
-
-export interface MessageContentDetail {
-  type: "text" | "image_url";
-  text: string;
-  image_url: { url: string };
-}
-
-/**
- * Extended type for the content of a message that allows for multi-modal messages.
- */
-export type MessageContent = string | MessageContentDetail[];
-
-/**
- * HistoryChatEngine is a ChatEngine that uses a `ChatHistory` object
- * to keeps track of chat's message history.
- * A `ChatHistory` object is passed as a parameter for each call to the `chat` method,
- * so the state of the chat engine is preserved between calls.
- * Optionally, a `ContextGenerator` can be used to generate an additional context for each call to `chat`.
- */
-export class HistoryChatEngine {
-  llm: LLM;
-  contextGenerator?: ContextGenerator;
-
-  constructor(init?: Partial<HistoryChatEngine>) {
-    this.llm = init?.llm ?? new OpenAI();
-    this.contextGenerator = init?.contextGenerator;
-  }
-
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
-    message: MessageContent,
-    chatHistory: ChatHistory,
-    streaming?: T,
-  ): Promise<R> {
-    //Streaming option
-    if (streaming) {
-      return this.streamChat(message, chatHistory) as R;
-    }
-    const requestMessages = await this.prepareRequestMessages(
-      message,
-      chatHistory,
-    );
-    const response = await this.llm.chat(requestMessages);
-    chatHistory.addMessage(response.message);
-    return new Response(response.message.content) as R;
-  }
-
-  protected async *streamChat(
-    message: MessageContent,
-    chatHistory: ChatHistory,
-  ): AsyncGenerator<string, void, unknown> {
-    const requestMessages = await this.prepareRequestMessages(
-      message,
-      chatHistory,
-    );
-    const response_stream = await this.llm.chat(
-      requestMessages,
-      undefined,
-      true,
-    );
-
-    var accumulator = "";
-    for await (const part of response_stream) {
-      accumulator += part;
-      yield part;
-    }
-    chatHistory.addMessage({
-      content: accumulator,
-      role: "assistant",
-    });
-    return;
-  }
-
-  private async prepareRequestMessages(
-    message: MessageContent,
-    chatHistory: ChatHistory,
-  ) {
-    chatHistory.addMessage({
-      content: message,
-      role: "user",
-    });
-    let requestMessages;
-    let context;
-    if (this.contextGenerator) {
-      if (Array.isArray(message)) {
-        // message is of type MessageContentDetail[] - retrieve just the text parts and concatenate them
-        // so we can pass them to the context generator
-        message = (message as MessageContentDetail[])
-          .filter((c) => c.type === "text")
-          .map((c) => c.text)
-          .join("\n\n");
-      }
-      context = await this.contextGenerator.generate(message);
-    }
-    requestMessages = await chatHistory.requestMessages(
-      context ? [context.message] : undefined,
-    );
-    return requestMessages;
-  }
-}

packages/core/src/ChatHistory.ts

@@ -1,200 +0,0 @@
-import { ChatMessage, LLM, MessageType, OpenAI } from "./llm/LLM";
-import {
-  defaultSummaryPrompt,
-  messagesToHistoryStr,
-  SummaryPrompt,
-} from "./Prompt";
-
-/**
- * A ChatHistory is used to keep the state of back and forth chat messages
- */
-export interface ChatHistory {
-  messages: ChatMessage[];
-  /**
-   * Adds a message to the chat history.
-   * @param message
-   */
-  addMessage(message: ChatMessage): void;
-
-  /**
-   * Returns the messages that should be used as input to the LLM.
-   */
-  requestMessages(transientMessages?: ChatMessage[]): Promise<ChatMessage[]>;
-
-  /**
-   * Resets the chat history so that it's empty.
-   */
-  reset(): void;
-
-  /**
-   * Returns the new messages since the last call to this function (or since calling the constructor)
-   */
-  newMessages(): ChatMessage[];
-}
-
-export class SimpleChatHistory implements ChatHistory {
-  messages: ChatMessage[];
-  private messagesBefore: number;
-
-  constructor(init?: Partial<SimpleChatHistory>) {
-    this.messages = init?.messages ?? [];
-    this.messagesBefore = this.messages.length;
-  }
-
-  addMessage(message: ChatMessage) {
-    this.messages.push(message);
-  }
-
-  async requestMessages(transientMessages?: ChatMessage[]) {
-    return [...(transientMessages ?? []), ...this.messages];
-  }
-
-  reset() {
-    this.messages = [];
-  }
-
-  newMessages() {
-    const newMessages = this.messages.slice(this.messagesBefore);
-    this.messagesBefore = this.messages.length;
-    return newMessages;
-  }
-}
-
-export class SummaryChatHistory implements ChatHistory {
-  tokensToSummarize: number;
-  messages: ChatMessage[];
-  summaryPrompt: SummaryPrompt;
-  llm: LLM;
-  private messagesBefore: number;
-
-  constructor(init?: Partial<SummaryChatHistory>) {
-    this.messages = init?.messages ?? [];
-    this.messagesBefore = this.messages.length;
-    this.summaryPrompt = init?.summaryPrompt ?? defaultSummaryPrompt;
-    this.llm = init?.llm ?? new OpenAI();
-    if (!this.llm.metadata.maxTokens) {
-      throw new Error(
-        "LLM maxTokens is not set. Needed so the summarizer ensures the context window size of the LLM.",
-      );
-    }
-    this.tokensToSummarize =
-      this.llm.metadata.contextWindow - this.llm.metadata.maxTokens;
-  }
-
-  private async summarize(): Promise<ChatMessage> {
-    // get the conversation messages to create summary
-    const messagesToSummarize = this.calcConversationMessages();
-
-    let promptMessages;
-    do {
-      promptMessages = [
-        {
-          content: this.summaryPrompt({
-            context: messagesToHistoryStr(messagesToSummarize),
-          }),
-          role: "user" as MessageType,
-        },
-      ];
-      // remove oldest message until the chat history is short enough for the context window
-      messagesToSummarize.shift();
-    } while (this.llm.tokens(promptMessages) > this.tokensToSummarize);
-
-    const response = await this.llm.chat(promptMessages);
-    return { content: response.message.content, role: "memory" };
-  }
-
-  addMessage(message: ChatMessage) {
-    this.messages.push(message);
-  }
-
-  // Find last summary message
-  private getLastSummaryIndex(): number | null {
-    const reversedMessages = this.messages.slice().reverse();
-    const index = reversedMessages.findIndex(
-      (message) => message.role === "memory",
-    );
-    if (index === -1) {
-      return null;
-    }
-    return this.messages.length - 1 - index;
-  }
-
-  private get systemMessages() {
-    // get array of all system messages
-    return this.messages.filter((message) => message.role === "system");
-  }
-
-  private get nonSystemMessages() {
-    // get array of all non-system messages
-    return this.messages.filter((message) => message.role !== "system");
-  }
-
-  /**
-   * Calculates the messages that describe the conversation so far.
-   * If there's no memory, all non-system messages are used.
-   * If there's a memory, uses all messages after the last summary message.
-   */
-  private calcConversationMessages(transformSummary?: boolean): ChatMessage[] {
-    const lastSummaryIndex = this.getLastSummaryIndex();
-    if (!lastSummaryIndex) {
-      // there's no memory, so just use all non-system messages
-      return this.nonSystemMessages;
-    } else {
-      // there's a memory, so use all messages after the last summary message
-      // and convert summary message so it can be send to the LLM
-      const summaryMessage: ChatMessage = transformSummary
-        ? {
-            content: `Summary of the conversation so far: ${this.messages[lastSummaryIndex].content}`,
-            role: "system",
-          }
-        : this.messages[lastSummaryIndex];
-      return [summaryMessage, ...this.messages.slice(lastSummaryIndex + 1)];
-    }
-  }
-
-  private calcCurrentRequestMessages(transientMessages?: ChatMessage[]) {
-    // TODO: check order: currently, we're sending:
-    // system messages first, then transient messages and then the messages that describe the conversation so far
-    return [
-      ...this.systemMessages,
-      ...(transientMessages ? transientMessages : []),
-      ...this.calcConversationMessages(true),
-    ];
-  }
-
-  async requestMessages(transientMessages?: ChatMessage[]) {
-    const requestMessages = this.calcCurrentRequestMessages(transientMessages);
-
-    // get tokens of current request messages and the transient messages
-    const tokens = this.llm.tokens(requestMessages);
-    if (tokens > this.tokensToSummarize) {
-      // if there are too many tokens for the next request, call summarize
-      const memoryMessage = await this.summarize();
-      const lastMessage = this.messages.at(-1);
-      if (lastMessage && lastMessage.role === "user") {
-        // if last message is a user message, ensure that it's sent after the new memory message
-        this.messages.pop();
-        this.messages.push(memoryMessage);
-        this.messages.push(lastMessage);
-      } else {
-        // otherwise just add the memory message
-        this.messages.push(memoryMessage);
-      }
-      // TODO: we still might have too many tokens
-      // e.g. too large system messages or transient messages
-      // how should we deal with that?
-      return this.calcCurrentRequestMessages(transientMessages);
-    }
-    return requestMessages;
-  }
-
-  reset() {
-    this.messages = [];
-  }
-
-  newMessages() {
-    const newMessages = this.messages.slice(this.messagesBefore);
-    this.messagesBefore = this.messages.length;
-    return newMessages;
-  }
-}

packages/core/src/Embedding.ts

@@ -8,7 +8,7 @@ import {
   getAzureModel,
   shouldUseAzure,
 } from "./llm/azure";
-import { OpenAISession, getOpenAISession } from "./llm/openai";
+import { getOpenAISession, OpenAISession } from "./llm/openai";
 import { VectorStoreQueryMode } from "./storage/vectorStore/types";
 
 /**
@@ -280,6 +280,9 @@ export class OpenAIEmbedding extends BaseEmbedding {
   }
 
   private async getOpenAIEmbedding(input: string) {
+    input = input.replace(/\n/g, " ");
+    //^ NOTE this performance helper is in the OpenAI python library but may not be in the JS library
+
     const { data } = await this.session.openai.embeddings.create({
       model: this.model,
       input,

packages/core/src/GlobalsHelper.ts

@@ -1,54 +1,28 @@
-import { encodingForModel } from "js-tiktoken";
-
-import { v4 as uuidv4 } from "uuid";
 import { Event, EventTag, EventType } from "./callbacks/CallbackManager";
-
-export enum Tokenizers {
-  CL100K_BASE = "cl100k_base",
-}
+import { v4 as uuidv4 } from "uuid";
 
 /**
  * Helper class singleton
  */
 class GlobalsHelper {
   defaultTokenizer: {
-    encode: (text: string) => Uint32Array;
-    decode: (tokens: Uint32Array) => string;
+    encode: (text: string) => number[];
+    decode: (tokens: number[]) => string;
   } | null = null;
 
-  private initDefaultTokenizer() {
-    const encoding = encodingForModel("text-embedding-ada-002"); // cl100k_base
-
-    this.defaultTokenizer = {
-      encode: (text: string) => {
-        return new Uint32Array(encoding.encode(text));
-      },
-      decode: (tokens: Uint32Array) => {
-        const numberArray = Array.from(tokens);
-        const text = encoding.decode(numberArray);
-        const uint8Array = new TextEncoder().encode(text);
-        return new TextDecoder().decode(uint8Array);
-      },
-    };
-  }
-
-  tokenizer(encoding?: string) {
-    if (encoding && encoding !== Tokenizers.CL100K_BASE) {
-      throw new Error(`Tokenizer encoding ${encoding} not yet supported`);
-    }
+  tokenizer() {
     if (!this.defaultTokenizer) {
-      this.initDefaultTokenizer();
+      const tiktoken = require("tiktoken-node");
+      this.defaultTokenizer = tiktoken.getEncoding("gpt2");
     }
 
     return this.defaultTokenizer!.encode.bind(this.defaultTokenizer);
   }
 
-  tokenizerDecoder(encoding?: string) {
-    if (encoding && encoding !== Tokenizers.CL100K_BASE) {
-      throw new Error(`Tokenizer encoding ${encoding} not yet supported`);
-    }
+  tokenizerDecoder() {
     if (!this.defaultTokenizer) {
-      this.initDefaultTokenizer();
+      const tiktoken = require("tiktoken-node");
+      this.defaultTokenizer = tiktoken.getEncoding("gpt2");
     }
 
     return this.defaultTokenizer!.decode.bind(this.defaultTokenizer);

packages/core/src/Node.ts

@@ -23,23 +23,19 @@ export enum MetadataMode {
   NONE = "NONE",
 }
 
-export type Metadata = Record<string, any>;
-
-export interface RelatedNodeInfo<T extends Metadata = Metadata> {
+export interface RelatedNodeInfo {
   nodeId: string;
   nodeType?: ObjectType;
-  metadata: T;
+  metadata: Record<string, any>;
   hash?: string;
 }
 
-export type RelatedNodeType<T extends Metadata = Metadata> =
-  | RelatedNodeInfo<T>
-  | RelatedNodeInfo<T>[];
+export type RelatedNodeType = RelatedNodeInfo | RelatedNodeInfo[];
 
 /**
  * Generic abstract class for retrievable nodes
  */
-export abstract class BaseNode<T extends Metadata = Metadata> {
+export abstract class BaseNode {
   /**
    * The unique ID of the Node/Document. The trailing underscore is here
    * to avoid collisions with the id keyword in Python.
@@ -50,13 +46,13 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
   embedding?: number[];
 
   // Metadata fields
-  metadata: T = {} as T;
+  metadata: Record<string, any> = {};
   excludedEmbedMetadataKeys: string[] = [];
   excludedLlmMetadataKeys: string[] = [];
-  relationships: Partial<Record<NodeRelationship, RelatedNodeType<T>>> = {};
+  relationships: Partial<Record<NodeRelationship, RelatedNodeType>> = {};
   hash: string = "";
 
-  constructor(init?: Partial<BaseNode<T>>) {
+  constructor(init?: Partial<BaseNode>) {
     Object.assign(this, init);
   }
 
@@ -66,7 +62,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
   abstract getMetadataStr(metadataMode: MetadataMode): string;
   abstract setContent(value: any): void;
 
-  get sourceNode(): RelatedNodeInfo<T> | undefined {
+  get sourceNode(): RelatedNodeInfo | undefined {
     const relationship = this.relationships[NodeRelationship.SOURCE];
 
     if (Array.isArray(relationship)) {
@@ -76,7 +72,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
     return relationship;
   }
 
-  get prevNode(): RelatedNodeInfo<T> | undefined {
+  get prevNode(): RelatedNodeInfo | undefined {
     const relationship = this.relationships[NodeRelationship.PREVIOUS];
 
     if (Array.isArray(relationship)) {
@@ -88,7 +84,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
     return relationship;
   }
 
-  get nextNode(): RelatedNodeInfo<T> | undefined {
+  get nextNode(): RelatedNodeInfo | undefined {
     const relationship = this.relationships[NodeRelationship.NEXT];
 
     if (Array.isArray(relationship)) {
@@ -98,7 +94,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
     return relationship;
   }
 
-  get parentNode(): RelatedNodeInfo<T> | undefined {
+  get parentNode(): RelatedNodeInfo | undefined {
     const relationship = this.relationships[NodeRelationship.PARENT];
 
     if (Array.isArray(relationship)) {
@@ -108,7 +104,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
     return relationship;
   }
 
-  get childNodes(): RelatedNodeInfo<T>[] | undefined {
+  get childNodes(): RelatedNodeInfo[] | undefined {
     const relationship = this.relationships[NodeRelationship.CHILD];
 
     if (!Array.isArray(relationship)) {
@@ -130,7 +126,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
     return this.embedding;
   }
 
-  asRelatedNodeInfo(): RelatedNodeInfo<T> {
+  asRelatedNodeInfo(): RelatedNodeInfo {
     return {
       nodeId: this.id_,
       metadata: this.metadata,
@@ -150,7 +146,7 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
 /**
  * TextNode is the default node type for text. Most common node type in LlamaIndex.TS
  */
-export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
+export class TextNode extends BaseNode {
   text: string = "";
   startCharIdx?: number;
   endCharIdx?: number;
@@ -158,7 +154,7 @@ export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
   // metadataTemplate: NOTE write your own formatter if needed
   metadataSeparator: string = "\n";
 
-  constructor(init?: Partial<TextNode<T>>) {
+  constructor(init?: Partial<TextNode>) {
     super(init);
     Object.assign(this, init);
 
@@ -237,10 +233,10 @@ export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
 //   }
 // }
 
-export class IndexNode<T extends Metadata = Metadata> extends TextNode<T> {
+export class IndexNode extends TextNode {
   indexId: string = "";
 
-  constructor(init?: Partial<IndexNode<T>>) {
+  constructor(init?: Partial<IndexNode>) {
     super(init);
     Object.assign(this, init);
 
@@ -257,8 +253,8 @@ export class IndexNode<T extends Metadata = Metadata> extends TextNode<T> {
 /**
  * A document is just a special text node with a docId.
  */
-export class Document<T extends Metadata = Metadata> extends TextNode<T> {
-  constructor(init?: Partial<Document<T>>) {
+export class Document extends TextNode {
+  constructor(init?: Partial<Document>) {
     super(init);
     Object.assign(this, init);
 
@@ -296,7 +292,7 @@ export function jsonToNode(json: any) {
 /**
  * A node with a similarity score
  */
-export interface NodeWithScore<T extends Metadata = Metadata> {
-  node: BaseNode<T>;
-  score?: number;
+export interface NodeWithScore {
+  node: BaseNode;
+  score: number;
 }

packages/core/src/OutputParser.ts

@@ -28,7 +28,7 @@ class OutputParserError extends Error {
 
   constructor(
     message: string,
-    options: { cause?: Error; output?: string } = {},
+    options: { cause?: Error; output?: string } = {}
   ) {
     // @ts-ignore
     super(message, options); // https://github.com/tc39/proposal-error-cause
@@ -53,31 +53,30 @@ class OutputParserError extends Error {
  * @param text A markdown block with JSON
  * @returns parsed JSON object
  */
-export function parseJsonMarkdown(text: string) {
+function parseJsonMarkdown(text: string) {
   text = text.trim();
 
-  const left_square = text.indexOf("[");
-  const left_brace = text.indexOf("{");
+  const beginDelimiter = "```json";
+  const endDelimiter = "```";
 
-  var left: number;
-  var right: number;
-  if (left_square < left_brace && left_square != -1) {
-    left = left_square;
-    right = text.lastIndexOf("]");
-  } else {
-    left = left_brace;
-    right = text.lastIndexOf("}");
+  const beginIndex = text.indexOf(beginDelimiter);
+  const endIndex = text.indexOf(
+    endDelimiter,
+    beginIndex + beginDelimiter.length
+  );
+  if (beginIndex === -1 || endIndex === -1) {
+    throw new OutputParserError("Not a json markdown", { output: text });
   }
-  const jsonText = text.substring(left, right + 1);
+
+  const jsonText = text.substring(beginIndex + beginDelimiter.length, endIndex);
+
   try {
-    //Single JSON object case
-    if (left_square === -1) {
-      return [JSON.parse(jsonText)];
-    }
-    //Multiple JSON object case.
     return JSON.parse(jsonText);
   } catch (e) {
-    throw new OutputParserError("Not a json markdown", { output: text });
+    throw new OutputParserError("Not a valid json", {
+      cause: e as Error,
+      output: text,
+    });
   }
 }
 

packages/core/src/Prompt.ts

@@ -7,9 +7,7 @@ import { ToolMetadata } from "./Tool";
  * NOTE this is a different interface compared to LlamaIndex Python
  * NOTE 2: we default to empty string to make it easy to calculate prompt sizes
  */
-export type SimplePrompt = (
-  input: Record<string, string | undefined>,
-) => string;
+export type SimplePrompt = (input: Record<string, string>) => string;
 
 /*
 DEFAULT_TEXT_QA_PROMPT_TMPL = (
@@ -24,7 +22,9 @@ DEFAULT_TEXT_QA_PROMPT_TMPL = (
 )
 */
 
-export const defaultTextQaPrompt = ({ context = "", query = "" }) => {
+export const defaultTextQaPrompt: SimplePrompt = (input) => {
+  const { context = "", query = "" } = input;
+
   return `Context information is below.
 ---------------------
 ${context}
@@ -34,8 +34,6 @@ Query: ${query}
 Answer:`;
 };
 
-export type TextQaPrompt = typeof defaultTextQaPrompt;
-
 /*
 DEFAULT_SUMMARY_PROMPT_TMPL = (
     "Write a summary of the following. Try to use only the "
@@ -50,7 +48,9 @@ DEFAULT_SUMMARY_PROMPT_TMPL = (
 )
 */
 
-export const defaultSummaryPrompt = ({ context = "" }) => {
+export const defaultSummaryPrompt: SimplePrompt = (input) => {
+  const { context = "" } = input;
+
   return `Write a summary of the following. Try to use only the information provided. Try to include as many key details as possible.
 
 
@@ -61,8 +61,6 @@ SUMMARY:"""
 `;
 };
 
-export type SummaryPrompt = typeof defaultSummaryPrompt;
-
 /*
 DEFAULT_REFINE_PROMPT_TMPL = (
     "The original query is as follows: {query_str}\n"
@@ -79,11 +77,9 @@ DEFAULT_REFINE_PROMPT_TMPL = (
 )
 */
 
-export const defaultRefinePrompt = ({
-  query = "",
-  existingAnswer = "",
-  context = "",
-}) => {
+export const defaultRefinePrompt: SimplePrompt = (input) => {
+  const { query = "", existingAnswer = "", context = "" } = input;
+
   return `The original query is as follows: ${query}
 We have provided an existing answer: ${existingAnswer}
 We have the opportunity to refine the existing answer (only if needed) with some more context below.
@@ -94,8 +90,6 @@ Given the new context, refine the original answer to better answer the query. If
 Refined Answer:`;
 };
 
-export type RefinePrompt = typeof defaultRefinePrompt;
-
 /*
 DEFAULT_TREE_SUMMARIZE_TMPL = (
   "Context information from multiple sources is below.\n"
@@ -109,7 +103,9 @@ DEFAULT_TREE_SUMMARIZE_TMPL = (
 )
 */
 
-export const defaultTreeSummarizePrompt = ({ context = "", query = "" }) => {
+export const defaultTreeSummarizePrompt: SimplePrompt = (input) => {
+  const { context = "", query = "" } = input;
+
   return `Context information from multiple sources is below.
 ---------------------
 ${context}
@@ -119,9 +115,9 @@ Query: ${query}
 Answer:`;
 };
 
-export type TreeSummarizePrompt = typeof defaultTreeSummarizePrompt;
+export const defaultChoiceSelectPrompt: SimplePrompt = (input) => {
+  const { context = "", query = "" } = input;
 
-export const defaultChoiceSelectPrompt = ({ context = "", query = "" }) => {
   return `A list of documents is shown below. Each document has a number next to it along 
 with a summary of the document. A question is also provided.
 Respond with the numbers of the documents
@@ -153,8 +149,6 @@ Question: ${query}
 Answer:`;
 };
 
-export type ChoiceSelectPrompt = typeof defaultChoiceSelectPrompt;
-
 /*
 PREFIX = """\
 Given a user question, and a list of tools, output a list of relevant sub-questions \
@@ -272,7 +266,9 @@ const exampleOutput: SubQuestion[] = [
   },
 ];
 
-export const defaultSubQuestionPrompt = ({ toolsStr = "", queryStr = "" }) => {
+export const defaultSubQuestionPrompt: SimplePrompt = (input) => {
+  const { toolsStr, queryStr } = input;
+
   return `Given a user question, and a list of tools, output a list of relevant sub-questions that when composed can help answer the full user question:
 
 # Example 1
@@ -302,8 +298,6 @@ ${queryStr}
 `;
 };
 
-export type SubQuestionPrompt = typeof defaultSubQuestionPrompt;
-
 // DEFAULT_TEMPLATE = """\
 // Given a conversation (between Human and Assistant) and a follow up message from Human, \
 // rewrite the message to be a standalone question that captures all relevant context \
@@ -318,10 +312,9 @@ export type SubQuestionPrompt = typeof defaultSubQuestionPrompt;
 // <Standalone question>
 // """
 
-export const defaultCondenseQuestionPrompt = ({
-  chatHistory = "",
-  question = "",
-}) => {
+export const defaultCondenseQuestionPrompt: SimplePrompt = (input) => {
+  const { chatHistory, question } = input;
+
   return `Given a conversation (between Human and Assistant) and a follow up message from Human, rewrite the message to be a standalone question that captures all relevant context from the conversation.
 
 <Chat History>
@@ -334,8 +327,6 @@ ${question}
 `;
 };
 
-export type CondenseQuestionPrompt = typeof defaultCondenseQuestionPrompt;
-
 export function messagesToHistoryStr(messages: ChatMessage[]) {
   return messages.reduce((acc, message) => {
     acc += acc ? "\n" : "";
@@ -348,42 +339,11 @@ export function messagesToHistoryStr(messages: ChatMessage[]) {
   }, "");
 }
 
-export const defaultContextSystemPrompt = ({ context = "" }) => {
+export const contextSystemPrompt: SimplePrompt = (input) => {
+  const { context } = input;
+
   return `Context information is below.
 ---------------------
 ${context}
 ---------------------`;
 };
-
-export type ContextSystemPrompt = typeof defaultContextSystemPrompt;
-
-export const defaultKeywordExtractPrompt = ({
-  context = "",
-  maxKeywords = 10,
-}) => {
-  return `
-Some text is provided below. Given the text, extract up to ${maxKeywords} keywords from the text. Avoid stopwords.
----------------------
-${context}
----------------------
-Provide keywords in the following comma-separated format: 'KEYWORDS: <keywords>'
-`;
-};
-
-export type KeywordExtractPrompt = typeof defaultKeywordExtractPrompt;
-
-export const defaultQueryKeywordExtractPrompt = ({
-  question = "",
-  maxKeywords = 10,
-}) => {
-  return `(
-  "A question is provided below. Given the question, extract up to ${maxKeywords} "
-  "keywords from the text. Focus on extracting the keywords that we can use "
-  "to best lookup answers to the question. Avoid stopwords."
-  "---------------------"
-  "${question}"
-  "---------------------"
-  "Provide keywords in the following comma-separated format: 'KEYWORDS: <keywords>'"
-)`;
-};
-export type QueryKeywordExtractPrompt = typeof defaultQueryKeywordExtractPrompt;

packages/core/src/PromptHelper.ts

@@ -34,7 +34,7 @@ export class PromptHelper {
   numOutput = DEFAULT_NUM_OUTPUTS;
   chunkOverlapRatio = DEFAULT_CHUNK_OVERLAP_RATIO;
   chunkSizeLimit?: number;
-  tokenizer: (text: string) => Uint32Array;
+  tokenizer: (text: string) => number[];
   separator = " ";
 
   constructor(
@@ -42,7 +42,7 @@ export class PromptHelper {
     numOutput = DEFAULT_NUM_OUTPUTS,
     chunkOverlapRatio = DEFAULT_CHUNK_OVERLAP_RATIO,
     chunkSizeLimit?: number,
-    tokenizer?: (text: string) => Uint32Array,
+    tokenizer?: (text: string) => number[],
     separator = " ",
   ) {
     this.contextWindow = contextWindow;

packages/core/src/QueryEngine.ts

@@ -1,4 +1,3 @@
-import { v4 as uuidv4 } from "uuid";
 import { NodeWithScore, TextNode } from "./Node";
 import {
   BaseQuestionGenerator,
@@ -8,10 +7,10 @@ import {
 import { Response } from "./Response";
 import { CompactAndRefine, ResponseSynthesizer } from "./ResponseSynthesizer";
 import { BaseRetriever } from "./Retriever";
+import { v4 as uuidv4 } from "uuid";
+import { Event } from "./callbacks/CallbackManager";
 import { ServiceContext, serviceContextFromDefaults } from "./ServiceContext";
 import { QueryEngineTool, ToolMetadata } from "./Tool";
-import { Event } from "./callbacks/CallbackManager";
-import { BaseNodePostprocessor } from "./indices/BaseNodePostprocessor";
 
 /**
  * A query engine is a question answerer that can use one or more steps.
@@ -31,39 +30,16 @@ export interface BaseQueryEngine {
 export class RetrieverQueryEngine implements BaseQueryEngine {
   retriever: BaseRetriever;
   responseSynthesizer: ResponseSynthesizer;
-  nodePostprocessors: BaseNodePostprocessor[];
-  preFilters?: unknown;
 
   constructor(
     retriever: BaseRetriever,
-    responseSynthesizer?: ResponseSynthesizer,
-    preFilters?: unknown,
-    nodePostprocessors?: BaseNodePostprocessor[],
+    responseSynthesizer?: ResponseSynthesizer
   ) {
     this.retriever = retriever;
     const serviceContext: ServiceContext | undefined =
       this.retriever.getServiceContext();
     this.responseSynthesizer =
       responseSynthesizer || new ResponseSynthesizer({ serviceContext });
-    this.preFilters = preFilters;
-    this.nodePostprocessors = nodePostprocessors || [];
-  }
-
-  private applyNodePostprocessors(nodes: NodeWithScore[]) {
-    return this.nodePostprocessors.reduce(
-      (nodes, nodePostprocessor) => nodePostprocessor.postprocessNodes(nodes),
-      nodes,
-    );
-  }
-
-  private async retrieve(query: string, parentEvent: Event) {
-    const nodes = await this.retriever.retrieve(
-      query,
-      parentEvent,
-      this.preFilters,
-    );
-
-    return this.applyNodePostprocessors(nodes);
   }
 
   async query(query: string, parentEvent?: Event) {
@@ -72,7 +48,7 @@ export class RetrieverQueryEngine implements BaseQueryEngine {
       type: "wrapper",
       tags: ["final"],
     };
-    const nodes = await this.retrieve(query, _parentEvent);
+    const nodes = await this.retriever.retrieve(query, _parentEvent);
     return this.responseSynthesizer.synthesize(query, nodes, _parentEvent);
   }
 }
@@ -146,7 +122,7 @@ export class SubQuestionQueryEngine implements BaseQueryEngine {
     };
 
     const subQNodes = await Promise.all(
-      subQuestions.map((subQ) => this.querySubQ(subQ, subQueryParentEvent)),
+      subQuestions.map((subQ) => this.querySubQ(subQ, subQueryParentEvent))
     );
 
     const nodes = subQNodes
@@ -157,7 +133,7 @@ export class SubQuestionQueryEngine implements BaseQueryEngine {
 
   private async querySubQ(
     subQ: SubQuestion,
-    parentEvent?: Event,
+    parentEvent?: Event
   ): Promise<NodeWithScore | null> {
     try {
       const question = subQ.subQuestion;

packages/core/src/QuestionGenerator.ts

@@ -4,7 +4,7 @@ import {
   SubQuestionOutputParser,
 } from "./OutputParser";
 import {
-  SubQuestionPrompt,
+  SimplePrompt,
   buildToolsText,
   defaultSubQuestionPrompt,
 } from "./Prompt";
@@ -28,7 +28,7 @@ export interface BaseQuestionGenerator {
  */
 export class LLMQuestionGenerator implements BaseQuestionGenerator {
   llm: LLM;
-  prompt: SubQuestionPrompt;
+  prompt: SimplePrompt;
   outputParser: BaseOutputParser<StructuredOutput<SubQuestion[]>>;
 
   constructor(init?: Partial<LLMQuestionGenerator>) {
@@ -45,7 +45,7 @@ export class LLMQuestionGenerator implements BaseQuestionGenerator {
         this.prompt({
           toolsStr,
           queryStr,
-        }),
+        })
       )
     ).message.content;
 

packages/core/src/ResponseSynthesizer.ts

@@ -1,18 +1,15 @@
-import { Event } from "./callbacks/CallbackManager";
-import { LLM } from "./llm/LLM";
 import { MetadataMode, NodeWithScore } from "./Node";
 import {
+  SimplePrompt,
   defaultRefinePrompt,
   defaultTextQaPrompt,
   defaultTreeSummarizePrompt,
-  RefinePrompt,
-  SimplePrompt,
-  TextQaPrompt,
-  TreeSummarizePrompt,
 } from "./Prompt";
 import { getBiggestPrompt } from "./PromptHelper";
 import { Response } from "./Response";
 import { ServiceContext, serviceContextFromDefaults } from "./ServiceContext";
+import { Event } from "./callbacks/CallbackManager";
+import { LLM } from "./llm/LLM";
 
 /**
  * Response modes of the response synthesizer
@@ -76,13 +73,13 @@ export class SimpleResponseBuilder implements BaseResponseBuilder {
  */
 export class Refine implements BaseResponseBuilder {
   serviceContext: ServiceContext;
-  textQATemplate: TextQaPrompt;
-  refineTemplate: RefinePrompt;
+  textQATemplate: SimplePrompt;
+  refineTemplate: SimplePrompt;
 
   constructor(
     serviceContext: ServiceContext,
-    textQATemplate?: TextQaPrompt,
-    refineTemplate?: RefinePrompt,
+    textQATemplate?: SimplePrompt,
+    refineTemplate?: SimplePrompt,
   ) {
     this.serviceContext = serviceContext;
     this.textQATemplate = textQATemplate ?? defaultTextQaPrompt;
@@ -212,14 +209,9 @@ export class CompactAndRefine extends Refine {
  */
 export class TreeSummarize implements BaseResponseBuilder {
   serviceContext: ServiceContext;
-  summaryTemplate: TreeSummarizePrompt;
 
-  constructor(
-    serviceContext: ServiceContext,
-    summaryTemplate?: TreeSummarizePrompt,
-  ) {
+  constructor(serviceContext: ServiceContext) {
     this.serviceContext = serviceContext;
-    this.summaryTemplate = summaryTemplate ?? defaultTreeSummarizePrompt;
   }
 
   async getResponse(
@@ -227,22 +219,22 @@ export class TreeSummarize implements BaseResponseBuilder {
     textChunks: string[],
     parentEvent?: Event,
   ): Promise<string> {
+    const summaryTemplate: SimplePrompt = defaultTreeSummarizePrompt;
+
     if (!textChunks || textChunks.length === 0) {
       throw new Error("Must have at least one text chunk");
     }
 
-    // Should we send the query here too?
     const packedTextChunks = this.serviceContext.promptHelper.repack(
-      this.summaryTemplate,
+      summaryTemplate,
       textChunks,
     );
 
     if (packedTextChunks.length === 1) {
       return (
         await this.serviceContext.llm.complete(
-          this.summaryTemplate({
+          summaryTemplate({
             context: packedTextChunks[0],
-            query,
           }),
           parentEvent,
         )
@@ -251,9 +243,8 @@ export class TreeSummarize implements BaseResponseBuilder {
       const summaries = await Promise.all(
         packedTextChunks.map((chunk) =>
           this.serviceContext.llm.complete(
-            this.summaryTemplate({
+            summaryTemplate({
               context: chunk,
-              query,
             }),
             parentEvent,
           ),
@@ -290,30 +281,22 @@ export function getResponseBuilder(
 export class ResponseSynthesizer {
   responseBuilder: BaseResponseBuilder;
   serviceContext: ServiceContext;
-  metadataMode: MetadataMode;
 
   constructor({
     responseBuilder,
     serviceContext,
-    metadataMode = MetadataMode.NONE,
   }: {
     responseBuilder?: BaseResponseBuilder;
     serviceContext?: ServiceContext;
-    metadataMode?: MetadataMode;
   } = {}) {
     this.serviceContext = serviceContext ?? serviceContextFromDefaults();
     this.responseBuilder =
       responseBuilder ?? getResponseBuilder(this.serviceContext);
-    this.metadataMode = metadataMode;
   }
 
-  async synthesize(
-    query: string,
-    nodesWithScore: NodeWithScore[],
-    parentEvent?: Event,
-  ) {
-    let textChunks: string[] = nodesWithScore.map(({ node }) =>
-      node.getContent(this.metadataMode),
+  async synthesize(query: string, nodes: NodeWithScore[], parentEvent?: Event) {
+    let textChunks: string[] = nodes.map((node) =>
+      node.node.getContent(MetadataMode.NONE),
     );
     const response = await this.responseBuilder.getResponse(
       query,
@@ -322,7 +305,7 @@ export class ResponseSynthesizer {
     );
     return new Response(
       response,
-      nodesWithScore.map(({ node }) => node),
+      nodes.map((node) => node.node),
     );
   }
 }

packages/core/src/Retriever.ts

@@ -1,15 +1,11 @@
-import { Event } from "./callbacks/CallbackManager";
 import { NodeWithScore } from "./Node";
 import { ServiceContext } from "./ServiceContext";
+import { Event } from "./callbacks/CallbackManager";
 
 /**
  * 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(query: string, parentEvent?: Event): Promise<NodeWithScore[]>;
   getServiceContext(): ServiceContext;
 }

packages/core/src/ServiceContext.ts

@@ -1,8 +1,8 @@
 import { BaseEmbedding, OpenAIEmbedding } from "./Embedding";
+import { LLM, OpenAI } from "./llm/LLM";
 import { NodeParser, SimpleNodeParser } from "./NodeParser";
 import { PromptHelper } from "./PromptHelper";
 import { CallbackManager } from "./callbacks/CallbackManager";
-import { LLM, OpenAI } from "./llm/LLM";
 
 /**
  * The ServiceContext is a collection of components that are used in different parts of the application.
@@ -47,7 +47,7 @@ export function serviceContextFromDefaults(options?: ServiceContextOptions) {
 
 export function serviceContextFromServiceContext(
   serviceContext: ServiceContext,
-  options: ServiceContextOptions,
+  options: ServiceContextOptions
 ) {
   const newServiceContext = { ...serviceContext };
   if (options.llm) {

packages/core/src/callbacks/CallbackManager.ts

@@ -20,8 +20,7 @@ interface BaseCallbackResponse {
   event: Event;
 }
 
-//Specify StreamToken per mainstream LLM
-export interface DefaultStreamToken {
+export interface StreamToken {
   id: string;
   object: string;
   created: number;
@@ -30,34 +29,16 @@ export interface DefaultStreamToken {
     index: number;
     delta: {
       content?: string | null;
-      role?: "user" | "assistant" | "system" | "function" | "tool";
+      role?: "user" | "assistant" | "system" | "function";
     };
     finish_reason: string | null;
   }[];
 }
 
-//OpenAI stream token schema is the default.
-//Note: Anthropic and Replicate also use similar token schemas.
-export type OpenAIStreamToken = DefaultStreamToken;
-export type AnthropicStreamToken = {
-  completion: string;
-  model: string;
-  stop_reason: string | undefined;
-  stop?: boolean | undefined;
-  log_id?: string;
-};
-
-//
-//Callback Responses
-//
-//TODO: Write Embedding Callbacks
-
-//StreamCallbackResponse should let practitioners implement callbacks out of the box...
-//When custom streaming LLMs are involved, people are expected to write their own StreamCallbackResponses
 export interface StreamCallbackResponse extends BaseCallbackResponse {
   index: number;
   isDone?: boolean;
-  token?: DefaultStreamToken;
+  token?: StreamToken;
 }
 
 export interface RetrievalCallbackResponse extends BaseCallbackResponse {

packages/core/src/callbacks/utility/handleOpenAIStream.ts

@@ -0,0 +1,45 @@
+import { ChatCompletionChunk } from "openai/resources/chat";
+import { Stream } from "openai/streaming";
+import { globalsHelper } from "../../GlobalsHelper";
+import { MessageType } from "../../llm/LLM";
+import { Event, StreamCallbackResponse } from "../CallbackManager";
+
+/**
+ * Handles the OpenAI streaming interface and pipes it to the callback function
+ * @param response - The response from the OpenAI API.
+ * @param onLLMStream - A callback function to handle the LLM stream.
+ * @param parentEvent - An optional parent event.
+ * @returns A promise that resolves to an object with a message and a role.
+ */
+export async function handleOpenAIStream({
+  response,
+  onLLMStream,
+  parentEvent,
+}: {
+  response: Stream<ChatCompletionChunk>;
+  onLLMStream: (data: StreamCallbackResponse) => void;
+  parentEvent?: Event;
+}): Promise<{ message: string; role: MessageType }> {
+  const event = globalsHelper.createEvent({
+    parentEvent,
+    type: "llmPredict",
+  });
+  let index = 0;
+  let cumulativeText = "";
+  let messageRole: MessageType = "assistant";
+  for await (const part of response) {
+    const { content = "", role = "assistant" } = part.choices[0].delta;
+
+    // ignore the first token
+    if (!content && role === "assistant" && index === 0) {
+      continue;
+    }
+
+    cumulativeText += content;
+    messageRole = role;
+    onLLMStream?.({ event, index, token: part });
+    index++;
+  }
+  onLLMStream?.({ event, index, isDone: true });
+  return { message: cumulativeText, role: messageRole };
+}

packages/core/src/index.ts

@@ -1,29 +1,28 @@
 export * from "./ChatEngine";
-export * from "./ChatHistory";
+export * from "./constants";
 export * from "./Embedding";
 export * from "./GlobalsHelper";
+export * from "./llm/LLM";
 export * from "./Node";
 export * from "./NodeParser";
 export * from "./OutputParser";
 export * from "./Prompt";
-export * from "./PromptHelper";
-export * from "./QueryEngine";
 export * from "./QuestionGenerator";
+export * from "./QueryEngine";
 export * from "./Response";
 export * from "./ResponseSynthesizer";
 export * from "./Retriever";
 export * from "./ServiceContext";
 export * from "./TextSplitter";
 export * from "./Tool";
-export * from "./callbacks/CallbackManager";
-export * from "./constants";
+
 export * from "./indices";
-export * from "./llm/LLM";
-export * from "./readers/CSVReader";
-export * from "./readers/HTMLReader";
-export * from "./readers/MarkdownReader";
-export * from "./readers/NotionReader";
-export * from "./readers/PDFReader";
-export * from "./readers/SimpleDirectoryReader";
+
+export * from "./callbacks/CallbackManager";
+
 export * from "./readers/base";
+export * from "./readers/PDFReader";
+export * from "./readers/CSVReader";
+export * from "./readers/SimpleDirectoryReader";
+
 export * from "./storage";

packages/core/src/indices/BaseIndex.ts

@@ -4,9 +4,9 @@ import { BaseQueryEngine } from "../QueryEngine";
 import { ResponseSynthesizer } from "../ResponseSynthesizer";
 import { BaseRetriever } from "../Retriever";
 import { ServiceContext } from "../ServiceContext";
-import { StorageContext } from "../storage/StorageContext";
 import { BaseDocumentStore } from "../storage/docStore/types";
 import { BaseIndexStore } from "../storage/indexStore/types";
+import { StorageContext } from "../storage/StorageContext";
 import { VectorStore } from "../storage/vectorStore/types";
 
 /**
@@ -39,7 +39,6 @@ export abstract class IndexStruct {
 export enum IndexStructType {
   SIMPLE_DICT = "simple_dict",
   LIST = "list",
-  KEYWORD_TABLE = "keyword_table",
 }
 
 export class IndexDict extends IndexStruct {
@@ -107,36 +106,6 @@ export class IndexList extends IndexStruct {
   }
 }
 
-// A table of keywords mapping keywords to text chunks.
-export class KeywordTable extends IndexStruct {
-  table: Map<string, Set<string>> = new Map();
-  type: IndexStructType = IndexStructType.KEYWORD_TABLE;
-  addNode(keywords: string[], nodeId: string): void {
-    keywords.forEach((keyword) => {
-      if (!this.table.has(keyword)) {
-        this.table.set(keyword, new Set());
-      }
-      this.table.get(keyword)!.add(nodeId);
-    });
-  }
-
-  deleteNode(keywords: string[], nodeId: string) {
-    keywords.forEach((keyword) => {
-      if (this.table.has(keyword)) {
-        this.table.get(keyword)!.delete(nodeId);
-      }
-    });
-  }
-
-  toJson(): Record<string, unknown> {
-    return {
-      ...super.toJson(),
-      table: this.table,
-      type: this.type,
-    };
-  }
-}
-
 export interface BaseIndexInit<T> {
   serviceContext: ServiceContext;
   storageContext: StorageContext;

packages/core/src/indices/BaseNodePostprocessor.ts

@@ -1,20 +0,0 @@
-import { NodeWithScore } from "../Node";
-
-export interface BaseNodePostprocessor {
-  postprocessNodes: (nodes: NodeWithScore[]) => NodeWithScore[];
-}
-
-export class SimilarityPostprocessor implements BaseNodePostprocessor {
-  similarityCutoff?: number;
-
-  constructor(options?: { similarityCutoff?: number }) {
-    this.similarityCutoff = options?.similarityCutoff;
-  }
-
-  postprocessNodes(nodes: NodeWithScore[]) {
-    if (this.similarityCutoff === undefined) return nodes;
-
-    const cutoff = this.similarityCutoff || 0;
-    return nodes.filter((node) => node.score && node.score >= cutoff);
-  }
-}

packages/core/src/indices/index.ts

@@ -1,5 +1,3 @@
 export * from "./BaseIndex";
-export * from "./BaseNodePostprocessor";
-export * from "./keyword";
-export * from "./summary";
+export * from "./list";
 export * from "./vectorStore";

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

@@ -1,274 +0,0 @@
-import { BaseNode, Document, MetadataMode } from "../../Node";
-import { defaultKeywordExtractPrompt } from "../../Prompt";
-import { BaseQueryEngine, RetrieverQueryEngine } from "../../QueryEngine";
-import { ResponseSynthesizer } from "../../ResponseSynthesizer";
-import { BaseRetriever } from "../../Retriever";
-import {
-  ServiceContext,
-  serviceContextFromDefaults,
-} from "../../ServiceContext";
-import { StorageContext, storageContextFromDefaults } from "../../storage";
-import { BaseDocumentStore } from "../../storage/docStore/types";
-import {
-  BaseIndex,
-  BaseIndexInit,
-  IndexStructType,
-  KeywordTable,
-} from "../BaseIndex";
-import { BaseNodePostprocessor } from "../BaseNodePostprocessor";
-import {
-  KeywordTableLLMRetriever,
-  KeywordTableRAKERetriever,
-  KeywordTableSimpleRetriever,
-} from "./KeywordTableIndexRetriever";
-import { extractKeywordsGivenResponse } from "./utils";
-
-export interface KeywordIndexOptions {
-  nodes?: BaseNode[];
-  indexStruct?: KeywordTable;
-  indexId?: string;
-  serviceContext?: ServiceContext;
-  storageContext?: StorageContext;
-}
-export enum KeywordTableRetrieverMode {
-  DEFAULT = "DEFAULT",
-  SIMPLE = "SIMPLE",
-  RAKE = "RAKE",
-}
-
-const KeywordTableRetrieverMap = {
-  [KeywordTableRetrieverMode.DEFAULT]: KeywordTableLLMRetriever,
-  [KeywordTableRetrieverMode.SIMPLE]: KeywordTableSimpleRetriever,
-  [KeywordTableRetrieverMode.RAKE]: KeywordTableRAKERetriever,
-};
-
-/**
- * The KeywordTableIndex, an index that extracts keywords from each Node and builds a mapping from each keyword to the corresponding Nodes of that keyword.
- */
-export class KeywordTableIndex extends BaseIndex<KeywordTable> {
-  constructor(init: BaseIndexInit<KeywordTable>) {
-    super(init);
-  }
-
-  static async init(options: KeywordIndexOptions): Promise<KeywordTableIndex> {
-    const storageContext =
-      options.storageContext ?? (await storageContextFromDefaults({}));
-    const serviceContext =
-      options.serviceContext ?? serviceContextFromDefaults({});
-    const { docStore, indexStore } = storageContext;
-
-    // Setup IndexStruct from storage
-    let indexStructs = (await indexStore.getIndexStructs()) as KeywordTable[];
-    let indexStruct: KeywordTable | null;
-
-    if (options.indexStruct && indexStructs.length > 0) {
-      throw new Error(
-        "Cannot initialize index with both indexStruct and indexStore",
-      );
-    }
-
-    if (options.indexStruct) {
-      indexStruct = options.indexStruct;
-    } else if (indexStructs.length == 1) {
-      indexStruct = indexStructs[0];
-    } else if (indexStructs.length > 1 && options.indexId) {
-      indexStruct = (await indexStore.getIndexStruct(
-        options.indexId,
-      )) as KeywordTable;
-    } else {
-      indexStruct = null;
-    }
-
-    // check indexStruct type
-    if (indexStruct && indexStruct.type !== IndexStructType.KEYWORD_TABLE) {
-      throw new Error(
-        "Attempting to initialize KeywordTableIndex with non-keyword table indexStruct",
-      );
-    }
-
-    if (indexStruct) {
-      if (options.nodes) {
-        throw new Error(
-          "Cannot initialize KeywordTableIndex with both nodes and indexStruct",
-        );
-      }
-    } else {
-      if (!options.nodes) {
-        throw new Error(
-          "Cannot initialize KeywordTableIndex without nodes or indexStruct",
-        );
-      }
-      indexStruct = await KeywordTableIndex.buildIndexFromNodes(
-        options.nodes,
-        storageContext.docStore,
-        serviceContext,
-      );
-
-      await indexStore.addIndexStruct(indexStruct);
-    }
-
-    return new KeywordTableIndex({
-      storageContext,
-      serviceContext,
-      docStore,
-      indexStore,
-      indexStruct,
-    });
-  }
-
-  asRetriever(options?: any): BaseRetriever {
-    const { mode = KeywordTableRetrieverMode.DEFAULT, ...otherOptions } =
-      options ?? {};
-    const KeywordTableRetriever =
-      KeywordTableRetrieverMap[mode as KeywordTableRetrieverMode];
-    if (KeywordTableRetriever) {
-      return new KeywordTableRetriever({ index: this, ...otherOptions });
-    }
-    throw new Error(`Unknown retriever mode: ${mode}`);
-  }
-
-  asQueryEngine(options?: {
-    retriever?: BaseRetriever;
-    responseSynthesizer?: ResponseSynthesizer;
-    preFilters?: unknown;
-    nodePostprocessors?: BaseNodePostprocessor[];
-  }): BaseQueryEngine {
-    const { retriever, responseSynthesizer } = options ?? {};
-    return new RetrieverQueryEngine(
-      retriever ?? this.asRetriever(),
-      responseSynthesizer,
-      options?.preFilters,
-      options?.nodePostprocessors,
-    );
-  }
-
-  static async extractKeywords(
-    text: string,
-    serviceContext: ServiceContext,
-  ): Promise<Set<string>> {
-    const response = await serviceContext.llm.complete(
-      defaultKeywordExtractPrompt({
-        context: text,
-      }),
-    );
-    return extractKeywordsGivenResponse(response.message.content, "KEYWORDS:");
-  }
-
-  /**
-   * High level API: split documents, get keywords, and build index.
-   * @param documents
-   * @param storageContext
-   * @param serviceContext
-   * @returns
-   */
-  static async fromDocuments(
-    documents: Document[],
-    args: {
-      storageContext?: StorageContext;
-      serviceContext?: ServiceContext;
-    } = {},
-  ): Promise<KeywordTableIndex> {
-    let { storageContext, serviceContext } = args;
-    storageContext = storageContext ?? (await storageContextFromDefaults({}));
-    serviceContext = serviceContext ?? serviceContextFromDefaults({});
-    const docStore = storageContext.docStore;
-
-    docStore.addDocuments(documents, true);
-    for (const doc of documents) {
-      docStore.setDocumentHash(doc.id_, doc.hash);
-    }
-
-    const nodes = serviceContext.nodeParser.getNodesFromDocuments(documents);
-    const index = await KeywordTableIndex.init({
-      nodes,
-      storageContext,
-      serviceContext,
-    });
-    return index;
-  }
-
-  /**
-   * Get keywords for nodes and place them into the index.
-   * @param nodes
-   * @param serviceContext
-   * @param vectorStore
-   * @returns
-   */
-  static async buildIndexFromNodes(
-    nodes: BaseNode[],
-    docStore: BaseDocumentStore,
-    serviceContext: ServiceContext,
-  ): Promise<KeywordTable> {
-    const indexStruct = new KeywordTable();
-    await docStore.addDocuments(nodes, true);
-    for (const node of nodes) {
-      const keywords = await KeywordTableIndex.extractKeywords(
-        node.getContent(MetadataMode.LLM),
-        serviceContext,
-      );
-      indexStruct.addNode([...keywords], node.id_);
-    }
-    return indexStruct;
-  }
-
-  async insertNodes(nodes: BaseNode[]) {
-    for (let node of nodes) {
-      const keywords = await KeywordTableIndex.extractKeywords(
-        node.getContent(MetadataMode.LLM),
-        this.serviceContext,
-      );
-      this.indexStruct.addNode([...keywords], node.id_);
-    }
-  }
-
-  deleteNode(nodeId: string): void {
-    const keywordsToDelete: Set<string> = new Set();
-    for (const [keyword, existingNodeIds] of Object.entries(
-      this.indexStruct.table,
-    )) {
-      const index = existingNodeIds.indexOf(nodeId);
-      if (index !== -1) {
-        existingNodeIds.splice(index, 1);
-
-        // Delete keywords that have zero nodes
-        if (existingNodeIds.length === 0) {
-          keywordsToDelete.add(keyword);
-        }
-      }
-    }
-    this.indexStruct.deleteNode([...keywordsToDelete], nodeId);
-  }
-
-  async deleteNodes(nodeIds: string[], deleteFromDocStore: boolean) {
-    nodeIds.forEach((nodeId) => {
-      this.deleteNode(nodeId);
-    });
-
-    if (deleteFromDocStore) {
-      for (const nodeId of nodeIds) {
-        await this.docStore.deleteDocument(nodeId, false);
-      }
-    }
-
-    await this.storageContext.indexStore.addIndexStruct(this.indexStruct);
-  }
-
-  async deleteRefDoc(
-    refDocId: string,
-    deleteFromDocStore?: boolean,
-  ): Promise<void> {
-    const refDocInfo = await this.docStore.getRefDocInfo(refDocId);
-
-    if (!refDocInfo) {
-      return;
-    }
-
-    await this.deleteNodes(refDocInfo.nodeIds, false);
-
-    if (deleteFromDocStore) {
-      await this.docStore.deleteRefDoc(refDocId, false);
-    }
-
-    return;
-  }
-}

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

@@ -1,119 +0,0 @@
-import { NodeWithScore } from "../../Node";
-import {
-  defaultKeywordExtractPrompt,
-  defaultQueryKeywordExtractPrompt,
-  KeywordExtractPrompt,
-  QueryKeywordExtractPrompt,
-} from "../../Prompt";
-import { BaseRetriever } from "../../Retriever";
-import { ServiceContext } from "../../ServiceContext";
-import { BaseDocumentStore } from "../../storage/docStore/types";
-import { KeywordTable } from "../BaseIndex";
-import { KeywordTableIndex } from "./KeywordTableIndex";
-import {
-  extractKeywordsGivenResponse,
-  rakeExtractKeywords,
-  simpleExtractKeywords,
-} from "./utils";
-
-// Base Keyword Table Retriever
-abstract class BaseKeywordTableRetriever implements BaseRetriever {
-  protected index: KeywordTableIndex;
-  protected indexStruct: KeywordTable;
-  protected docstore: BaseDocumentStore;
-  protected serviceContext: ServiceContext;
-
-  protected maxKeywordsPerQuery: number; // Maximum number of keywords to extract from query.
-  protected numChunksPerQuery: number; // Maximum number of text chunks to query.
-  protected keywordExtractTemplate: KeywordExtractPrompt; // A Keyword Extraction Prompt
-  protected queryKeywordExtractTemplate: QueryKeywordExtractPrompt; // A Query Keyword Extraction Prompt
-
-  constructor({
-    index,
-    keywordExtractTemplate,
-    queryKeywordExtractTemplate,
-    maxKeywordsPerQuery = 10,
-    numChunksPerQuery = 10,
-  }: {
-    index: KeywordTableIndex;
-    keywordExtractTemplate?: KeywordExtractPrompt;
-    queryKeywordExtractTemplate?: QueryKeywordExtractPrompt;
-    maxKeywordsPerQuery: number;
-    numChunksPerQuery: number;
-  }) {
-    this.index = index;
-    this.indexStruct = index.indexStruct;
-    this.docstore = index.docStore;
-    this.serviceContext = index.serviceContext;
-
-    this.maxKeywordsPerQuery = maxKeywordsPerQuery;
-    this.numChunksPerQuery = numChunksPerQuery;
-    this.keywordExtractTemplate =
-      keywordExtractTemplate || defaultKeywordExtractPrompt;
-    this.queryKeywordExtractTemplate =
-      queryKeywordExtractTemplate || defaultQueryKeywordExtractPrompt;
-  }
-
-  abstract getKeywords(query: string): Promise<string[]>;
-
-  async retrieve(query: string): Promise<NodeWithScore[]> {
-    const keywords = await this.getKeywords(query);
-    const chunkIndicesCount: { [key: string]: number } = {};
-    const filteredKeywords = keywords.filter((keyword) =>
-      this.indexStruct.table.has(keyword),
-    );
-
-    for (let keyword of filteredKeywords) {
-      for (let nodeId of this.indexStruct.table.get(keyword) || []) {
-        chunkIndicesCount[nodeId] = (chunkIndicesCount[nodeId] ?? 0) + 1;
-      }
-    }
-
-    const sortedChunkIndices = Object.keys(chunkIndicesCount)
-      .sort((a, b) => chunkIndicesCount[b] - chunkIndicesCount[a])
-      .slice(0, this.numChunksPerQuery);
-
-    const sortedNodes = await this.docstore.getNodes(sortedChunkIndices);
-
-    return sortedNodes.map((node) => ({ node }));
-  }
-
-  getServiceContext(): ServiceContext {
-    return this.index.serviceContext;
-  }
-}
-
-// Extracts keywords using LLMs.
-export class KeywordTableLLMRetriever extends BaseKeywordTableRetriever {
-  async getKeywords(query: string): Promise<string[]> {
-    const response = await this.serviceContext.llm.complete(
-      this.queryKeywordExtractTemplate({
-        question: query,
-        maxKeywords: this.maxKeywordsPerQuery,
-      }),
-    );
-    const keywords = extractKeywordsGivenResponse(
-      response.message.content,
-      "KEYWORDS:",
-    );
-    return [...keywords];
-  }
-}
-
-// Extracts keywords using simple regex-based keyword extractor.
-export class KeywordTableSimpleRetriever extends BaseKeywordTableRetriever {
-  getKeywords(query: string): Promise<string[]> {
-    return Promise.resolve([
-      ...simpleExtractKeywords(query, this.maxKeywordsPerQuery),
-    ]);
-  }
-}
-
-// Extracts keywords using RAKE keyword extractor
-export class KeywordTableRAKERetriever extends BaseKeywordTableRetriever {
-  getKeywords(query: string): Promise<string[]> {
-    return Promise.resolve([
-      ...rakeExtractKeywords(query, this.maxKeywordsPerQuery),
-    ]);
-  }
-}

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

@@ -1,9 +0,0 @@
-export {
-  KeywordTableIndex,
-  KeywordTableRetrieverMode,
-} from "./KeywordTableIndex";
-export {
-  KeywordTableLLMRetriever,
-  KeywordTableRAKERetriever,
-  KeywordTableSimpleRetriever,
-} from "./KeywordTableIndexRetriever";

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

@@ -1,81 +0,0 @@
-// @ts-ignore
-import rake from "rake-modified";
-
-// Get subtokens from a list of tokens., filtering for stopwords.
-export function expandTokensWithSubtokens(tokens: Set<string>): Set<string> {
-  const results: Set<string> = new Set();
-  const regex: RegExp = /\w+/g;
-
-  for (let token of tokens) {
-    results.add(token);
-    const subTokens: RegExpMatchArray | null = token.match(regex);
-    if (subTokens && subTokens.length > 1) {
-      for (let w of subTokens) {
-        results.add(w);
-      }
-    }
-  }
-  return results;
-}
-
-export function extractKeywordsGivenResponse(
-  response: string,
-  startToken: string = "",
-  lowercase: boolean = true,
-): Set<string> {
-  const results: string[] = [];
-  response = response.trim();
-
-  if (response.startsWith(startToken)) {
-    response = response.substring(startToken.length);
-  }
-
-  const keywords: string[] = response.split(",");
-  for (let k of keywords) {
-    let rk: string = k;
-    if (lowercase) {
-      rk = rk.toLowerCase();
-    }
-    results.push(rk.trim());
-  }
-
-  return expandTokensWithSubtokens(new Set(results));
-}
-
-export function simpleExtractKeywords(
-  textChunk: string,
-  maxKeywords?: number,
-): Set<string> {
-  const regex: RegExp = /\w+/g;
-  let tokens: string[] = [...textChunk.matchAll(regex)].map((token) =>
-    token[0].toLowerCase().trim(),
-  );
-
-  // Creating a frequency map
-  const valueCounts: { [key: string]: number } = {};
-  for (let token of tokens) {
-    valueCounts[token] = (valueCounts[token] || 0) + 1;
-  }
-
-  // Sorting tokens by frequency
-  const sortedTokens: string[] = Object.keys(valueCounts).sort(
-    (a, b) => valueCounts[b] - valueCounts[a],
-  );
-
-  const keywords: string[] = maxKeywords
-    ? sortedTokens.slice(0, maxKeywords)
-    : sortedTokens;
-
-  return new Set(keywords);
-}
-
-export function rakeExtractKeywords(
-  textChunk: string,
-  maxKeywords?: number,
-): Set<string> {
-  const keywords = Object.keys(rake(textChunk));
-  const limitedKeywords = maxKeywords
-    ? keywords.slice(0, maxKeywords)
-    : keywords;
-  return new Set(limitedKeywords);
-}

packages/core/src/indices/list/ListIndex.ts

@@ -10,30 +10,29 @@ import {
   ServiceContext,
   serviceContextFromDefaults,
 } from "../../ServiceContext";
+import { BaseDocumentStore, RefDocInfo } from "../../storage/docStore/types";
 import {
   StorageContext,
   storageContextFromDefaults,
 } from "../../storage/StorageContext";
-import { BaseDocumentStore, RefDocInfo } from "../../storage/docStore/types";
 import {
   BaseIndex,
   BaseIndexInit,
   IndexList,
   IndexStructType,
 } from "../BaseIndex";
-import { BaseNodePostprocessor } from "../BaseNodePostprocessor";
 import {
-  SummaryIndexLLMRetriever,
-  SummaryIndexRetriever,
-} from "./SummaryIndexRetriever";
+  ListIndexLLMRetriever,
+  ListIndexRetriever,
+} from "./ListIndexRetriever";
 
-export enum SummaryRetrieverMode {
+export enum ListRetrieverMode {
   DEFAULT = "default",
   // EMBEDDING = "embedding",
   LLM = "llm",
 }
 
-export interface SummaryIndexOptions {
+export interface ListIndexOptions {
   nodes?: BaseNode[];
   indexStruct?: IndexList;
   indexId?: string;
@@ -42,14 +41,14 @@ export interface SummaryIndexOptions {
 }
 
 /**
- * A SummaryIndex keeps nodes in a sequential order for use with summarization.
+ * A ListIndex keeps nodes in a sequential list structure
  */
-export class SummaryIndex extends BaseIndex<IndexList> {
+export class ListIndex extends BaseIndex<IndexList> {
   constructor(init: BaseIndexInit<IndexList>) {
     super(init);
   }
 
-  static async init(options: SummaryIndexOptions): Promise<SummaryIndex> {
+  static async init(options: ListIndexOptions): Promise<ListIndex> {
     const storageContext =
       options.storageContext ?? (await storageContextFromDefaults({}));
     const serviceContext =
@@ -81,23 +80,23 @@ export class SummaryIndex extends BaseIndex<IndexList> {
     // check indexStruct type
     if (indexStruct && indexStruct.type !== IndexStructType.LIST) {
       throw new Error(
-        "Attempting to initialize SummaryIndex with non-list indexStruct",
+        "Attempting to initialize ListIndex with non-list indexStruct",
       );
     }
 
     if (indexStruct) {
       if (options.nodes) {
         throw new Error(
-          "Cannot initialize SummaryIndex with both nodes and indexStruct",
+          "Cannot initialize VectorStoreIndex with both nodes and indexStruct",
         );
       }
     } else {
       if (!options.nodes) {
         throw new Error(
-          "Cannot initialize SummaryIndex without nodes or indexStruct",
+          "Cannot initialize VectorStoreIndex without nodes or indexStruct",
         );
       }
-      indexStruct = await SummaryIndex.buildIndexFromNodes(
+      indexStruct = await ListIndex.buildIndexFromNodes(
         options.nodes,
         storageContext.docStore,
       );
@@ -105,7 +104,7 @@ export class SummaryIndex extends BaseIndex<IndexList> {
       await indexStore.addIndexStruct(indexStruct);
     }
 
-    return new SummaryIndex({
+    return new ListIndex({
       storageContext,
       serviceContext,
       docStore,
@@ -120,7 +119,7 @@ export class SummaryIndex extends BaseIndex<IndexList> {
       storageContext?: StorageContext;
       serviceContext?: ServiceContext;
     } = {},
-  ): Promise<SummaryIndex> {
+  ): Promise<ListIndex> {
     let { storageContext, serviceContext } = args;
     storageContext = storageContext ?? (await storageContextFromDefaults({}));
     serviceContext = serviceContext ?? serviceContextFromDefaults({});
@@ -132,7 +131,7 @@ export class SummaryIndex extends BaseIndex<IndexList> {
     }
 
     const nodes = serviceContext.nodeParser.getNodesFromDocuments(documents);
-    const index = await SummaryIndex.init({
+    const index = await ListIndex.init({
       nodes,
       storageContext,
       serviceContext,
@@ -140,14 +139,14 @@ export class SummaryIndex extends BaseIndex<IndexList> {
     return index;
   }
 
-  asRetriever(options?: { mode: SummaryRetrieverMode }): BaseRetriever {
-    const { mode = SummaryRetrieverMode.DEFAULT } = options ?? {};
+  asRetriever(options?: { mode: ListRetrieverMode }): BaseRetriever {
+    const { mode = ListRetrieverMode.DEFAULT } = options ?? {};
 
     switch (mode) {
-      case SummaryRetrieverMode.DEFAULT:
-        return new SummaryIndexRetriever(this);
-      case SummaryRetrieverMode.LLM:
-        return new SummaryIndexLLMRetriever(this);
+      case ListRetrieverMode.DEFAULT:
+        return new ListIndexRetriever(this);
+      case ListRetrieverMode.LLM:
+        return new ListIndexLLMRetriever(this);
       default:
         throw new Error(`Unknown retriever mode: ${mode}`);
     }
@@ -156,8 +155,6 @@ export class SummaryIndex extends BaseIndex<IndexList> {
   asQueryEngine(options?: {
     retriever?: BaseRetriever;
     responseSynthesizer?: ResponseSynthesizer;
-    preFilters?: unknown;
-    nodePostprocessors?: BaseNodePostprocessor[];
   }): BaseQueryEngine {
     let { retriever, responseSynthesizer } = options ?? {};
 
@@ -173,12 +170,7 @@ export class SummaryIndex extends BaseIndex<IndexList> {
       });
     }
 
-    return new RetrieverQueryEngine(
-      retriever,
-      responseSynthesizer,
-      options?.preFilters,
-      options?.nodePostprocessors,
-    );
+    return new RetrieverQueryEngine(retriever, responseSynthesizer);
   }
 
   static async buildIndexFromNodes(
@@ -261,5 +253,4 @@ export class SummaryIndex extends BaseIndex<IndexList> {
 }
 
 // Legacy
-export type ListIndex = SummaryIndex;
-export type ListRetrieverMode = SummaryRetrieverMode;
+export type GPTListIndex = ListIndex;

packages/core/src/indices/list/ListIndexRetriever.ts

@@ -1,25 +1,25 @@
-import _ from "lodash";
-import { globalsHelper } from "../../GlobalsHelper";
-import { NodeWithScore } from "../../Node";
-import { ChoiceSelectPrompt, defaultChoiceSelectPrompt } from "../../Prompt";
 import { BaseRetriever } from "../../Retriever";
+import { NodeWithScore } from "../../Node";
+import { ListIndex } from "./ListIndex";
 import { ServiceContext } from "../../ServiceContext";
-import { Event } from "../../callbacks/CallbackManager";
-import { SummaryIndex } from "./SummaryIndex";
 import {
-  ChoiceSelectParserFunction,
   NodeFormatterFunction,
+  ChoiceSelectParserFunction,
   defaultFormatNodeBatchFn,
   defaultParseChoiceSelectAnswerFn,
 } from "./utils";
+import { SimplePrompt, defaultChoiceSelectPrompt } from "../../Prompt";
+import _ from "lodash";
+import { globalsHelper } from "../../GlobalsHelper";
+import { Event } from "../../callbacks/CallbackManager";
 
 /**
- * Simple retriever for SummaryIndex that returns all nodes
+ * Simple retriever for ListIndex that returns all nodes
  */
-export class SummaryIndexRetriever implements BaseRetriever {
-  index: SummaryIndex;
+export class ListIndexRetriever implements BaseRetriever {
+  index: ListIndex;
 
-  constructor(index: SummaryIndex) {
+  constructor(index: ListIndex) {
     this.index = index;
   }
 
@@ -51,23 +51,23 @@ export class SummaryIndexRetriever implements BaseRetriever {
 }
 
 /**
- * LLM retriever for SummaryIndex which lets you select the most relevant chunks.
+ * LLM retriever for ListIndex.
  */
-export class SummaryIndexLLMRetriever implements BaseRetriever {
-  index: SummaryIndex;
-  choiceSelectPrompt: ChoiceSelectPrompt;
+export class ListIndexLLMRetriever implements BaseRetriever {
+  index: ListIndex;
+  choiceSelectPrompt: SimplePrompt;
   choiceBatchSize: number;
   formatNodeBatchFn: NodeFormatterFunction;
   parseChoiceSelectAnswerFn: ChoiceSelectParserFunction;
   serviceContext: ServiceContext;
 
   constructor(
-    index: SummaryIndex,
-    choiceSelectPrompt?: ChoiceSelectPrompt,
+    index: ListIndex,
+    choiceSelectPrompt?: SimplePrompt,
     choiceBatchSize: number = 10,
     formatNodeBatchFn?: NodeFormatterFunction,
     parseChoiceSelectAnswerFn?: ChoiceSelectParserFunction,
-    serviceContext?: ServiceContext,
+    serviceContext?: ServiceContext
   ) {
     this.index = index;
     this.choiceSelectPrompt = choiceSelectPrompt || defaultChoiceSelectPrompt;
@@ -95,7 +95,7 @@ export class SummaryIndexLLMRetriever implements BaseRetriever {
       // parseResult is a map from doc number to relevance score
       const parseResult = this.parseChoiceSelectAnswerFn(
         rawResponse,
-        nodesBatch.length,
+        nodesBatch.length
       );
       const choiceNodeIds = nodeIdsBatch.filter((nodeId, idx) => {
         return `${idx}` in parseResult;
@@ -128,7 +128,3 @@ export class SummaryIndexLLMRetriever implements BaseRetriever {
     return this.serviceContext;
   }
 }
-
-// Legacy
-export type ListIndexRetriever = SummaryIndexRetriever;
-export type ListIndexLLMRetriever = SummaryIndexLLMRetriever;

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

@@ -0,0 +1,5 @@
+export { ListIndex, ListRetrieverMode } from "./ListIndex";
+export {
+  ListIndexRetriever,
+  ListIndexLLMRetriever,
+} from "./ListIndexRetriever";

packages/core/src/indices/list/utils.ts

@@ -1,9 +1,9 @@
-import _ from "lodash";
 import { BaseNode, MetadataMode } from "../../Node";
+import _ from "lodash";
 
 export type NodeFormatterFunction = (summaryNodes: BaseNode[]) => string;
 export const defaultFormatNodeBatchFn: NodeFormatterFunction = (
-  summaryNodes: BaseNode[],
+  summaryNodes: BaseNode[]
 ): string => {
   return summaryNodes
     .map((node, idx) => {
@@ -20,13 +20,13 @@ export type ChoiceSelectParseResult = { [docNumber: number]: number };
 export type ChoiceSelectParserFunction = (
   answer: string,
   numChoices: number,
-  raiseErr?: boolean,
+  raiseErr?: boolean
 ) => ChoiceSelectParseResult;
 
 export const defaultParseChoiceSelectAnswerFn: ChoiceSelectParserFunction = (
   answer: string,
   numChoices: number,
-  raiseErr: boolean = false,
+  raiseErr: boolean = false
 ): ChoiceSelectParseResult => {
   // split the line into the answer number and relevance score portions
   const lineTokens: string[][] = answer
@@ -36,7 +36,7 @@ export const defaultParseChoiceSelectAnswerFn: ChoiceSelectParserFunction = (
       if (lineTokens.length !== 2) {
         if (raiseErr) {
           throw new Error(
-            `Invalid answer line: ${line}. Answer line must be of the form: answer_num: <int>, answer_relevance: <float>`,
+            `Invalid answer line: ${line}. Answer line must be of the form: answer_num: <int>, answer_relevance: <float>`
           );
         } else {
           return null;
@@ -55,7 +55,7 @@ export const defaultParseChoiceSelectAnswerFn: ChoiceSelectParserFunction = (
         if (docNum < 1 || docNum > numChoices) {
           if (raiseErr) {
             throw new Error(
-              `Invalid answer number: ${docNum}. Answer number must be between 1 and ${numChoices}`,
+              `Invalid answer number: ${docNum}. Answer number must be between 1 and ${numChoices}`
             );
           }
         } else {
@@ -68,6 +68,6 @@ export const defaultParseChoiceSelectAnswerFn: ChoiceSelectParserFunction = (
       }
       return parseResult;
     },
-    {},
+    {}
   );
 };

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

@@ -1,10 +0,0 @@
-export { SummaryIndex, SummaryRetrieverMode } from "./SummaryIndex";
-export type { ListIndex, ListRetrieverMode } from "./SummaryIndex";
-export {
-  SummaryIndexLLMRetriever,
-  SummaryIndexRetriever,
-} from "./SummaryIndexRetriever";
-export type {
-  ListIndexLLMRetriever,
-  ListIndexRetriever,
-} from "./SummaryIndexRetriever";

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

@@ -1,14 +1,14 @@
-import { Event } from "../../callbacks/CallbackManager";
-import { DEFAULT_SIMILARITY_TOP_K } from "../../constants";
+import { VectorStoreIndex } from "./VectorStoreIndex";
 import { globalsHelper } from "../../GlobalsHelper";
 import { NodeWithScore } from "../../Node";
-import { BaseRetriever } from "../../Retriever";
 import { ServiceContext } from "../../ServiceContext";
+import { Event } from "../../callbacks/CallbackManager";
+import { DEFAULT_SIMILARITY_TOP_K } from "../../constants";
 import {
   VectorStoreQuery,
   VectorStoreQueryMode,
 } from "../../storage/vectorStore/types";
-import { VectorStoreIndex } from "./VectorStoreIndex";
+import { BaseRetriever } from "../../Retriever";
 
 /**
  * VectorIndexRetriever retrieves nodes from a VectorIndex.
@@ -32,11 +32,7 @@ export class VectorIndexRetriever implements BaseRetriever {
     this.similarityTopK = similarityTopK ?? DEFAULT_SIMILARITY_TOP_K;
   }
 
-  async retrieve(
-    query: string,
-    parentEvent?: Event,
-    preFilters?: unknown,
-  ): Promise<NodeWithScore[]> {
+  async retrieve(query: string, parentEvent?: Event): Promise<NodeWithScore[]> {
     const queryEmbedding =
       await this.serviceContext.embedModel.getQueryEmbedding(query);
 
@@ -45,15 +41,10 @@ export class VectorIndexRetriever implements BaseRetriever {
       mode: VectorStoreQueryMode.DEFAULT,
       similarityTopK: this.similarityTopK,
     };
-    const result = await this.index.vectorStore.query(q, preFilters);
+    const result = await this.index.vectorStore.query(q);
 
     let nodesWithScores: NodeWithScore[] = [];
     for (let i = 0; i < result.ids.length; i++) {
-      const nodeFromResult = result.nodes?.[i];
-      if (!this.index.indexStruct.nodesDict[result.ids[i]] && nodeFromResult) {
-        this.index.indexStruct.nodesDict[result.ids[i]] = nodeFromResult;
-      }
-
       const node = this.index.indexStruct.nodesDict[result.ids[i]];
       nodesWithScores.push({
         node: node,

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

@@ -18,7 +18,6 @@ import {
   IndexDict,
   IndexStructType,
 } from "../BaseIndex";
-import { BaseNodePostprocessor } from "../BaseNodePostprocessor";
 import { VectorIndexRetriever } from "./VectorIndexRetriever";
 
 export interface VectorIndexOptions {
@@ -88,23 +87,24 @@ export class VectorStoreIndex extends BaseIndex<IndexDict> {
       );
     }
 
-    if (options.nodes) {
-      // If nodes are passed in, then we need to update the index
-      indexStruct = await VectorStoreIndex.buildIndexFromNodes(
-        options.nodes,
-        serviceContext,
-        vectorStore,
-        docStore,
-        indexStruct,
-      );
-
-      await indexStore.addIndexStruct(indexStruct);
-    } else if (!indexStruct) {
+    if (!indexStruct && !options.nodes) {
       throw new Error(
         "Cannot initialize VectorStoreIndex without nodes or indexStruct",
       );
     }
 
+    const nodes = options.nodes ?? [];
+
+    indexStruct = await VectorStoreIndex.buildIndexFromNodes(
+      nodes,
+      serviceContext,
+      vectorStore,
+      docStore,
+      indexStruct,
+    );
+
+    await indexStore.addIndexStruct(indexStruct);
+
     return new VectorStoreIndex({
       storageContext,
       serviceContext,
@@ -219,27 +219,6 @@ export class VectorStoreIndex extends BaseIndex<IndexDict> {
     return index;
   }
 
-  static async fromVectorStore(
-    vectorStore: VectorStore,
-    serviceContext: ServiceContext,
-  ) {
-    if (!vectorStore.storesText) {
-      throw new Error(
-        "Cannot initialize from a vector store that does not store text",
-      );
-    }
-
-    const storageContext = await storageContextFromDefaults({ vectorStore });
-
-    const index = await VectorStoreIndex.init({
-      nodes: [],
-      storageContext,
-      serviceContext,
-    });
-
-    return index;
-  }
-
   asRetriever(options?: any): VectorIndexRetriever {
     return new VectorIndexRetriever({ index: this, ...options });
   }
@@ -247,15 +226,11 @@ export class VectorStoreIndex extends BaseIndex<IndexDict> {
   asQueryEngine(options?: {
     retriever?: BaseRetriever;
     responseSynthesizer?: ResponseSynthesizer;
-    preFilters?: unknown;
-    nodePostprocessors?: BaseNodePostprocessor[];
   }): BaseQueryEngine {
     const { retriever, responseSynthesizer } = options ?? {};
     return new RetrieverQueryEngine(
       retriever ?? this.asRetriever(),
       responseSynthesizer,
-      options?.preFilters,
-      options?.nodePostprocessors,
     );
   }
 

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

@@ -1,2 +1,2 @@
-export { VectorIndexRetriever } from "./VectorIndexRetriever";
 export { VectorStoreIndex } from "./VectorStoreIndex";
+export { VectorIndexRetriever } from "./VectorIndexRetriever";

packages/core/src/llm/LLM.ts

@@ -1,16 +1,6 @@
 import OpenAILLM, { ClientOptions as OpenAIClientOptions } from "openai";
-import {
-  AnthropicStreamToken,
-  CallbackManager,
-  Event,
-  EventType,
-  OpenAIStreamToken,
-  StreamCallbackResponse,
-} from "../callbacks/CallbackManager";
-
-import { ChatCompletionMessageParam } from "openai/resources";
-import { LLMOptions } from "portkey-ai";
-import { globalsHelper, Tokenizers } from "../GlobalsHelper";
+import { CallbackManager, Event } from "../callbacks/CallbackManager";
+import { handleOpenAIStream } from "../callbacks/utility/handleOpenAIStream";
 import {
   ANTHROPIC_AI_PROMPT,
   ANTHROPIC_HUMAN_PROMPT,
@@ -24,8 +14,7 @@ import {
   getAzureModel,
   shouldUseAzure,
 } from "./azure";
-import { getOpenAISession, OpenAISession } from "./openai";
-import { getPortkeySession, PortkeySession } from "./portkey";
+import { OpenAISession, getOpenAISession } from "./openai";
 import { ReplicateSession } from "./replicate";
 
 export type MessageType =
@@ -33,11 +22,10 @@ export type MessageType =
   | "assistant"
   | "system"
   | "generic"
-  | "function"
-  | "memory";
+  | "function";
 
 export interface ChatMessage {
-  content: any;
+  content: string;
   role: MessageType;
 }
 
@@ -50,67 +38,31 @@ export interface ChatResponse {
 // NOTE in case we need CompletionResponse to diverge from ChatResponse in the future
 export type CompletionResponse = ChatResponse;
 
-export interface LLMMetadata {
-  model: string;
-  temperature: number;
-  topP: number;
-  maxTokens?: number;
-  contextWindow: number;
-  tokenizer: Tokenizers | undefined;
-}
-
 /**
  * Unified language model interface
  */
 export interface LLM {
-  metadata: LLMMetadata;
-  // Whether a LLM has streaming support
-  hasStreaming: boolean;
   /**
    * Get a chat response from the LLM
    * @param messages
-   *
-   * The return type of chat() and complete() are set by the "streaming" parameter being set to True.
    */
-  chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
-    messages: ChatMessage[],
-    parentEvent?: Event,
-    streaming?: T,
-  ): Promise<R>;
+  chat(messages: ChatMessage[], parentEvent?: Event): Promise<ChatResponse>;
 
   /**
    * Get a prompt completion from the LLM
    * @param prompt the prompt to complete
    */
-  complete<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
-    prompt: string,
-    parentEvent?: Event,
-    streaming?: T,
-  ): Promise<R>;
-
-  /**
-   * Calculates the number of tokens needed for the given chat messages
-   */
-  tokens(messages: ChatMessage[]): number;
+  complete(prompt: string, parentEvent?: Event): Promise<CompletionResponse>;
 }
 
 export const GPT4_MODELS = {
   "gpt-4": { contextWindow: 8192 },
   "gpt-4-32k": { contextWindow: 32768 },
-  "gpt-4-1106-preview": { contextWindow: 128000 },
-  "gpt-4-vision-preview": { contextWindow: 8192 },
 };
 
-export const GPT35_MODELS = {
+export const TURBO_MODELS = {
   "gpt-3.5-turbo": { contextWindow: 4096 },
   "gpt-3.5-turbo-16k": { contextWindow: 16384 },
-  "gpt-3.5-turbo-1106": { contextWindow: 16384 },
 };
 
 /**
@@ -118,22 +70,20 @@ export const GPT35_MODELS = {
  */
 export const ALL_AVAILABLE_OPENAI_MODELS = {
   ...GPT4_MODELS,
-  ...GPT35_MODELS,
+  ...TURBO_MODELS,
 };
 
 /**
  * OpenAI LLM implementation
  */
 export class OpenAI implements LLM {
-  hasStreaming: boolean = true;
-
   // Per completion OpenAI params
   model: keyof typeof ALL_AVAILABLE_OPENAI_MODELS;
   temperature: number;
   topP: number;
   maxTokens?: number;
   additionalChatOptions?: Omit<
-    Partial<OpenAILLM.Chat.ChatCompletionCreateParams>,
+    Partial<OpenAILLM.Chat.CompletionCreateParams>,
     "max_tokens" | "messages" | "model" | "temperature" | "top_p" | "streaming"
   >;
 
@@ -203,32 +153,6 @@ export class OpenAI implements LLM {
     this.callbackManager = init?.callbackManager;
   }
 
-  get metadata() {
-    return {
-      model: this.model,
-      temperature: this.temperature,
-      topP: this.topP,
-      maxTokens: this.maxTokens,
-      contextWindow: ALL_AVAILABLE_OPENAI_MODELS[this.model].contextWindow,
-      tokenizer: Tokenizers.CL100K_BASE,
-    };
-  }
-
-  tokens(messages: ChatMessage[]): number {
-    // for latest OpenAI models, see https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
-    const tokenizer = globalsHelper.tokenizer(this.metadata.tokenizer);
-    const tokensPerMessage = 3;
-    let numTokens = 0;
-    for (const message of messages) {
-      numTokens += tokensPerMessage;
-      for (const value of Object.values(message)) {
-        numTokens += tokenizer(value).length;
-      }
-    }
-    numTokens += 3; // every reply is primed with <|im_start|>assistant<|im_sep|>
-    return numTokens;
-  }
-
   mapMessageType(
     messageType: MessageType,
   ): "user" | "assistant" | "system" | "function" {
@@ -246,124 +170,52 @@ export class OpenAI implements LLM {
     }
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(messages: ChatMessage[], parentEvent?: Event, streaming?: T): Promise<R> {
-    const baseRequestParams: OpenAILLM.Chat.ChatCompletionCreateParams = {
-      model: this.model,
-      temperature: this.temperature,
-      max_tokens: this.maxTokens,
-      messages: messages.map(
-        (message) =>
-          ({
-            role: this.mapMessageType(message.role),
-            content: message.content,
-          }) as ChatCompletionMessageParam,
-      ),
-      top_p: this.topP,
-      ...this.additionalChatOptions,
-    };
-    // Streaming
-    if (streaming) {
-      if (!this.hasStreaming) {
-        throw Error("No streaming support for this LLM.");
-      }
-      return this.streamChat(messages, parentEvent) as R;
-    }
-    // Non-streaming
-    const response = await this.session.openai.chat.completions.create({
-      ...baseRequestParams,
-      stream: false,
-    });
-
-    const content = response.choices[0].message?.content ?? "";
-    return {
-      message: { content, role: response.choices[0].message.role },
-    } as R;
-  }
-
-  async complete<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(prompt: string, parentEvent?: Event, streaming?: T): Promise<R> {
-    return this.chat(
-      [{ content: prompt, role: "user" }],
-      parentEvent,
-      streaming,
-    );
-  }
-
-  //We can wrap a stream in a generator to add some additional logging behavior
-  //For future edits: syntax for generator type is <typeof Yield, typeof Return, typeof Accept>
-  //"typeof Accept" refers to what types you'll accept when you manually call generator.next(<AcceptType>)
-  protected async *streamChat(
+  async chat(
     messages: ChatMessage[],
     parentEvent?: Event,
-  ): AsyncGenerator<string, void, unknown> {
-    const baseRequestParams: OpenAILLM.Chat.ChatCompletionCreateParams = {
+  ): Promise<ChatResponse> {
+    const baseRequestParams: OpenAILLM.Chat.CompletionCreateParams = {
       model: this.model,
       temperature: this.temperature,
       max_tokens: this.maxTokens,
-      messages: messages.map(
-        (message) =>
-          ({
-            role: this.mapMessageType(message.role),
-            content: message.content,
-          }) as ChatCompletionMessageParam,
-      ),
+      messages: messages.map((message) => ({
+        role: this.mapMessageType(message.role),
+        content: message.content,
+      })),
       top_p: this.topP,
       ...this.additionalChatOptions,
     };
 
-    //Now let's wrap our stream in a callback
-    const onLLMStream = this.callbackManager?.onLLMStream
-      ? this.callbackManager.onLLMStream
-      : () => {};
-
-    const chunk_stream: AsyncIterable<OpenAIStreamToken> =
-      await this.session.openai.chat.completions.create({
+    if (this.callbackManager?.onLLMStream) {
+      // Streaming
+      const response = await this.session.openai.chat.completions.create({
         ...baseRequestParams,
         stream: true,
       });
 
-    const event: Event = parentEvent
-      ? parentEvent
-      : {
-          id: "unspecified",
-          type: "llmPredict" as EventType,
-        };
+      const { message, role } = await handleOpenAIStream({
+        response,
+        onLLMStream: this.callbackManager.onLLMStream,
+        parentEvent,
+      });
+      return { message: { content: message, role } };
+    } else {
+      // Non-streaming
+      const response = await this.session.openai.chat.completions.create({
+        ...baseRequestParams,
+        stream: false,
+      });
 
-    //Indices
-    var idx_counter: number = 0;
-    for await (const part of chunk_stream) {
-      //Increment
-      part.choices[0].index = idx_counter;
-      const is_done: boolean =
-        part.choices[0].finish_reason === "stop" ? true : false;
-      //onLLMStream Callback
-
-      const stream_callback: StreamCallbackResponse = {
-        event: event,
-        index: idx_counter,
-        isDone: is_done,
-        token: part,
-      };
-      onLLMStream(stream_callback);
-
-      idx_counter++;
-
-      yield part.choices[0].delta.content ? part.choices[0].delta.content : "";
+      const content = response.choices[0].message?.content ?? "";
+      return { message: { content, role: response.choices[0].message.role } };
     }
-    return;
   }
 
-  //streamComplete doesn't need to be async because it's child function is already async
-  protected streamComplete(
-    query: string,
+  async complete(
+    prompt: string,
     parentEvent?: Event,
-  ): AsyncGenerator<string, void, unknown> {
-    return this.streamChat([{ content: query, role: "user" }], parentEvent);
+  ): Promise<CompletionResponse> {
+    return this.chat([{ content: prompt, role: "user" }], parentEvent);
   }
 }
 
@@ -377,10 +229,10 @@ export const ALL_AVAILABLE_LLAMADEUCE_MODELS = {
   "Llama-2-70b-chat-4bit": {
     contextWindow: 4096,
     replicateApi:
-      "meta/llama-2-70b-chat:02e509c789964a7ea8736978a43525956ef40397be9033abf9fd2badfe68c9e3",
+      "replicate/llama70b-v2-chat:2c1608e18606fad2812020dc541930f2d0495ce32eee50074220b87300bc16e1",
     //^ Model is based off of exllama 4bit.
   },
-  "Llama-2-13b-chat-old": {
+  "Llama-2-13b-chat": {
     contextWindow: 4096,
     replicateApi:
       "a16z-infra/llama13b-v2-chat:df7690f1994d94e96ad9d568eac121aecf50684a0b0963b25a41cc40061269e5",
@@ -389,9 +241,9 @@ export const ALL_AVAILABLE_LLAMADEUCE_MODELS = {
   "Llama-2-13b-chat-4bit": {
     contextWindow: 4096,
     replicateApi:
-      "meta/llama-2-13b-chat:f4e2de70d66816a838a89eeeb621910adffb0dd0baba3976c96980970978018d",
+      "a16z-infra/llama13b-v2-chat:2a7f981751ec7fdf87b5b91ad4db53683a98082e9ff7bfd12c8cd5ea85980a52",
   },
-  "Llama-2-7b-chat-old": {
+  "Llama-2-7b-chat": {
     contextWindow: 4096,
     replicateApi:
       "a16z-infra/llama7b-v2-chat:4f0a4744c7295c024a1de15e1a63c880d3da035fa1f49bfd344fe076074c8eea",
@@ -403,7 +255,7 @@ export const ALL_AVAILABLE_LLAMADEUCE_MODELS = {
   "Llama-2-7b-chat-4bit": {
     contextWindow: 4096,
     replicateApi:
-      "meta/llama-2-7b-chat:13c3cdee13ee059ab779f0291d29054dab00a47dad8261375654de5540165fb0",
+      "a16z-infra/llama7b-v2-chat:4f0b260b6a13eb53a6b1891f089d57c08f41003ae79458be5011303d81a394dc",
   },
 };
 
@@ -415,8 +267,6 @@ export enum DeuceChatStrategy {
   // Unfortunately any string only API won't support these properly.
   REPLICATE4BIT = "replicate4bit",
   //^ To satisfy Replicate's 4 bit models' requirements where they also insert some INST tags
-  REPLICATE4BITWNEWLINES = "replicate4bitwnewlines",
-  //^ Replicate's documentation recommends using newlines: https://replicate.com/blog/how-to-prompt-llama
 }
 
 /**
@@ -429,14 +279,13 @@ export class LlamaDeuce implements LLM {
   topP: number;
   maxTokens?: number;
   replicateSession: ReplicateSession;
-  hasStreaming: boolean;
 
   constructor(init?: Partial<LlamaDeuce>) {
     this.model = init?.model ?? "Llama-2-70b-chat-4bit";
     this.chatStrategy =
       init?.chatStrategy ??
       (this.model.endsWith("4bit")
-        ? DeuceChatStrategy.REPLICATE4BITWNEWLINES // With the newer Replicate models they do the system message themselves.
+        ? DeuceChatStrategy.REPLICATE4BIT // With the newer A16Z/Replicate models they do the system message themselves.
         : DeuceChatStrategy.METAWBOS); // With BOS and EOS seems to work best, although they all have problems past a certain point
     this.temperature = init?.temperature ?? 0.1; // minimum temperature is 0.01 for Replicate endpoint
     this.topP = init?.topP ?? 1;
@@ -444,22 +293,6 @@ export class LlamaDeuce implements LLM {
       init?.maxTokens ??
       ALL_AVAILABLE_LLAMADEUCE_MODELS[this.model].contextWindow; // For Replicate, the default is 500 tokens which is too low.
     this.replicateSession = init?.replicateSession ?? new ReplicateSession();
-    this.hasStreaming = init?.hasStreaming ?? false;
-  }
-
-  tokens(messages: ChatMessage[]): number {
-    throw new Error("Method not implemented.");
-  }
-
-  get metadata() {
-    return {
-      model: this.model,
-      temperature: this.temperature,
-      topP: this.topP,
-      maxTokens: this.maxTokens,
-      contextWindow: ALL_AVAILABLE_LLAMADEUCE_MODELS[this.model].contextWindow,
-      tokenizer: undefined,
-    };
   }
 
   mapMessagesToPrompt(messages: ChatMessage[]) {
@@ -470,15 +303,7 @@ export class LlamaDeuce implements LLM {
     } else if (this.chatStrategy === DeuceChatStrategy.METAWBOS) {
       return this.mapMessagesToPromptMeta(messages, { withBos: true });
     } else if (this.chatStrategy === DeuceChatStrategy.REPLICATE4BIT) {
-      return this.mapMessagesToPromptMeta(messages, {
-        replicate4Bit: true,
-        withNewlines: true,
-      });
-    } else if (this.chatStrategy === DeuceChatStrategy.REPLICATE4BITWNEWLINES) {
-      return this.mapMessagesToPromptMeta(messages, {
-        replicate4Bit: true,
-        withNewlines: true,
-      });
+      return this.mapMessagesToPromptMeta(messages, { replicate4Bit: true });
     } else {
       return this.mapMessagesToPromptMeta(messages);
     }
@@ -513,17 +338,9 @@ export class LlamaDeuce implements LLM {
 
   mapMessagesToPromptMeta(
     messages: ChatMessage[],
-    opts?: {
-      withBos?: boolean;
-      replicate4Bit?: boolean;
-      withNewlines?: boolean;
-    },
+    opts?: { withBos?: boolean; replicate4Bit?: boolean },
   ) {
-    const {
-      withBos = false,
-      replicate4Bit = false,
-      withNewlines = false,
-    } = opts ?? {};
+    const { withBos = false, replicate4Bit = false } = opts ?? {};
     const DEFAULT_SYSTEM_PROMPT = `You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
 
 If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.`;
@@ -571,28 +388,21 @@ If a question does not make any sense, or is not factually coherent, explain why
     return {
       prompt: messages.reduce((acc, message, index) => {
         if (index % 2 === 0) {
-          return (
-            `${acc}${
-              withBos ? BOS : ""
-            }${B_INST} ${message.content.trim()} ${E_INST}` +
-            (withNewlines ? "\n" : "")
-          );
+          return `${acc}${
+            withBos ? BOS : ""
+          }${B_INST} ${message.content.trim()} ${E_INST}`;
         } else {
-          return (
-            `${acc} ${message.content.trim()}` +
-            (withNewlines ? "\n" : " ") +
-            (withBos ? EOS : "")
-          ); // Yes, the EOS comes after the space. This is not a mistake.
+          return `${acc} ${message.content.trim()} ` + (withBos ? EOS : ""); // Yes, the EOS comes after the space. This is not a mistake.
         }
       }, ""),
       systemPrompt,
     };
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(messages: ChatMessage[], _parentEvent?: Event, streaming?: T): Promise<R> {
+  async chat(
+    messages: ChatMessage[],
+    _parentEvent?: Event,
+  ): Promise<ChatResponse> {
     const api = ALL_AVAILABLE_LLAMADEUCE_MODELS[this.model]
       .replicateApi as `${string}/${string}:${string}`;
 
@@ -613,9 +423,6 @@ If a question does not make any sense, or is not factually coherent, explain why
       replicateOptions.input.max_length = this.maxTokens;
     }
 
-    //TODO: Add streaming for this
-
-    //Non-streaming
     const response = await this.replicateSession.replicate.run(
       api,
       replicateOptions,
@@ -626,32 +433,24 @@ If a question does not make any sense, or is not factually coherent, explain why
         //^ We need to do this because Replicate returns a list of strings (for streaming functionality which is not exposed by the run function)
         role: "assistant",
       },
-    } as R;
+    };
   }
 
-  async complete<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(prompt: string, parentEvent?: Event, streaming?: T): Promise<R> {
+  async complete(
+    prompt: string,
+    parentEvent?: Event,
+  ): Promise<CompletionResponse> {
     return this.chat([{ content: prompt, role: "user" }], parentEvent);
   }
 }
 
-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: 100000 },
-  "claude-instant-1": { contextWindow: 100000 },
-};
-
 /**
  * Anthropic LLM implementation
  */
 
 export class Anthropic implements LLM {
-  hasStreaming: boolean = true;
-
   // Per completion Anthropic params
-  model: keyof typeof ALL_AVAILABLE_ANTHROPIC_MODELS;
+  model: string;
   temperature: number;
   topP: number;
   maxTokens?: number;
@@ -684,21 +483,6 @@ export class Anthropic implements LLM {
     this.callbackManager = init?.callbackManager;
   }
 
-  tokens(messages: ChatMessage[]): number {
-    throw new Error("Method not implemented.");
-  }
-
-  get metadata() {
-    return {
-      model: this.model,
-      temperature: this.temperature,
-      topP: this.topP,
-      maxTokens: this.maxTokens,
-      contextWindow: ALL_AVAILABLE_ANTHROPIC_MODELS[this.model].contextWindow,
-      tokenizer: undefined,
-    };
-  }
-
   mapMessagesToPrompt(messages: ChatMessage[]) {
     return (
       messages.reduce((acc, message) => {
@@ -714,22 +498,10 @@ export class Anthropic implements LLM {
     );
   }
 
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
+  async chat(
     messages: ChatMessage[],
     parentEvent?: Event | undefined,
-    streaming?: T,
-  ): Promise<R> {
-    //Streaming
-    if (streaming) {
-      if (!this.hasStreaming) {
-        throw Error("No streaming support for this LLM.");
-      }
-      return this.streamChat(messages, parentEvent) as R;
-    }
-    //Non-streaming
+  ): Promise<ChatResponse> {
     const response = await this.session.anthropic.completions.create({
       model: this.model,
       prompt: this.mapMessagesToPrompt(messages),
@@ -742,182 +514,12 @@ export class Anthropic implements LLM {
       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.
-    } as R;
+    };
   }
-
-  protected async *streamChat(
-    messages: ChatMessage[],
-    parentEvent?: Event | undefined,
-  ): AsyncGenerator<string, void, unknown> {
-    // 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,
-      });
-
-    var idx_counter: number = 0;
-    for await (const part of stream) {
-      //TODO: LLM Stream Callback, pending re-work.
-
-      idx_counter++;
-      yield part.completion;
-    }
-    return;
-  }
-
-  async complete<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
+  async complete(
     prompt: string,
     parentEvent?: Event | undefined,
-    streaming?: T,
-  ): Promise<R> {
-    if (streaming) {
-      return this.streamComplete(prompt, parentEvent) as R;
-    }
-    return this.chat(
-      [{ content: prompt, role: "user" }],
-      parentEvent,
-      streaming,
-    ) as R;
-  }
-
-  protected streamComplete(
-    prompt: string,
-    parentEvent?: Event | undefined,
-  ): AsyncGenerator<string, void, unknown> {
-    return this.streamChat([{ content: prompt, role: "user" }], parentEvent);
-  }
-}
-
-export class Portkey implements LLM {
-  hasStreaming: boolean = true;
-
-  apiKey?: string = undefined;
-  baseURL?: string = undefined;
-  mode?: string = undefined;
-  llms?: [LLMOptions] | null = undefined;
-  session: PortkeySession;
-  callbackManager?: CallbackManager;
-
-  constructor(init?: Partial<Portkey>) {
-    this.apiKey = init?.apiKey;
-    this.baseURL = init?.baseURL;
-    this.mode = init?.mode;
-    this.llms = init?.llms;
-    this.session = getPortkeySession({
-      apiKey: this.apiKey,
-      baseURL: this.baseURL,
-      llms: this.llms,
-      mode: this.mode,
-    });
-    this.callbackManager = init?.callbackManager;
-  }
-
-  tokens(messages: ChatMessage[]): number {
-    throw new Error("Method not implemented.");
-  }
-
-  get metadata(): LLMMetadata {
-    throw new Error("metadata not implemented for Portkey");
-  }
-
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
-    messages: ChatMessage[],
-    parentEvent?: Event | undefined,
-    streaming?: T,
-    params?: Record<string, any>,
-  ): Promise<R> {
-    if (streaming) {
-      return this.streamChat(messages, parentEvent, params) as R;
-    } else {
-      const resolvedParams = params || {};
-      const response = await this.session.portkey.chatCompletions.create({
-        messages,
-        ...resolvedParams,
-      });
-
-      const content = response.choices[0].message?.content ?? "";
-      const role = response.choices[0].message?.role || "assistant";
-      return { message: { content, role: role as MessageType } } as R;
-    }
-  }
-
-  async complete<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : ChatResponse,
-  >(
-    prompt: string,
-    parentEvent?: Event | undefined,
-    streaming?: T,
-  ): Promise<R> {
-    return this.chat(
-      [{ content: prompt, role: "user" }],
-      parentEvent,
-      streaming,
-    );
-  }
-
-  async *streamChat(
-    messages: ChatMessage[],
-    parentEvent?: Event,
-    params?: Record<string, any>,
-  ): AsyncGenerator<string, void, unknown> {
-    // Wrapping the stream in a callback.
-    const onLLMStream = this.callbackManager?.onLLMStream
-      ? this.callbackManager.onLLMStream
-      : () => {};
-
-    const chunkStream = await this.session.portkey.chatCompletions.create({
-      messages,
-      ...params,
-      stream: true,
-    });
-
-    const event: Event = parentEvent
-      ? parentEvent
-      : {
-          id: "unspecified",
-          type: "llmPredict" as EventType,
-        };
-
-    //Indices
-    var idx_counter: number = 0;
-    for await (const part of chunkStream) {
-      //Increment
-      part.choices[0].index = idx_counter;
-      const is_done: boolean =
-        part.choices[0].finish_reason === "stop" ? true : false;
-      //onLLMStream Callback
-
-      const stream_callback: StreamCallbackResponse = {
-        event: event,
-        index: idx_counter,
-        isDone: is_done,
-        // token: part,
-      };
-      onLLMStream(stream_callback);
-
-      idx_counter++;
-
-      yield part.choices[0].delta?.content ?? "";
-    }
-    return;
-  }
-
-  streamComplete(
-    query: string,
-    parentEvent?: Event,
-  ): AsyncGenerator<string, void, unknown> {
-    return this.streamChat([{ content: query, role: "user" }], parentEvent);
+  ): Promise<CompletionResponse> {
+    return this.chat([{ content: prompt, role: "user" }], parentEvent);
   }
 }