chore: manual export modules

Co-authored-by: Alex Yang <himself65@outlook.com>

.changeset/chatty-parents-smell.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+feat(qdrant): Add Qdrant Vector DB

.changeset/mighty-chicken-trade.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+Preview: Add ingestion pipeline (incl. different strategies to handle doc store duplicates)

.changeset/moody-crabs-sniff.md

@@ -1,5 +0,0 @@
----
-"create-llama": patch
----
-
-fix: re-organize file structure

.changeset/nervous-numbers-swim.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Add an option that allows the user to run the generated app

.changeset/popular-grapes-cheat.md

@@ -0,0 +1,16 @@
+---
+"llamaindex": patch
+---
+
+feat: use conditional exports
+
+The benefit of conditional exports is we split the llamaindex into different files. This will improve the tree shake if you are building web apps.
+
+This also requires node16 (see https://nodejs.org/api/packages.html#conditional-exports).
+
+If you are seeing typescript issue `TS2724`('llamaindex' has no exported member named XXX):
+
+1. update `moduleResolution` to `bundler` in `tsconfig.json`, more for the web applications like Next.js, and vite, but still works for ts-node or tsx.
+2. consider the ES module in your project, add `"type": "module"` into `package.json` and update `moduleResolution` to `node16` or `nodenext` in `tsconfig.json`.
+
+We still support both cjs and esm, but you should update `tsconfig.json` to make the typescript happy.

.changeset/selfish-games-happen.md

@@ -1,5 +0,0 @@
----
-"llamaindex": patch
----
-
-feat: support together AI

.changeset/three-hounds-count.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+feat(extractors): add keyword extractor and base extractor

.devcontainer/devcontainer.json

@@ -4,6 +4,6 @@
     "ghcr.io/devcontainers/features/node:1": {},
     "ghcr.io/devcontainers-contrib/features/turborepo-npm:1": {},
     "ghcr.io/devcontainers-contrib/features/typescript:2": {},
-    "ghcr.io/devcontainers-contrib/features/pnpm:2": {}
-  }
+    "ghcr.io/devcontainers-contrib/features/pnpm:2": {},
+  },
 }

.eslintrc.js

@@ -7,4 +7,8 @@ module.exports = {
       rootDir: ["apps/*/"],
     },
   },
+  rules: {
+    "max-params": ["error", 4],
+  },
+  ignorePatterns: ["dist/"],
 };

.github/workflows/e2e.yml

@@ -8,6 +8,9 @@ on:
       - ".github/workflows/e2e.yml"
     branches: [main]
 
+env:
+  POETRY_VERSION: "1.6.1"
+
 jobs:
   e2e:
     name: create-llama
@@ -16,10 +19,22 @@ jobs:
       fail-fast: true
       matrix:
         node-version: [18, 20]
+        python-version: ["3.11"]
         os: [macos-latest, windows-latest]
+    defaults:
+      run:
+        shell: bash
     runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
+      - name: Set up python ${{ matrix.python-version }}
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install Poetry
+        uses: snok/install-poetry@v1
+        with:
+          version: ${{ env.POETRY_VERSION }}
       - uses: pnpm/action-setup@v2
       - name: Setup Node.js ${{ matrix.node-version }}
         uses: actions/setup-node@v4

.github/workflows/lint_on_push_or_pull.yml

@@ -14,6 +14,8 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - uses: pnpm/action-setup@v2
+        with:
+          version: latest
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:

.gitignore

@@ -39,9 +39,6 @@ yarn-error.log*
 dist/
 lib/
 
-# vs code
-.vscode/launch.json
-
 .cache
 test-results/
 playwright-report/

.prettierignore

@@ -2,3 +2,4 @@ apps/docs/i18n
 pnpm-lock.yaml
 lib/
 dist/
+.docusaurus/

.vscode/launch.json

@@ -0,0 +1,17 @@
+{
+  // Use IntelliSense to learn about possible attributes.
+  // Hover to view descriptions of existing attributes.
+  // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+  "version": "0.2.0",
+  "configurations": [
+    {
+      "type": "node",
+      "request": "launch",
+      "name": "Debug Example",
+      "skipFiles": ["<node_internals>/**"],
+      "runtimeExecutable": "pnpm",
+      "cwd": "${workspaceFolder}/examples",
+      "runtimeArgs": ["ts-node", "${fileBasename}"]
+    }
+  ]
+}

README.md

@@ -52,9 +52,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());
@@ -106,7 +106,6 @@ const nextConfig = {
       ...config.resolve.alias,
       sharp$: false,
       "onnxruntime-node$": false,
-      mongodb$: false,
     };
     return config;
   },

apps/docs/docs/end_to_end.md

@@ -26,7 +26,7 @@ Create and load a vector index. Persistance to disk in LlamaIndex.TS happens aut
 
 ## [Customized Vector Index](https://github.com/run-llama/LlamaIndexTS/blob/main/examples/vectorIndexCustomize.ts)
 
-Create a vector index and query it, while also configuring the the `LLM`, the `ServiceContext`, and the `similarity_top_k`.
+Create a vector index and query it, while also configuring the `LLM`, the `ServiceContext`, and the `similarity_top_k`.
 
 ## [OpenAI LLM](https://github.com/run-llama/LlamaIndexTS/blob/main/examples/openai.ts)
 

apps/docs/docs/modules/high_level/chat_engine.md

@@ -11,7 +11,16 @@ const retriever = index.asRetriever();
 const chatEngine = new ContextChatEngine({ retriever });
 
 // start chatting
-const response = await chatEngine.chat(query);
+const response = await chatEngine.chat({ message: query });
+```
+
+The `chat` function also supports streaming, just add `stream: true` as an option:
+
+```typescript
+const stream = await chatEngine.chat({ message: query, stream: true });
+for await (const chunk of stream) {
+  process.stdout.write(chunk.response);
+}
 ```
 
 ## Api References

apps/docs/docs/modules/high_level/query_engine.md

@@ -8,7 +8,16 @@ A query engine wraps a `Retriever` and a `ResponseSynthesizer` into a pipeline,
 
 ```typescript
 const queryEngine = index.asQueryEngine();
-const response = await queryEngine.query("query string");
+const response = await queryEngine.query({ query: "query string" });
+```
+
+The `query` function also supports streaming, just add `stream: true` as an option:
+
+```typescript
+const stream = await queryEngine.query({ query: "query string", stream: true });
+for await (const chunk of stream) {
+  process.stdout.write(chunk.response);
+}
 ```
 
 ## Sub Question Query Engine

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

@@ -35,13 +35,26 @@ const nodesWithScore: NodeWithScore[] = [
   },
 ];
 
-const response = await responseSynthesizer.synthesize(
-  "What age am I?",
+const response = await responseSynthesizer.synthesize({
+  query: "What age am I?",
   nodesWithScore,
-);
+});
 console.log(response.response);
 ```
 
+The `synthesize` function also supports streaming, just add `stream: true` as an option:
+
+```typescript
+const stream = await responseSynthesizer.synthesize({
+  query: "What age am I?",
+  nodesWithScore,
+  stream: true,
+});
+for await (const chunk of stream) {
+  process.stdout.write(chunk.response);
+}
+```
+
 ## API Reference
 
 - [ResponseSynthesizer](../../api/classes/ResponseSynthesizer.md)

apps/docs/docs/modules/observability/_category_.yml

@@ -0,0 +1 @@
+label: Observability

apps/docs/docs/modules/observability/index.md

@@ -0,0 +1,35 @@
+# Observability
+
+LlamaIndex provides **one-click observability** 🔭 to allow you to build principled LLM applications in a production setting.
+
+A key requirement for principled development of LLM applications over your data (RAG systems, agents) is being able to observe, debug, and evaluate
+your system - both as a whole and for each component.
+
+This feature allows you to seamlessly integrate the LlamaIndex library with powerful observability/evaluation tools offered by our partners.
+Configure a variable once, and you'll be able to do things like the following:
+
+- View LLM/prompt inputs/outputs
+- Ensure that the outputs of any component (LLMs, embeddings) are performing as expected
+- View call traces for both indexing and querying
+
+Each provider has similarities and differences. Take a look below for the full set of guides for each one!
+
+## OpenLLMetry
+
+[OpenLLMetry](https://github.com/traceloop/openllmetry-js) is an open-source project based on OpenTelemetry for tracing and monitoring
+LLM applications. It connects to [all major observability platforms](https://www.traceloop.com/docs/openllmetry/integrations/introduction) and installs in minutes.
+
+### Usage Pattern
+
+```bash
+npm install @traceloop/node-server-sdk
+```
+
+```js
+import * as traceloop from "@traceloop/node-server-sdk";
+
+traceloop.initialize({
+  apiKey: process.env.TRACELOOP_API_KEY,
+  disableBatch: true,
+});
+```

apps/docs/tsconfig.json

@@ -6,6 +6,6 @@
     "composite": true,
     "incremental": true,
     "outDir": "./lib",
-    "tsBuildInfoFile": "./lib/.tsbuildinfo"
-  }
+    "tsBuildInfoFile": "./lib/.tsbuildinfo",
+  },
 }

examples/anthropic.ts

@@ -2,13 +2,15 @@ import { Anthropic } from "llamaindex";
 
 (async () => {
   const anthropic = new Anthropic();
-  const result = await anthropic.chat([
-    { content: "You want to talk in rhymes.", role: "system" },
-    {
-      content:
-        "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
-      role: "user",
-    },
-  ]);
+  const result = await anthropic.chat({
+    messages: [
+      { content: "You want to talk in rhymes.", role: "system" },
+      {
+        content:
+          "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
+        role: "user",
+      },
+    ],
+  });
   console.log(result);
 })();

examples/astradb/query.ts

@@ -18,7 +18,9 @@ async function main() {
 
     const queryEngine = await index.asQueryEngine({ retriever });
 
-    const results = await queryEngine.query("What is the best reviewed movie?");
+    const results = await queryEngine.query({
+      query: "What is the best reviewed movie?",
+    });
 
     console.log(results.response);
   } catch (e) {

examples/chatEngine.ts

@@ -10,7 +10,7 @@ import {
   VectorStoreIndex,
 } from "llamaindex";
 
-import essay from "./essay";
+import essay from "./essay.js";
 
 async function main() {
   const document = new Document({ text: essay });
@@ -25,8 +25,11 @@ async function main() {
 
   while (true) {
     const query = await rl.question("Query: ");
-    const response = await chatEngine.chat(query);
-    console.log(response.toString());
+    const stream = await chatEngine.chat({ message: query, stream: true });
+    console.log();
+    for await (const chunk of stream) {
+      process.stdout.write(chunk.response);
+    }
   }
 }
 

examples/chatHistory.ts

@@ -0,0 +1,32 @@
+import { stdin as input, stdout as output } from "node:process";
+import readline from "node:readline/promises";
+
+import { OpenAI, SimpleChatEngine, SummaryChatHistory } from "llamaindex";
+
+async function main() {
+  // Set maxTokens to 75% of the context window size of 4096
+  // This will trigger the summarizer once the chat history reaches 25% of the context window size (1024 tokens)
+  const llm = new OpenAI({ model: "gpt-3.5-turbo", maxTokens: 4096 * 0.75 });
+  const chatHistory = new SummaryChatHistory({ llm });
+  const chatEngine = new SimpleChatEngine({ llm });
+  const rl = readline.createInterface({ input, output });
+
+  while (true) {
+    const query = await rl.question("Query: ");
+    const stream = await chatEngine.chat({
+      message: query,
+      chatHistory,
+      stream: true,
+    });
+    if (chatHistory.getLastSummary()) {
+      // Print the summary of the conversation so far that is produced by the SummaryChatHistory
+      console.log(`Summary: ${chatHistory.getLastSummary()?.content}`);
+    }
+    for await (const chunk of stream) {
+      process.stdout.write(chunk.response);
+    }
+    console.log();
+  }
+}
+
+main().catch(console.error);

examples/chromadb/preFilters.ts

@@ -0,0 +1,57 @@
+import {
+  ChromaVectorStore,
+  Document,
+  VectorStoreIndex,
+  storageContextFromDefaults,
+} from "llamaindex";
+
+const collectionName = "dog_colors";
+
+async function main() {
+  try {
+    const docs = [
+      new Document({
+        text: "The dog is brown",
+        metadata: {
+          dogId: "1",
+        },
+      }),
+      new Document({
+        text: "The dog is red",
+        metadata: {
+          dogId: "2",
+        },
+      }),
+    ];
+
+    console.log("Creating ChromaDB vector store");
+    const chromaVS = new ChromaVectorStore({ collectionName });
+    const ctx = await storageContextFromDefaults({ vectorStore: chromaVS });
+
+    console.log("Embedding documents and adding to index");
+    const index = await VectorStoreIndex.fromDocuments(docs, {
+      storageContext: ctx,
+    });
+
+    console.log("Querying index");
+    const queryEngine = index.asQueryEngine({
+      preFilters: {
+        filters: [
+          {
+            key: "dogId",
+            value: "2",
+            filterType: "ExactMatch",
+          },
+        ],
+      },
+    });
+    const response = await queryEngine.query({
+      query: "What is the color of the dog?",
+    });
+    console.log(response.toString());
+  } catch (e) {
+    console.error(e);
+  }
+}
+
+main();

examples/chromadb/test.ts

@@ -28,9 +28,9 @@ async function main() {
 
     console.log("Querying index");
     const queryEngine = index.asQueryEngine();
-    const response = await queryEngine.query(
-      "Tell me about Godfrey Cheshire's rating of La Sapienza.",
-    );
+    const response = await queryEngine.query({
+      query: "Tell me about Godfrey Cheshire's rating of La Sapienza.",
+    });
     console.log(response.toString());
   } catch (e) {
     console.error(e);

examples/extractors/keywordExtractor.ts

@@ -0,0 +1,24 @@
+import {
+  Document,
+  KeywordExtractor,
+  OpenAI,
+  SimpleNodeParser,
+} from "llamaindex";
+
+(async () => {
+  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0 });
+
+  const nodeParser = new SimpleNodeParser();
+
+  const nodes = nodeParser.getNodesFromDocuments([
+    new Document({ text: "banana apple orange pear peach watermelon" }),
+  ]);
+
+  console.log(nodes);
+
+  const keywordExtractor = new KeywordExtractor(openaiLLM, 5);
+
+  const nodesWithKeywordMetadata = await keywordExtractor.processNodes(nodes);
+
+  process.stdout.write(JSON.stringify(nodesWithKeywordMetadata, null, 2));
+})();

examples/extractors/questionsAnsweredExtractor.ts

@@ -0,0 +1,31 @@
+import {
+  Document,
+  OpenAI,
+  QuestionsAnsweredExtractor,
+  SimpleNodeParser,
+} from "llamaindex";
+
+(async () => {
+  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0 });
+
+  const nodeParser = new SimpleNodeParser();
+
+  const nodes = nodeParser.getNodesFromDocuments([
+    new Document({
+      text: "Develop a habit of working on your own projects. Don't let work mean something other people tell you to do. If you do manage to do great work one day, it will probably be on a project of your own. It may be within some bigger project, but you'll be driving your part of it.",
+    }),
+    new Document({
+      text: "The best way to get a good idea is to get a lot of ideas. The best way to get a lot of ideas is to get a lot of bad ideas. The best way to get a lot of bad ideas is to get a lot of ideas.",
+    }),
+  ]);
+
+  const questionsAnsweredExtractor = new QuestionsAnsweredExtractor(
+    openaiLLM,
+    5,
+  );
+
+  const nodesWithQuestionsMetadata =
+    await questionsAnsweredExtractor.processNodes(nodes);
+
+  process.stdout.write(JSON.stringify(nodesWithQuestionsMetadata, null, 2));
+})();

examples/extractors/summaryExtractor.ts

@@ -0,0 +1,24 @@
+import {
+  Document,
+  OpenAI,
+  SimpleNodeParser,
+  SummaryExtractor,
+} from "llamaindex";
+
+(async () => {
+  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0 });
+
+  const nodeParser = new SimpleNodeParser();
+
+  const nodes = nodeParser.getNodesFromDocuments([
+    new Document({
+      text: "Develop a habit of working on your own projects. Don't let work mean something other people tell you to do. If you do manage to do great work one day, it will probably be on a project of your own. It may be within some bigger project, but you'll be driving your part of it.",
+    }),
+  ]);
+
+  const summaryExtractor = new SummaryExtractor(openaiLLM);
+
+  const nodesWithSummaryMetadata = await summaryExtractor.processNodes(nodes);
+
+  process.stdout.write(JSON.stringify(nodesWithSummaryMetadata, null, 2));
+})();

examples/extractors/titleExtractor.ts

@@ -0,0 +1,19 @@
+import { Document, OpenAI, SimpleNodeParser, TitleExtractor } from "llamaindex";
+
+(async () => {
+  const openaiLLM = new OpenAI({ model: "gpt-3.5-turbo", temperature: 0 });
+
+  const nodeParser = new SimpleNodeParser();
+
+  const nodes = nodeParser.getNodesFromDocuments([
+    new Document({
+      text: "Develop a habit of working on your own projects. Don't let work mean something other people tell you to do. If you do manage to do great work one day, it will probably be on a project of your own. It may be within some bigger project, but you'll be driving your part of it.",
+    }),
+  ]);
+
+  const titleExtractor = new TitleExtractor(openaiLLM, 1);
+
+  const nodesWithTitledMetadata = await titleExtractor.processNodes(nodes);
+
+  process.stdout.write(JSON.stringify(nodesWithTitledMetadata, null, 2));
+})();

examples/gptllama.ts

@@ -25,12 +25,12 @@ import { ChatMessage, LlamaDeuce, OpenAI } from "llamaindex";
 
   while (true) {
     const next = history.length % 2 === 1 ? gpt4 : l2;
-    const r = await next.chat(
-      history.map(({ content, role }) => ({
+    const r = await next.chat({
+      messages: history.map(({ content, role }) => ({
         content,
         role: next === l2 ? role : role === "user" ? "assistant" : "user",
       })),
-    );
+    });
     history.push({
       content: r.message.content,
       role: next === l2 ? "assistant" : "user",

examples/huggingface.ts

@@ -32,12 +32,15 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const stream = await queryEngine.query({
+    query: "What did the author do in college?",
+    stream: true,
+  });
 
   // Output response
-  console.log(response.toString());
+  for await (const chunk of stream) {
+    process.stdout.write(chunk.response);
+  }
 }
 
 main().catch(console.error);

examples/jsonExtract.ts

@@ -21,18 +21,20 @@ async function main() {
     action_items: ["action item 1", "action item 2"],
   };
 
-  const response = await llm.chat([
-    {
-      role: "system",
-      content: `You are an expert assistant for summarizing and extracting insights from sales call transcripts.\n\nGenerate a valid JSON in the following format:\n\n${JSON.stringify(
-        example,
-      )}`,
-    },
-    {
-      role: "user",
-      content: `Here is the transcript: \n------\n${transcript}\n------`,
-    },
-  ]);
+  const response = await llm.chat({
+    messages: [
+      {
+        role: "system",
+        content: `You are an expert assistant for summarizing and extracting insights from sales call transcripts.\n\nGenerate a valid JSON in the following format:\n\n${JSON.stringify(
+          example,
+        )}`,
+      },
+      {
+        role: "user",
+        content: `Here is the transcript: \n------\n${transcript}\n------`,
+      },
+    ],
+  });
 
   const json = JSON.parse(response.message.content);
 

examples/keywordIndex.ts

@@ -3,7 +3,7 @@ import {
   KeywordTableIndex,
   KeywordTableRetrieverMode,
 } from "llamaindex";
-import essay from "./essay";
+import essay from "./essay.js";
 
 async function main() {
   const document = new Document({ text: essay, id_: "essay" });
@@ -20,9 +20,9 @@ async function main() {
         mode,
       }),
     });
-    const response = await queryEngine.query(
-      "What did the author do growing up?",
-    );
+    const response = await queryEngine.query({
+      query: "What did the author do growing up?",
+    });
     console.log(response.toString());
   });
 }

examples/llamadeuce.ts

@@ -2,6 +2,8 @@ import { DeuceChatStrategy, LlamaDeuce } from "llamaindex";
 
 (async () => {
   const deuce = new LlamaDeuce({ chatStrategy: DeuceChatStrategy.META });
-  const result = await deuce.chat([{ content: "Hello, world!", role: "user" }]);
+  const result = await deuce.chat({
+    messages: [{ content: "Hello, world!", role: "user" }],
+  });
   console.log(result);
 })();

examples/lowlevel.ts

@@ -27,9 +27,12 @@ import {
     },
   ];
 
-  const response = await responseSynthesizer.synthesize(
-    "What age am I?",
+  const stream = await responseSynthesizer.synthesize({
+    query: "What age am I?",
     nodesWithScore,
-  );
-  console.log(response.response);
+    stream: true,
+  });
+  for await (const chunk of stream) {
+    process.stdout.write(chunk.response);
+  }
 })();

examples/markdown.ts

@@ -11,7 +11,9 @@ async function main() {
   // Query the index
   const queryEngine = index.asQueryEngine();
 
-  const response = await queryEngine.query("What does the example code do?");
+  const response = await queryEngine.query({
+    query: "What does the example code do?",
+  });
 
   // Output response
   console.log(response.toString());

examples/mistral.ts

@@ -27,7 +27,7 @@ async function rag(llm: LLM, embedModel: BaseEmbedding, query: string) {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(query);
+  const response = await queryEngine.query({ query });
   return response.response;
 }
 
@@ -43,19 +43,20 @@ async function rag(llm: LLM, embedModel: BaseEmbedding, query: string) {
 
   // chat api (non-streaming)
   const llm = new MistralAI({ model: "mistral-tiny" });
-  const response = await llm.chat([
-    { content: "What is the best French cheese?", role: "user" },
-  ]);
+  const response = await llm.chat({
+    messages: [{ content: "What is the best French cheese?", role: "user" }],
+  });
   console.log(response.message.content);
 
   // chat api (streaming)
-  const stream = await llm.chat(
-    [{ content: "Who is the most renowned French painter?", role: "user" }],
-    undefined,
-    true,
-  );
+  const stream = await llm.chat({
+    messages: [
+      { content: "Who is the most renowned French painter?", role: "user" },
+    ],
+    stream: true,
+  });
   for await (const chunk of stream) {
-    process.stdout.write(chunk);
+    process.stdout.write(chunk.delta);
   }
 
   // rag

examples/mongo.ts

@@ -54,7 +54,7 @@ async function main() {
       break;
     }
 
-    const response = await queryEngine.query(query);
+    const response = await queryEngine.query({ query });
 
     // Output response
     console.log(response.toString());

examples/mongodb/3_query.ts

@@ -24,9 +24,9 @@ async function query() {
 
   const retriever = index.asRetriever({ similarityTopK: 20 });
   const queryEngine = index.asQueryEngine({ retriever });
-  const result = await queryEngine.query(
-    "What does the author think of web frameworks?",
-  );
+  const result = await queryEngine.query({
+    query: "What does the author think of web frameworks?",
+  });
   console.log(result.response);
   await client.close();
 }

examples/multimodal/rag.ts

@@ -46,9 +46,9 @@ async function main() {
     responseSynthesizer: new MultiModalResponseSynthesizer({ serviceContext }),
     retriever: index.asRetriever({ similarityTopK: 3, imageSimilarityTopK: 1 }),
   });
-  const result = await queryEngine.query(
-    "Tell me more about Vincent van Gogh's famous paintings",
-  );
+  const result = await queryEngine.query({
+    query: "Tell me more about Vincent van Gogh's famous paintings",
+  });
   console.log(result.response, "\n");
   images.forEach((image) =>
     console.log(`Image retrieved and used in inference: ${image.toString()}`),

examples/ollama.ts

@@ -3,32 +3,34 @@ import { Ollama } from "llamaindex";
 (async () => {
   const llm = new Ollama({ model: "llama2", temperature: 0.75 });
   {
-    const response = await llm.chat([
-      { content: "Tell me a joke.", role: "user" },
-    ]);
+    const response = await llm.chat({
+      messages: [{ content: "Tell me a joke.", role: "user" }],
+    });
     console.log("Response 1:", response.message.content);
   }
   {
-    const response = await llm.complete("How are you?");
-    console.log("Response 2:", response.message.content);
+    const response = await llm.complete({ prompt: "How are you?" });
+    console.log("Response 2:", response.text);
   }
   {
-    const response = await llm.chat(
-      [{ content: "Tell me a joke.", role: "user" }],
-      undefined,
-      true,
-    );
+    const response = await llm.chat({
+      messages: [{ content: "Tell me a joke.", role: "user" }],
+      stream: true,
+    });
     console.log("Response 3:");
     for await (const message of response) {
-      process.stdout.write(message); // no newline
+      process.stdout.write(message.delta); // no newline
     }
     console.log(); // newline
   }
   {
-    const response = await llm.complete("How are you?", undefined, true);
+    const response = await llm.complete({
+      prompt: "How are you?",
+      stream: true,
+    });
     console.log("Response 4:");
     for await (const message of response) {
-      process.stdout.write(message); // no newline
+      process.stdout.write(message.text); // no newline
     }
     console.log(); // newline
   }

examples/openai.ts

@@ -4,12 +4,12 @@ import { OpenAI } from "llamaindex";
   const llm = new OpenAI({ model: "gpt-4-1106-preview", temperature: 0.1 });
 
   // complete api
-  const response1 = await llm.complete("How are you?");
-  console.log(response1.message.content);
+  const response1 = await llm.complete({ prompt: "How are you?" });
+  console.log(response1.text);
 
   // chat api
-  const response2 = await llm.chat([
-    { content: "Tell me a joke.", role: "user" },
-  ]);
+  const response2 = await llm.chat({
+    messages: [{ content: "Tell me a joke.", role: "user" }],
+  });
   console.log(response2.message.content);
 })();

examples/pg-vector-store/query.ts

@@ -31,7 +31,7 @@ async function main() {
       }
 
       try {
-        const answer = await queryEngine.query(question);
+        const answer = await queryEngine.query({ query: question });
         console.log(answer.response);
       } catch (error) {
         console.error("Error:", error);

examples/pinecone-vector-store/query.ts

@@ -29,7 +29,7 @@ async function main() {
       }
 
       try {
-        const answer = await queryEngine.query(question);
+        const answer = await queryEngine.query({ query: question });
         console.log(answer.response);
       } catch (error) {
         console.error("Error:", error);

examples/pipeline/ingestion.ts

@@ -0,0 +1,36 @@
+import fs from "node:fs/promises";
+
+import {
+  Document,
+  IngestionPipeline,
+  MetadataMode,
+  OpenAIEmbedding,
+  SimpleNodeParser,
+} from "llamaindex";
+
+async function main() {
+  // Load essay from abramov.txt in Node
+  const path = "node_modules/llamaindex/examples/abramov.txt";
+
+  const essay = await fs.readFile(path, "utf-8");
+
+  // Create Document object with essay
+  const document = new Document({ text: essay, id_: path });
+  const pipeline = new IngestionPipeline({
+    transformations: [
+      new SimpleNodeParser({ chunkSize: 1024, chunkOverlap: 20 }),
+      // new TitleExtractor(llm),
+      new OpenAIEmbedding(),
+    ],
+  });
+
+  // run the pipeline
+  const nodes = await pipeline.run({ documents: [document] });
+
+  // print out the result of the pipeline run
+  for (const node of nodes) {
+    console.log(node.getContent(MetadataMode.NONE));
+  }
+}
+
+main().catch(console.error);

examples/portkey.ts

@@ -12,11 +12,14 @@ import { Portkey } from "llamaindex";
       },
     ],
   });
-  const result = portkey.streamChat([
-    { role: "system", content: "You are a helpful assistant." },
-    { role: "user", content: "Tell me a joke." },
-  ]);
+  const result = await portkey.chat({
+    messages: [
+      { role: "system", content: "You are a helpful assistant." },
+      { role: "user", content: "Tell me a joke." },
+    ],
+    stream: true,
+  });
   for await (const res of result) {
-    process.stdout.write(res);
+    process.stdout.write(res.delta);
   }
 })();

examples/readers/load-assemblyai.ts

@@ -48,7 +48,7 @@ program
         break;
       }
 
-      const response = await queryEngine.query(query);
+      const response = await queryEngine.query({ query });
 
       console.log(response.toString());
     }

examples/readers/load-csv.ts

@@ -38,9 +38,9 @@ Given the CSV file, generate me Typescript code to answer the question: ${query}
   const queryEngine = index.asQueryEngine({ responseSynthesizer });
 
   // Query the index
-  const response = await queryEngine.query(
-    "What is the correlation between survival and age?",
-  );
+  const response = await queryEngine.query({
+    query: "What is the correlation between survival and age?",
+  });
 
   // Output response
   console.log(response.toString());

examples/readers/load-docx.ts

@@ -15,7 +15,7 @@ async function main() {
 
   // Test query
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(SAMPLE_QUERY);
+  const response = await queryEngine.query({ query: SAMPLE_QUERY });
   console.log(`Test query > ${SAMPLE_QUERY}:\n`, response.toString());
 }
 

examples/readers/load-html.ts

@@ -10,9 +10,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What were the notable changes in 18.1?",
-  );
+  const response = await queryEngine.query({
+    query: "What were the notable changes in 18.1?",
+  });
 
   // Output response
   console.log(response.toString());

examples/readers/load-md.ts

@@ -15,7 +15,7 @@ async function main() {
 
   // Test query
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(SAMPLE_QUERY);
+  const response = await queryEngine.query({ query: SAMPLE_QUERY });
   console.log(`Test query > ${SAMPLE_QUERY}:\n`, response.toString());
 }
 

examples/readers/load-notion.ts

@@ -79,7 +79,7 @@ program
         break;
       }
 
-      const response = await queryEngine.query(query);
+      const response = await queryEngine.query({ query });
 
       // Output response
       console.log(response.toString());

examples/readers/load-pdf.ts

@@ -10,7 +10,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query("What mistakes did they make?");
+  const response = await queryEngine.query({
+    query: "What mistakes did they make?",
+  });
 
   // Output response
   console.log(response.toString());

examples/sentenceWindow.ts

@@ -6,7 +6,7 @@ import {
   VectorStoreIndex,
   serviceContextFromDefaults,
 } from "llamaindex";
-import essay from "./essay";
+import essay from "./essay.js";
 
 async function main() {
   const document = new Document({ text: essay, id_: "essay" });
@@ -31,9 +31,9 @@ async function main() {
   const queryEngine = index.asQueryEngine({
     nodePostprocessors: [new MetadataReplacementPostProcessor("window")],
   });
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());

examples/storageContext.ts

@@ -3,7 +3,7 @@ import {
   storageContextFromDefaults,
   VectorStoreIndex,
 } from "llamaindex";
-import essay from "./essay";
+import essay from "./essay.js";
 
 async function main() {
   // Create Document object with essay
@@ -20,9 +20,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());
@@ -35,9 +35,9 @@ async function main() {
     storageContext: secondStorageContext,
   });
   const loadedQueryEngine = loadedIndex.asQueryEngine();
-  const loadedResponse = await loadedQueryEngine.query(
-    "What did the author do growing up?",
-  );
+  const loadedResponse = await loadedQueryEngine.query({
+    query: "What did the author do growing up?",
+  });
   console.log(loadedResponse.toString());
 }
 

examples/subquestion.ts

@@ -1,6 +1,6 @@
 import { Document, SubQuestionQueryEngine, VectorStoreIndex } from "llamaindex";
 
-import essay from "./essay";
+import essay from "./essay.js";
 
 (async () => {
   const document = new Document({ text: essay, id_: essay });
@@ -18,9 +18,9 @@ import essay from "./essay";
     ],
   });
 
-  const response = await queryEngine.query(
-    "How was Paul Grahams life different before and after YC?",
-  );
+  const response = await queryEngine.query({
+    query: "How was Paul Grahams life different before and after YC?",
+  });
 
   console.log(response.toString());
 })();

examples/summaryIndex.ts

@@ -5,7 +5,7 @@ import {
   SummaryRetrieverMode,
   serviceContextFromDefaults,
 } from "llamaindex";
-import essay from "./essay";
+import essay from "./essay.js";
 
 async function main() {
   const serviceContext = serviceContextFromDefaults({
@@ -20,9 +20,9 @@ async function main() {
   const queryEngine = index.asQueryEngine({
     retriever: index.asRetriever({ mode: SummaryRetrieverMode.LLM }),
   });
-  const response = await queryEngine.query(
-    "What did the author do growing up?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do growing up?",
+  });
   console.log(response.toString());
 }
 

examples/together-ai.ts

@@ -6,8 +6,8 @@ const together = new TogetherLLM({
 });
 
 (async () => {
-  const generator = await together.chat(
-    [
+  const generator = await together.chat({
+    messages: [
       {
         role: "system",
         content: "You are an AI assistant",
@@ -17,12 +17,11 @@ const together = new TogetherLLM({
         content: "Tell me about San Francisco",
       },
     ],
-    undefined,
-    true,
-  );
+    stream: true,
+  });
   console.log("Chatting with Together AI...");
   for await (const message of generator) {
-    process.stdout.write(message);
+    process.stdout.write(message.delta);
   }
   const embedding = new TogetherEmbedding();
   const vector = await embedding.getTextEmbedding("Hello world!");

examples/together-ai/vector-index.ts

@@ -0,0 +1,39 @@
+import fs from "node:fs/promises";
+
+import {
+  Document,
+  TogetherEmbedding,
+  TogetherLLM,
+  VectorStoreIndex,
+  serviceContextFromDefaults,
+} from "llamaindex";
+
+async function main() {
+  const apiKey = process.env.TOGETHER_API_KEY;
+  if (!apiKey) {
+    throw new Error("Missing TOGETHER_API_KEY");
+  }
+  const path = require.resolve("llamaindex/examples/abramov.txt");
+  const essay = await fs.readFile(path, "utf-8");
+
+  const document = new Document({ text: essay, id_: path });
+
+  const serviceContext = serviceContextFromDefaults({
+    llm: new TogetherLLM({ model: "mistralai/Mixtral-8x7B-Instruct-v0.1" }),
+    embedModel: new TogetherEmbedding(),
+  });
+
+  const index = await VectorStoreIndex.fromDocuments([document], {
+    serviceContext,
+  });
+
+  const queryEngine = index.asQueryEngine();
+
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
+
+  console.log(response.toString());
+}
+
+main().catch(console.error);

examples/tsconfig.json

@@ -1,12 +1,10 @@
 {
   "extends": "../tsconfig.json",
-  "compilerOptions": {
-    "target": "es2016",
-    "module": "commonjs",
-    "esModuleInterop": true,
-    "forceConsistentCasingInFileNames": true,
-    "strict": true,
-    "skipLibCheck": true
+  "ts-node": {
+    "files": true,
+    "compilerOptions": {
+      "module": "commonjs",
+    },
   },
-  "include": ["./**/*.ts"]
+  "include": ["./**/*.ts"],
 }

examples/vectorIndex.ts

@@ -16,9 +16,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());

examples/vectorIndexAnthropic.ts

@@ -35,9 +35,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine({ responseSynthesizer });
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());

examples/vectorIndexCustomize.ts

@@ -6,7 +6,7 @@ import {
   SimilarityPostprocessor,
   VectorStoreIndex,
 } from "llamaindex";
-import essay from "./essay";
+import essay from "./essay.js";
 
 // Customize retrieval and query args
 async function main() {
@@ -33,9 +33,9 @@ async function main() {
     [nodePostprocessor],
   );
 
-  const response = await queryEngine.query(
-    "What did the author do growing up?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do growing up?",
+  });
   console.log(response.response);
 }
 

examples/vectorIndexFromVectorStore.ts

@@ -189,7 +189,9 @@ async function main() {
     },
   });
 
-  const response = await queryEngine.query("How many results do you have?");
+  const response = await queryEngine.query({
+    query: "How many results do you have?",
+  });
 
   console.log(response.toString());
 }

examples/vectorIndexGPT4.ts

@@ -25,9 +25,9 @@ async function main() {
 
   // Query the index
   const queryEngine = index.asQueryEngine();
-  const response = await queryEngine.query(
-    "What did the author do in college?",
-  );
+  const response = await queryEngine.query({
+    query: "What did the author do in college?",
+  });
 
   // Output response
   console.log(response.toString());

examples/vision.ts

@@ -4,12 +4,12 @@ import { OpenAI } from "llamaindex";
   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);
+  const response1 = await llm.complete({ prompt: "How are you?" });
+  console.log(response1.text);
 
   // chat api
-  const response2 = await llm.chat([
-    { content: "Tell me a joke!", role: "user" },
-  ]);
+  const response2 = await llm.chat({
+    messages: [{ content: "Tell me a joke!", role: "user" }],
+  });
   console.log(response2.message.content);
 })();

package.json

@@ -1,7 +1,9 @@
 {
+  "name": "@llamaindex/monorepo",
   "private": true,
   "scripts": {
     "build": "turbo run build",
+    "build:release": "turbo run build lint test --filter=\"!docs\"",
     "dev": "turbo run dev",
     "format": "prettier --ignore-unknown --cache --check .",
     "format:write": "prettier --ignore-unknown --write .",
@@ -9,8 +11,9 @@
     "prepare": "husky install",
     "test": "turbo run test",
     "type-check": "tsc -b --diagnostics",
-    "new-version": "turbo run build lint test --filter=\"!docs\" && changeset version",
-    "new-snapshot": "turbo run build lint test --filter=\"!docs\" && changeset version --snapshot"
+    "release": "pnpm run build:release && changeset publish",
+    "new-version": "pnpm run build:release && changeset version",
+    "new-snapshot": "pnpm run build:release && changeset version --snapshot"
   },
   "devDependencies": {
     "@changesets/cli": "^2.27.1",
@@ -21,7 +24,7 @@
     "husky": "^8.0.3",
     "jest": "^29.7.0",
     "lint-staged": "^15.2.0",
-    "prettier": "^3.1.1",
+    "prettier": "^3.2.4",
     "prettier-plugin-organize-imports": "^3.2.4",
     "ts-jest": "^29.1.1",
     "turbo": "^1.11.2",

packages/core/CHANGELOG.md

@@ -1,5 +1,32 @@
 # llamaindex
 
+## 0.0.48
+
+### Patch Changes
+
+- 34a26e5: Remove HistoryChatEngine and use ChatHistory for all chat engines
+
+## 0.0.47
+
+### Patch Changes
+
+- 844029d: Add streaming support for QueryEngine (and unify streaming interface with ChatEngine)
+- 844029d: Breaking: Use parameter object for query and chat methods of ChatEngine and QueryEngine
+
+## 0.0.46
+
+### Patch Changes
+
+- 977f284: fixing import statement
+- 5d3bb66: fix: class SimpleKVStore might throw error in ES module
+- f18c9f6: refactor: Updated low-level streaming interface
+
+## 0.0.45
+
+### Patch Changes
+
+- 2e6b36e: feat: support together AI
+
 ## 0.0.44
 
 ### Patch Changes

packages/core/README.md

@@ -106,7 +106,6 @@ const nextConfig = {
       ...config.resolve.alias,
       sharp$: false,
       "onnxruntime-node$": false,
-      mongodb$: false,
     };
     return config;
   },

packages/core/jest.config.cjs

@@ -2,4 +2,5 @@
 module.exports = {
   preset: "ts-jest",
   testEnvironment: "node",
+  testPathIgnorePatterns: ["/lib/"],
 };

packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llamaindex",
-  "version": "0.0.44",
+  "version": "0.0.48",
   "license": "MIT",
   "dependencies": {
     "@anthropic-ai/sdk": "^0.9.1",
@@ -8,6 +8,7 @@
     "@mistralai/mistralai": "^0.0.7",
     "@notionhq/client": "^2.2.14",
     "@pinecone-database/pinecone": "^1.1.2",
+    "@qdrant/js-client-rest": "^1.7.0",
     "@xenova/transformers": "^2.10.0",
     "assemblyai": "^4.0.0",
     "chromadb": "^1.7.3",
@@ -30,12 +31,14 @@
     "wink-nlp": "^1.14.3"
   },
   "devDependencies": {
+    "@aws-crypto/sha256-js": "^5.2.0",
     "@types/jest": "^29.5.11",
     "@types/lodash": "^4.14.202",
     "@types/node": "^18.19.6",
     "@types/papaparse": "^5.3.14",
     "@types/pg": "^8.10.9",
-    "bunchee": "^4.3.3",
+    "bunchee": "^4.4.1",
+    "madge": "^6.1.0",
     "node-stdlib-browser": "^1.2.0",
     "typescript": "^5.3.3"
   },
@@ -47,8 +50,151 @@
   "exports": {
     ".": {
       "types": "./dist/index.d.mts",
+      "edge-light": "./dist/index.edge-light.mjs",
       "import": "./dist/index.mjs",
       "require": "./dist/index.js"
+    },
+    "./env": {
+      "types": "./dist/env.d.mts",
+      "edge-light": "./dist/env.edge-light.mjs",
+      "import": "./dist/env.mjs",
+      "require": "./dist/env.js"
+    },
+    "./storage/FileSystem": {
+      "types": "./dist/storage/FileSystem.d.mts",
+      "edge-light": "./dist/storage/FileSystem.edge-light.mjs",
+      "import": "./dist/storage/FileSystem.mjs",
+      "require": "./dist/storage/FileSystem.js"
+    },
+    "./ChatHistory": {
+      "types": "./dist/ChatHistory.d.mts",
+      "import": "./dist/ChatHistory.mjs",
+      "require": "./dist/ChatHistory.js"
+    },
+    "./constants": {
+      "types": "./dist/constants.d.mts",
+      "import": "./dist/constants.mjs",
+      "require": "./dist/constants.js"
+    },
+    "./GlobalsHelper": {
+      "types": "./dist/GlobalsHelper.d.mts",
+      "import": "./dist/GlobalsHelper.mjs",
+      "require": "./dist/GlobalsHelper.js"
+    },
+    "./Node": {
+      "types": "./dist/Node.d.mts",
+      "import": "./dist/Node.mjs",
+      "require": "./dist/Node.js"
+    },
+    "./OutputParser": {
+      "types": "./dist/OutputParser.d.mts",
+      "import": "./dist/OutputParser.mjs",
+      "require": "./dist/OutputParser.js"
+    },
+    "./Prompt": {
+      "types": "./dist/Prompt.d.mts",
+      "import": "./dist/Prompt.mjs",
+      "require": "./dist/Prompt.js"
+    },
+    "./PromptHelper": {
+      "types": "./dist/PromptHelper.d.mts",
+      "import": "./dist/PromptHelper.mjs",
+      "require": "./dist/PromptHelper.js"
+    },
+    "./QueryEngine": {
+      "types": "./dist/QueryEngine.d.mts",
+      "import": "./dist/QueryEngine.mjs",
+      "require": "./dist/QueryEngine.js"
+    },
+    "./QuestionGenerator": {
+      "types": "./dist/QuestionGenerator.d.mts",
+      "import": "./dist/QuestionGenerator.mjs",
+      "require": "./dist/QuestionGenerator.js"
+    },
+    "./Response": {
+      "types": "./dist/Response.d.mts",
+      "import": "./dist/Response.mjs",
+      "require": "./dist/Response.js"
+    },
+    "./Retriever": {
+      "types": "./dist/Retriever.d.mts",
+      "import": "./dist/Retriever.mjs",
+      "require": "./dist/Retriever.js"
+    },
+    "./ServiceContext": {
+      "types": "./dist/ServiceContext.d.mts",
+      "import": "./dist/ServiceContext.mjs",
+      "require": "./dist/ServiceContext.js"
+    },
+    "./TextSplitter": {
+      "types": "./dist/TextSplitter.d.mts",
+      "import": "./dist/TextSplitter.mjs",
+      "require": "./dist/TextSplitter.js"
+    },
+    "./Tool": {
+      "types": "./dist/Tool.d.mts",
+      "import": "./dist/Tool.mjs",
+      "require": "./dist/Tool.js"
+    },
+    "./readers": {
+      "types": "./dist/readers.d.mts",
+      "import": "./dist/readers.mjs",
+      "require": "./dist/readers.js"
+    },
+    "./readers/AssemblyAI": {
+      "types": "./dist/readers/AssemblyAI.d.mts",
+      "import": "./dist/readers/AssemblyAI.mjs",
+      "require": "./dist/readers/AssemblyAI.js"
+    },
+    "./readers/base": {
+      "types": "./dist/readers/base.d.mts",
+      "import": "./dist/readers/base.mjs",
+      "require": "./dist/readers/base.js"
+    },
+    "./readers/PapaCSVReader": {
+      "types": "./dist/readers/PapaCSVReader.d.mts",
+      "import": "./dist/readers/PapaCSVReader.mjs",
+      "require": "./dist/readers/PapaCSVReader.js"
+    },
+    "./readers/DocxReader": {
+      "types": "./dist/readers/DocxReader.d.mts",
+      "import": "./dist/readers/DocxReader.mjs",
+      "require": "./dist/readers/DocxReader.js"
+    },
+    "./readers/HTMLReader": {
+      "types": "./dist/readers/HTMLReader.d.mts",
+      "import": "./dist/readers/HTMLReader.mjs",
+      "require": "./dist/readers/HTMLReader.js"
+    },
+    "./readers/ImageReader": {
+      "types": "./dist/readers/ImageReader.d.mts",
+      "import": "./dist/readers/ImageReader.mjs",
+      "require": "./dist/readers/ImageReader.js"
+    },
+    "./readers/MarkdownReader": {
+      "types": "./dist/readers/MarkdownReader.d.mts",
+      "import": "./dist/readers/MarkdownReader.mjs",
+      "require": "./dist/readers/MarkdownReader.js"
+    },
+    "./readers/NotionReader": {
+      "types": "./dist/readers/NotionReader.d.mts",
+      "import": "./dist/readers/NotionReader.mjs",
+      "require": "./dist/readers/NotionReader.js"
+    },
+    "./readers/PDFReader": {
+      "types": "./dist/readers/PDFReader.d.mts",
+      "import": "./dist/readers/PDFReader.mjs",
+      "require": "./dist/readers/PDFReader.js"
+    },
+    "./readers/SimpleDirectoryReader": {
+      "types": "./dist/readers/SimpleDirectoryReader.d.mts",
+      "import": "./dist/readers/SimpleDirectoryReader.mjs",
+      "require": "./dist/readers/SimpleDirectoryReader.js"
+    },
+    "./readers/SimpleMongoReader": {
+      "types": "./dist/readers/SimpleMongoReader.d.mts",
+      "import": "./dist/readers/SimpleMongoReader.mjs",
+      "require": "./dist/readers/SimpleMongoReader.js"
     }
   },
   "files": [
@@ -66,7 +212,8 @@
   "scripts": {
     "lint": "eslint .",
     "test": "jest",
-    "build": "bunchee",
-    "dev": "bunchee -w"
+    "build": "NODE_OPTIONS=--max_old_space_size=8192 bunchee",
+    "dev": "NODE_OPTIONS=--max_old_space_size=8192 bunchee -w",
+    "circular-check": "madge --circular ./src/*.ts"
   }
 }

packages/core/src/ChatEngine.ts

@@ -1,452 +0,0 @@
-import { randomUUID } from "node:crypto";
-import { ChatHistory } from "./ChatHistory";
-import { NodeWithScore, TextNode } from "./Node";
-import {
-  CondenseQuestionPrompt,
-  ContextSystemPrompt,
-  defaultCondenseQuestionPrompt,
-  defaultContextSystemPrompt,
-  messagesToHistoryStr,
-} from "./Prompt";
-import { BaseQueryEngine } from "./QueryEngine";
-import { Response } from "./Response";
-import { BaseRetriever } from "./Retriever";
-import { ServiceContext, serviceContextFromDefaults } from "./ServiceContext";
-import { Event } from "./callbacks/CallbackManager";
-import { ChatMessage, LLM, OpenAI } from "./llm";
-import { BaseNodePostprocessor } from "./postprocessors";
-
-/**
- * A ChatEngine is used to handle back and forth chats between the application and the LLM.
- */
-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: MessageContent,
-    chatHistory?: ChatMessage[],
-    streaming?: T,
-  ): Promise<R>;
-
-  /**
-   * Resets the chat history so that it's empty.
-   */
-  reset(): void;
-}
-
-/**
- * SimpleChatEngine is the simplest possible chat engine. Useful for using your own custom prompts.
- */
-export class SimpleChatEngine implements ChatEngine {
-  chatHistory: ChatMessage[];
-  llm: LLM;
-
-  constructor(init?: Partial<SimpleChatEngine>) {
-    this.chatHistory = init?.chatHistory ?? [];
-    this.llm = init?.llm ?? new OpenAI();
-  }
-
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
-    message: MessageContent,
-    chatHistory?: ChatMessage[],
-    streaming?: T,
-  ): Promise<R> {
-    //Streaming option
-    if (streaming) {
-      return this.streamChat(message, chatHistory) as R;
-    }
-
-    //Non-streaming option
-    chatHistory = chatHistory ?? this.chatHistory;
-    chatHistory.push({ content: message, role: "user" });
-    const response = await this.llm.chat(chatHistory, undefined);
-    chatHistory.push(response.message);
-    this.chatHistory = chatHistory;
-    return new Response(response.message.content) as R;
-  }
-
-  protected async *streamChat(
-    message: MessageContent,
-    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;
-  }
-
-  reset() {
-    this.chatHistory = [];
-  }
-}
-
-/**
- * CondenseQuestionChatEngine is used in conjunction with a Index (for example VectorStoreIndex).
- * It does two steps on taking a user's chat message: first, it condenses the chat message
- * with the previous chat history into a question with more context.
- * Then, it queries the underlying Index using the new question with context and returns
- * the response.
- * CondenseQuestionChatEngine performs well when the input is primarily questions about the
- * underlying data. It performs less well when the chat messages are not questions about the
- * data, or are very referential to previous context.
- */
-export class CondenseQuestionChatEngine implements ChatEngine {
-  queryEngine: BaseQueryEngine;
-  chatHistory: ChatMessage[];
-  serviceContext: ServiceContext;
-  condenseMessagePrompt: CondenseQuestionPrompt;
-
-  constructor(init: {
-    queryEngine: BaseQueryEngine;
-    chatHistory: ChatMessage[];
-    serviceContext?: ServiceContext;
-    condenseMessagePrompt?: CondenseQuestionPrompt;
-  }) {
-    this.queryEngine = init.queryEngine;
-    this.chatHistory = init?.chatHistory ?? [];
-    this.serviceContext =
-      init?.serviceContext ?? serviceContextFromDefaults({});
-    this.condenseMessagePrompt =
-      init?.condenseMessagePrompt ?? defaultCondenseQuestionPrompt;
-  }
-
-  private async condenseQuestion(chatHistory: ChatMessage[], question: string) {
-    const chatHistoryStr = messagesToHistoryStr(chatHistory);
-
-    return this.serviceContext.llm.complete(
-      defaultCondenseQuestionPrompt({
-        question: question,
-        chatHistory: chatHistoryStr,
-      }),
-    );
-  }
-
-  async chat<
-    T extends boolean | undefined = undefined,
-    R = T extends true ? AsyncGenerator<string, void, unknown> : Response,
-  >(
-    message: MessageContent,
-    chatHistory?: ChatMessage[] | undefined,
-    streaming?: T,
-  ): Promise<R> {
-    chatHistory = chatHistory ?? this.chatHistory;
-
-    const condensedQuestion = (
-      await this.condenseQuestion(chatHistory, extractText(message))
-    ).message.content;
-
-    const response = await this.queryEngine.query(condensedQuestion);
-
-    chatHistory.push({ content: message, role: "user" });
-    chatHistory.push({ content: response.response, role: "assistant" });
-
-    return response as R;
-  }
-
-  reset() {
-    this.chatHistory = [];
-  }
-}
-
-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: randomUUID(),
-        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;
-  chatHistory: ChatMessage[];
-  contextGenerator: ContextGenerator;
-
-  constructor(init: {
-    retriever: BaseRetriever;
-    chatModel?: LLM;
-    chatHistory?: ChatMessage[];
-    contextSystemPrompt?: ContextSystemPrompt;
-    nodePostprocessors?: BaseNodePostprocessor[];
-  }) {
-    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: MessageContent,
-    chatHistory?: ChatMessage[] | undefined,
-    streaming?: T,
-  ): Promise<R> {
-    chatHistory = chatHistory ?? this.chatHistory;
-
-    //Streaming option
-    if (streaming) {
-      return this.streamChat(message, chatHistory) as R;
-    }
-
-    const parentEvent: Event = {
-      id: randomUUID(),
-      type: "wrapper",
-      tags: ["final"],
-    };
-    const context = await this.contextGenerator.generate(
-      extractText(message),
-      parentEvent,
-    );
-
-    chatHistory.push({ content: message, role: "user" });
-
-    const response = await this.chatModel.chat(
-      [context.message, ...chatHistory],
-      parentEvent,
-    );
-    chatHistory.push(response.message);
-
-    this.chatHistory = chatHistory;
-
-    return new Response(
-      response.message.content,
-      context.nodes.map((r) => r.node),
-    ) as R;
-  }
-
-  protected async *streamChat(
-    message: MessageContent,
-    chatHistory?: ChatMessage[] | undefined,
-  ): AsyncGenerator<string, void, unknown> {
-    chatHistory = chatHistory ?? this.chatHistory;
-
-    const parentEvent: Event = {
-      id: randomUUID(),
-      type: "wrapper",
-      tags: ["final"],
-    };
-    const context = await this.contextGenerator.generate(
-      extractText(message),
-      parentEvent,
-    );
-
-    chatHistory.push({ content: message, role: "user" });
-
-    const response_stream = await this.chatModel.chat(
-      [context.message, ...chatHistory],
-      parentEvent,
-      true,
-    );
-    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[];
-
-/**
- * Extracts just the text from a multi-modal message or the message itself if it's just text.
- *
- * @param message The message to extract text from.
- * @returns The extracted text
- */
-function extractText(message: MessageContent): string {
-  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
-    return (message as MessageContentDetail[])
-      .filter((c) => c.type === "text")
-      .map((c) => c.text)
-      .join("\n\n");
-  }
-  return message;
-}
-
-/**
- * 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) {
-      const textOnly = extractText(message);
-      context = await this.contextGenerator.generate(textOnly);
-    }
-    requestMessages = await chatHistory.requestMessages(
-      context ? [context.message] : undefined,
-    );
-    return requestMessages;
-  }
-}

packages/core/src/ChatHistory.ts

@@ -1,4 +1,5 @@
-import { ChatMessage, LLM, MessageType, OpenAI } from "./llm/LLM";
+import { OpenAI } from "./llm/LLM";
+import { ChatMessage, LLM, MessageType } from "./llm/types";
 import {
   defaultSummaryPrompt,
   messagesToHistoryStr,
@@ -8,35 +9,38 @@ import {
 /**
  * A ChatHistory is used to keep the state of back and forth chat messages
  */
-export interface ChatHistory {
-  messages: ChatMessage[];
+export abstract class ChatHistory {
+  abstract get messages(): ChatMessage[];
   /**
    * Adds a message to the chat history.
    * @param message
    */
-  addMessage(message: ChatMessage): void;
+  abstract addMessage(message: ChatMessage): void;
 
   /**
    * Returns the messages that should be used as input to the LLM.
    */
-  requestMessages(transientMessages?: ChatMessage[]): Promise<ChatMessage[]>;
+  abstract requestMessages(
+    transientMessages?: ChatMessage[],
+  ): Promise<ChatMessage[]>;
 
   /**
    * Resets the chat history so that it's empty.
    */
-  reset(): void;
+  abstract reset(): void;
 
   /**
    * Returns the new messages since the last call to this function (or since calling the constructor)
    */
-  newMessages(): ChatMessage[];
+  abstract newMessages(): ChatMessage[];
 }
 
-export class SimpleChatHistory implements ChatHistory {
+export class SimpleChatHistory extends ChatHistory {
   messages: ChatMessage[];
   private messagesBefore: number;
 
   constructor(init?: Partial<SimpleChatHistory>) {
+    super();
     this.messages = init?.messages ?? [];
     this.messagesBefore = this.messages.length;
   }
@@ -60,7 +64,7 @@ export class SimpleChatHistory implements ChatHistory {
   }
 }
 
-export class SummaryChatHistory implements ChatHistory {
+export class SummaryChatHistory extends ChatHistory {
   tokensToSummarize: number;
   messages: ChatMessage[];
   summaryPrompt: SummaryPrompt;
@@ -68,6 +72,7 @@ export class SummaryChatHistory implements ChatHistory {
   private messagesBefore: number;
 
   constructor(init?: Partial<SummaryChatHistory>) {
+    super();
     this.messages = init?.messages ?? [];
     this.messagesBefore = this.messages.length;
     this.summaryPrompt = init?.summaryPrompt ?? defaultSummaryPrompt;
@@ -79,6 +84,11 @@ export class SummaryChatHistory implements ChatHistory {
     }
     this.tokensToSummarize =
       this.llm.metadata.contextWindow - this.llm.metadata.maxTokens;
+    if (this.tokensToSummarize < this.llm.metadata.contextWindow * 0.25) {
+      throw new Error(
+        "The number of tokens that trigger the summarize process are less than 25% of the context window. Try lowering maxTokens or use a model with a larger context window.",
+      );
+    }
   }
 
   private async summarize(): Promise<ChatMessage> {
@@ -99,7 +109,7 @@ export class SummaryChatHistory implements ChatHistory {
       messagesToSummarize.shift();
     } while (this.llm.tokens(promptMessages) > this.tokensToSummarize);
 
-    const response = await this.llm.chat(promptMessages);
+    const response = await this.llm.chat({ messages: promptMessages });
     return { content: response.message.content, role: "memory" };
   }
 
@@ -119,6 +129,11 @@ export class SummaryChatHistory implements ChatHistory {
     return this.messages.length - 1 - index;
   }
 
+  public getLastSummary(): ChatMessage | null {
+    const lastSummaryIndex = this.getLastSummaryIndex();
+    return lastSummaryIndex ? this.messages[lastSummaryIndex] : null;
+  }
+
   private get systemMessages() {
     // get array of all system messages
     return this.messages.filter((message) => message.role === "system");
@@ -198,3 +213,12 @@ export class SummaryChatHistory implements ChatHistory {
     return newMessages;
   }
 }
+
+export function getHistory(
+  chatHistory?: ChatMessage[] | ChatHistory,
+): ChatHistory {
+  if (chatHistory instanceof ChatHistory) {
+    return chatHistory;
+  }
+  return new SimpleChatHistory({ messages: chatHistory });
+}

packages/core/src/GlobalsHelper.ts

@@ -1,7 +1,7 @@
 import { encodingForModel } from "js-tiktoken";
 
-import { randomUUID } from "node:crypto";
 import { Event, EventTag, EventType } from "./callbacks/CallbackManager";
+import { randomUUID } from "./env";
 
 export enum Tokenizers {
   CL100K_BASE = "cl100k_base",

packages/core/src/Node.ts

@@ -1,6 +1,6 @@
 import _ from "lodash";
-import { createHash, randomUUID } from "node:crypto";
 import path from "node:path";
+import { createSHA256, randomUUID } from "./env";
 
 export enum NodeRelationship {
   SOURCE = "SOURCE",
@@ -168,6 +168,8 @@ export abstract class BaseNode<T extends Metadata = Metadata> {
  */
 export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
   text: string = "";
+  textTemplate: string = "";
+
   startCharIdx?: number;
   endCharIdx?: number;
   // textTemplate: NOTE write your own formatter if needed
@@ -181,7 +183,7 @@ export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
     if (new.target === TextNode) {
       // Don't generate the hash repeatedly so only do it if this is
       // constructing the derived class
-      this.hash = this.generateHash();
+      this.hash = init?.hash ?? this.generateHash();
     }
   }
 
@@ -191,13 +193,13 @@ export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
    * @returns
    */
   generateHash() {
-    const hashFunction = createHash("sha256");
+    const hashFunction = createSHA256();
     hashFunction.update(`type=${this.getType()}`);
     hashFunction.update(
       `startCharIdx=${this.startCharIdx} endCharIdx=${this.endCharIdx}`,
     );
     hashFunction.update(this.getContent(MetadataMode.ALL));
-    return hashFunction.digest("base64");
+    return hashFunction.digest();
   }
 
   getType(): ObjectType {
@@ -232,7 +234,6 @@ export class TextNode<T extends Metadata = Metadata> extends BaseNode<T> {
 
   setContent(value: string) {
     this.text = value;
-
     this.hash = this.generateHash();
   }
 
@@ -253,7 +254,7 @@ export class IndexNode<T extends Metadata = Metadata> extends TextNode<T> {
     Object.assign(this, init);
 
     if (new.target === IndexNode) {
-      this.hash = this.generateHash();
+      this.hash = init?.hash ?? this.generateHash();
     }
   }
 
@@ -271,7 +272,7 @@ export class Document<T extends Metadata = Metadata> extends TextNode<T> {
     Object.assign(this, init);
 
     if (new.target === Document) {
-      this.hash = this.generateHash();
+      this.hash = init?.hash ?? this.generateHash();
     }
   }
 
@@ -332,7 +333,7 @@ export class ImageDocument<T extends Metadata = Metadata> extends ImageNode<T> {
     super(init);
 
     if (new.target === ImageDocument) {
-      this.hash = this.generateHash();
+      this.hash = init?.hash ?? this.generateHash();
     }
   }
 

packages/core/src/Prompt.ts

@@ -1,4 +1,4 @@
-import { ChatMessage } from "./llm/LLM";
+import { ChatMessage } from "./llm/types";
 import { SubQuestion } from "./QuestionGenerator";
 import { ToolMetadata } from "./Tool";
 

packages/core/src/PromptHelper.ts

@@ -37,6 +37,7 @@ export class PromptHelper {
   tokenizer: (text: string) => Uint32Array;
   separator = " ";
 
+  // eslint-disable-next-line max-params
   constructor(
     contextWindow = DEFAULT_CONTEXT_WINDOW,
     numOutput = DEFAULT_NUM_OUTPUTS,

packages/core/src/QueryEngine.ts

@@ -1,4 +1,3 @@
-import { randomUUID } from "node:crypto";
 import { NodeWithScore, TextNode } from "./Node";
 import {
   BaseQuestionGenerator,
@@ -10,6 +9,7 @@ import { BaseRetriever } from "./Retriever";
 import { ServiceContext, serviceContextFromDefaults } from "./ServiceContext";
 import { QueryEngineTool, ToolMetadata } from "./Tool";
 import { Event } from "./callbacks/CallbackManager";
+import { randomUUID } from "./env";
 import { BaseNodePostprocessor } from "./postprocessors";
 import {
   BaseSynthesizer,
@@ -17,16 +17,32 @@ import {
   ResponseSynthesizer,
 } from "./synthesizers";
 
+/**
+ * Parameters for sending a query.
+ */
+export interface QueryEngineParamsBase {
+  query: string;
+  parentEvent?: Event;
+}
+
+export interface QueryEngineParamsStreaming extends QueryEngineParamsBase {
+  stream: true;
+}
+
+export interface QueryEngineParamsNonStreaming extends QueryEngineParamsBase {
+  stream?: false | null;
+}
+
 /**
  * A query engine is a question answerer that can use one or more steps.
  */
 export interface BaseQueryEngine {
   /**
    * Query the query engine and get a response.
-   * @param query
-   * @param parentEvent
+   * @param params
    */
-  query(query: string, parentEvent?: Event): Promise<Response>;
+  query(params: QueryEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  query(params: QueryEngineParamsNonStreaming): Promise<Response>;
 }
 
 /**
@@ -70,14 +86,31 @@ export class RetrieverQueryEngine implements BaseQueryEngine {
     return this.applyNodePostprocessors(nodes);
   }
 
-  async query(query: string, parentEvent?: Event) {
-    const _parentEvent: Event = parentEvent || {
+  query(params: QueryEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  query(params: QueryEngineParamsNonStreaming): Promise<Response>;
+  async query(
+    params: QueryEngineParamsStreaming | QueryEngineParamsNonStreaming,
+  ): Promise<Response | AsyncIterable<Response>> {
+    const { query, stream } = params;
+    const parentEvent: Event = params.parentEvent || {
       id: randomUUID(),
       type: "wrapper",
       tags: ["final"],
     };
-    const nodes = await this.retrieve(query, _parentEvent);
-    return this.responseSynthesizer.synthesize(query, nodes, _parentEvent);
+    const nodesWithScore = await this.retrieve(query, parentEvent);
+    if (stream) {
+      return this.responseSynthesizer.synthesize({
+        query,
+        nodesWithScore,
+        parentEvent,
+        stream: true,
+      });
+    }
+    return this.responseSynthesizer.synthesize({
+      query,
+      nodesWithScore,
+      parentEvent,
+    });
   }
 }
 
@@ -131,11 +164,16 @@ export class SubQuestionQueryEngine implements BaseQueryEngine {
     });
   }
 
-  async query(query: string): Promise<Response> {
+  query(params: QueryEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  query(params: QueryEngineParamsNonStreaming): Promise<Response>;
+  async query(
+    params: QueryEngineParamsStreaming | QueryEngineParamsNonStreaming,
+  ): Promise<Response | AsyncIterable<Response>> {
+    const { query, stream } = params;
     const subQuestions = await this.questionGen.generate(this.metadatas, query);
 
     // groups final retrieval+synthesis operation
-    const parentEvent: Event = {
+    const parentEvent: Event = params.parentEvent || {
       id: randomUUID(),
       type: "wrapper",
       tags: ["final"],
@@ -153,10 +191,22 @@ export class SubQuestionQueryEngine implements BaseQueryEngine {
       subQuestions.map((subQ) => this.querySubQ(subQ, subQueryParentEvent)),
     );
 
-    const nodes = subQNodes
+    const nodesWithScore = subQNodes
       .filter((node) => node !== null)
       .map((node) => node as NodeWithScore);
-    return this.responseSynthesizer.synthesize(query, nodes, parentEvent);
+    if (stream) {
+      return this.responseSynthesizer.synthesize({
+        query,
+        nodesWithScore,
+        parentEvent,
+        stream: true,
+      });
+    }
+    return this.responseSynthesizer.synthesize({
+      query,
+      nodesWithScore,
+      parentEvent,
+    });
   }
 
   private async querySubQ(
@@ -167,7 +217,10 @@ export class SubQuestionQueryEngine implements BaseQueryEngine {
       const question = subQ.subQuestion;
       const queryEngine = this.queryEngines[subQ.toolName];
 
-      const response = await queryEngine.query(question, parentEvent);
+      const response = await queryEngine.query({
+        query: question,
+        parentEvent,
+      });
       const responseText = response.response;
       const nodeText = `Sub question: ${question}\nResponse: ${responseText}`;
       const node = new TextNode({ text: nodeText });

packages/core/src/QuestionGenerator.ts

@@ -9,7 +9,8 @@ import {
   defaultSubQuestionPrompt,
 } from "./Prompt";
 import { ToolMetadata } from "./Tool";
-import { LLM, OpenAI } from "./llm/LLM";
+import { OpenAI } from "./llm/LLM";
+import { LLM } from "./llm/types";
 
 export interface SubQuestion {
   subQuestion: string;
@@ -41,13 +42,13 @@ export class LLMQuestionGenerator implements BaseQuestionGenerator {
     const toolsStr = buildToolsText(tools);
     const queryStr = query;
     const prediction = (
-      await this.llm.complete(
-        this.prompt({
+      await this.llm.complete({
+        prompt: this.prompt({
           toolsStr,
           queryStr,
         }),
-      )
-    ).message.content;
+      })
+    ).text;
 
     const structuredOutput = this.outputParser.parse(prediction);
 

packages/core/src/Response.ts

@@ -1,7 +1,7 @@
 import { BaseNode } from "./Node";
 
 /**
- * Respone is the output of a LLM
+ * Response is the output of a LLM
  */
 export class Response {
   response: string;

packages/core/src/TextSplitter.ts

@@ -1,4 +1,4 @@
-import { EOL } from "node:os";
+import { EOL } from "./env";
 // GitHub translated
 import { globalsHelper } from "./GlobalsHelper";
 import { DEFAULT_CHUNK_OVERLAP, DEFAULT_CHUNK_SIZE } from "./constants";

packages/core/src/embeddings/types.ts

@@ -1,3 +1,5 @@
+import { BaseNode, MetadataMode } from "../Node";
+import { TransformComponent } from "../ingestion";
 import { similarity } from "./utils";
 
 /**
@@ -10,7 +12,7 @@ export enum SimilarityType {
   EUCLIDEAN = "euclidean",
 }
 
-export abstract class BaseEmbedding {
+export abstract class BaseEmbedding implements TransformComponent {
   similarity(
     embedding1: number[],
     embedding2: number[],
@@ -21,4 +23,13 @@ export abstract class BaseEmbedding {
 
   abstract getTextEmbedding(text: string): Promise<number[]>;
   abstract getQueryEmbedding(query: string): Promise<number[]>;
+
+  async transform(nodes: BaseNode[], _options?: any): Promise<BaseNode[]> {
+    for (const node of nodes) {
+      node.embedding = await this.getTextEmbedding(
+        node.getContent(MetadataMode.EMBED),
+      );
+    }
+    return nodes;
+  }
 }

packages/core/src/embeddings/utils.ts

@@ -1,7 +1,8 @@
 import _ from "lodash";
 import { ImageType } from "../Node";
 import { DEFAULT_SIMILARITY_TOP_K } from "../constants";
-import { DEFAULT_FS, VectorStoreQueryMode } from "../storage";
+import { defaultFS } from "../env";
+import { VectorStoreQueryMode } from "../storage";
 import { SimilarityType } from "./types";
 
 /**
@@ -66,6 +67,7 @@ export function similarity(
  * @param similarityCutoff minimum similarity score
  * @returns
  */
+// eslint-disable-next-line max-params
 export function getTopKEmbeddings(
   queryEmbedding: number[],
   embeddings: number[][],
@@ -108,6 +110,7 @@ export function getTopKEmbeddings(
   return [resultSimilarities, resultIds];
 }
 
+// eslint-disable-next-line max-params
 export function getTopKEmbeddingsLearner(
   queryEmbedding: number[],
   embeddings: number[][],
@@ -120,6 +123,7 @@ export function getTopKEmbeddingsLearner(
   // https://github.com/mljs/libsvm which itself hasn't been updated in a while
 }
 
+// eslint-disable-next-line max-params
 export function getTopKMMREmbeddings(
   queryEmbedding: number[],
   embeddings: number[][],
@@ -241,8 +245,7 @@ export async function imageToDataUrl(input: ImageType): Promise<string> {
     _.isString(input)
   ) {
     // string or file URL
-    const fs = DEFAULT_FS;
-    const dataBuffer = await fs.readFile(
+    const dataBuffer = await defaultFS.readFile(
       input instanceof URL ? input.pathname : input,
     );
     input = new Blob([dataBuffer]);

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

@@ -0,0 +1,104 @@
+import { ChatHistory, getHistory } from "../../ChatHistory";
+import {
+  CondenseQuestionPrompt,
+  defaultCondenseQuestionPrompt,
+  messagesToHistoryStr,
+} from "../../Prompt";
+import { BaseQueryEngine } from "../../QueryEngine";
+import { Response } from "../../Response";
+import {
+  ServiceContext,
+  serviceContextFromDefaults,
+} from "../../ServiceContext";
+import { ChatMessage, LLM } from "../../llm";
+import { extractText, streamReducer } from "../../llm/utils";
+import {
+  ChatEngine,
+  ChatEngineParamsNonStreaming,
+  ChatEngineParamsStreaming,
+} from "./types";
+
+/**
+ * CondenseQuestionChatEngine is used in conjunction with a Index (for example VectorStoreIndex).
+ * It does two steps on taking a user's chat message: first, it condenses the chat message
+ * with the previous chat history into a question with more context.
+ * Then, it queries the underlying Index using the new question with context and returns
+ * the response.
+ * CondenseQuestionChatEngine performs well when the input is primarily questions about the
+ * underlying data. It performs less well when the chat messages are not questions about the
+ * data, or are very referential to previous context.
+ */
+
+export class CondenseQuestionChatEngine implements ChatEngine {
+  queryEngine: BaseQueryEngine;
+  chatHistory: ChatHistory;
+  llm: LLM;
+  condenseMessagePrompt: CondenseQuestionPrompt;
+
+  constructor(init: {
+    queryEngine: BaseQueryEngine;
+    chatHistory: ChatMessage[];
+    serviceContext?: ServiceContext;
+    condenseMessagePrompt?: CondenseQuestionPrompt;
+  }) {
+    this.queryEngine = init.queryEngine;
+    this.chatHistory = getHistory(init?.chatHistory);
+    this.llm = init?.serviceContext?.llm ?? serviceContextFromDefaults().llm;
+    this.condenseMessagePrompt =
+      init?.condenseMessagePrompt ?? defaultCondenseQuestionPrompt;
+  }
+
+  private async condenseQuestion(chatHistory: ChatHistory, question: string) {
+    const chatHistoryStr = messagesToHistoryStr(
+      await chatHistory.requestMessages(),
+    );
+
+    return this.llm.complete({
+      prompt: defaultCondenseQuestionPrompt({
+        question: question,
+        chatHistory: chatHistoryStr,
+      }),
+    });
+  }
+
+  chat(params: ChatEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  chat(params: ChatEngineParamsNonStreaming): Promise<Response>;
+  async chat(
+    params: ChatEngineParamsStreaming | ChatEngineParamsNonStreaming,
+  ): Promise<Response | AsyncIterable<Response>> {
+    const { message, stream } = params;
+    const chatHistory = params.chatHistory
+      ? getHistory(params.chatHistory)
+      : this.chatHistory;
+
+    const condensedQuestion = (
+      await this.condenseQuestion(chatHistory, extractText(message))
+    ).text;
+    chatHistory.addMessage({ content: message, role: "user" });
+
+    if (stream) {
+      const stream = await this.queryEngine.query({
+        query: condensedQuestion,
+        stream: true,
+      });
+      return streamReducer({
+        stream,
+        initialValue: "",
+        reducer: (accumulator, part) => (accumulator += part.response),
+        finished: (accumulator) => {
+          chatHistory.addMessage({ content: accumulator, role: "assistant" });
+        },
+      });
+    }
+    const response = await this.queryEngine.query({
+      query: condensedQuestion,
+    });
+    chatHistory.addMessage({ content: response.response, role: "assistant" });
+
+    return response;
+  }
+
+  reset() {
+    this.chatHistory.reset();
+  }
+}

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

@@ -0,0 +1,113 @@
+import { ChatHistory, getHistory } from "../../ChatHistory";
+import { ContextSystemPrompt } from "../../Prompt";
+import { Response } from "../../Response";
+import { BaseRetriever } from "../../Retriever";
+import { Event } from "../../callbacks/CallbackManager";
+import { randomUUID } from "../../env";
+import { ChatMessage, ChatResponseChunk, LLM, OpenAI } from "../../llm";
+import { MessageContent } from "../../llm/types";
+import { extractText, streamConverter, streamReducer } from "../../llm/utils";
+import { BaseNodePostprocessor } from "../../postprocessors";
+import { DefaultContextGenerator } from "./DefaultContextGenerator";
+import {
+  ChatEngine,
+  ChatEngineParamsNonStreaming,
+  ChatEngineParamsStreaming,
+  ContextGenerator,
+} from "./types";
+
+/**
+ * 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;
+  chatHistory: ChatHistory;
+  contextGenerator: ContextGenerator;
+
+  constructor(init: {
+    retriever: BaseRetriever;
+    chatModel?: LLM;
+    chatHistory?: ChatMessage[];
+    contextSystemPrompt?: ContextSystemPrompt;
+    nodePostprocessors?: BaseNodePostprocessor[];
+  }) {
+    this.chatModel =
+      init.chatModel ?? new OpenAI({ model: "gpt-3.5-turbo-16k" });
+    this.chatHistory = getHistory(init?.chatHistory);
+    this.contextGenerator = new DefaultContextGenerator({
+      retriever: init.retriever,
+      contextSystemPrompt: init?.contextSystemPrompt,
+      nodePostprocessors: init?.nodePostprocessors,
+    });
+  }
+
+  chat(params: ChatEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  chat(params: ChatEngineParamsNonStreaming): Promise<Response>;
+  async chat(
+    params: ChatEngineParamsStreaming | ChatEngineParamsNonStreaming,
+  ): Promise<Response | AsyncIterable<Response>> {
+    const { message, stream } = params;
+    const chatHistory = params.chatHistory
+      ? getHistory(params.chatHistory)
+      : this.chatHistory;
+    const parentEvent: Event = {
+      id: randomUUID(),
+      type: "wrapper",
+      tags: ["final"],
+    };
+    const requestMessages = await this.prepareRequestMessages(
+      message,
+      chatHistory,
+      parentEvent,
+    );
+
+    if (stream) {
+      const stream = await this.chatModel.chat({
+        messages: requestMessages.messages,
+        parentEvent,
+        stream: true,
+      });
+      return streamConverter(
+        streamReducer({
+          stream,
+          initialValue: "",
+          reducer: (accumulator, part) => (accumulator += part.delta),
+          finished: (accumulator) => {
+            chatHistory.addMessage({ content: accumulator, role: "assistant" });
+          },
+        }),
+        (r: ChatResponseChunk) => new Response(r.delta, requestMessages.nodes),
+      );
+    }
+    const response = await this.chatModel.chat({
+      messages: requestMessages.messages,
+      parentEvent,
+    });
+    chatHistory.addMessage(response.message);
+    return new Response(response.message.content, requestMessages.nodes);
+  }
+
+  reset() {
+    this.chatHistory.reset();
+  }
+
+  private async prepareRequestMessages(
+    message: MessageContent,
+    chatHistory: ChatHistory,
+    parentEvent?: Event,
+  ) {
+    chatHistory.addMessage({
+      content: message,
+      role: "user",
+    });
+    const textOnly = extractText(message);
+    const context = await this.contextGenerator.generate(textOnly, parentEvent);
+    const nodes = context.nodes.map((r) => r.node);
+    const messages = await chatHistory.requestMessages(
+      context ? [context.message] : undefined,
+    );
+    return { nodes, messages };
+  }
+}

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

@@ -0,0 +1,57 @@
+import { NodeWithScore, TextNode } from "../../Node";
+import { ContextSystemPrompt, defaultContextSystemPrompt } from "../../Prompt";
+import { BaseRetriever } from "../../Retriever";
+import { Event } from "../../callbacks/CallbackManager";
+import { randomUUID } from "../../env";
+import { BaseNodePostprocessor } from "../../postprocessors";
+import { Context, ContextGenerator } from "./types";
+
+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: randomUUID(),
+        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,
+    };
+  }
+}

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

@@ -0,0 +1,64 @@
+import { ChatHistory, getHistory } from "../../ChatHistory";
+import { Response } from "../../Response";
+import { ChatResponseChunk, LLM, OpenAI } from "../../llm";
+import { streamConverter, streamReducer } from "../../llm/utils";
+import {
+  ChatEngine,
+  ChatEngineParamsNonStreaming,
+  ChatEngineParamsStreaming,
+} from "./types";
+
+/**
+ * SimpleChatEngine is the simplest possible chat engine. Useful for using your own custom prompts.
+ */
+
+export class SimpleChatEngine implements ChatEngine {
+  chatHistory: ChatHistory;
+  llm: LLM;
+
+  constructor(init?: Partial<SimpleChatEngine>) {
+    this.chatHistory = getHistory(init?.chatHistory);
+    this.llm = init?.llm ?? new OpenAI();
+  }
+
+  chat(params: ChatEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  chat(params: ChatEngineParamsNonStreaming): Promise<Response>;
+  async chat(
+    params: ChatEngineParamsStreaming | ChatEngineParamsNonStreaming,
+  ): Promise<Response | AsyncIterable<Response>> {
+    const { message, stream } = params;
+
+    const chatHistory = params.chatHistory
+      ? getHistory(params.chatHistory)
+      : this.chatHistory;
+    chatHistory.addMessage({ content: message, role: "user" });
+
+    if (stream) {
+      const stream = await this.llm.chat({
+        messages: await chatHistory.requestMessages(),
+        stream: true,
+      });
+      return streamConverter(
+        streamReducer({
+          stream,
+          initialValue: "",
+          reducer: (accumulator, part) => (accumulator += part.delta),
+          finished: (accumulator) => {
+            chatHistory.addMessage({ content: accumulator, role: "assistant" });
+          },
+        }),
+        (r: ChatResponseChunk) => new Response(r.delta),
+      );
+    }
+
+    const response = await this.llm.chat({
+      messages: await chatHistory.requestMessages(),
+    });
+    chatHistory.addMessage(response.message);
+    return new Response(response.message.content);
+  }
+
+  reset() {
+    this.chatHistory.reset();
+  }
+}

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

@@ -0,0 +1,4 @@
+export { CondenseQuestionChatEngine } from "./CondenseQuestionChatEngine";
+export { ContextChatEngine } from "./ContextChatEngine";
+export { SimpleChatEngine } from "./SimpleChatEngine";
+export * from "./types";

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

@@ -0,0 +1,54 @@
+import { ChatHistory } from "../../ChatHistory";
+import { NodeWithScore } from "../../Node";
+import { Response } from "../../Response";
+import { Event } from "../../callbacks/CallbackManager";
+import { ChatMessage } from "../../llm";
+import { MessageContent } from "../../llm/types";
+
+/**
+ * Represents the base parameters for ChatEngine.
+ */
+export interface ChatEngineParamsBase {
+  message: MessageContent;
+  /**
+   * Optional chat history if you want to customize the chat history.
+   */
+  chatHistory?: ChatMessage[] | ChatHistory;
+}
+
+export interface ChatEngineParamsStreaming extends ChatEngineParamsBase {
+  stream: true;
+}
+
+export interface ChatEngineParamsNonStreaming extends ChatEngineParamsBase {
+  stream?: false | null;
+}
+
+/**
+ * A ChatEngine is used to handle back and forth chats between the application and the LLM.
+ */
+export interface ChatEngine {
+  /**
+   * Send message along with the class's current chat history to the LLM.
+   * @param params
+   */
+  chat(params: ChatEngineParamsStreaming): Promise<AsyncIterable<Response>>;
+  chat(params: ChatEngineParamsNonStreaming): Promise<Response>;
+
+  /**
+   * Resets the chat history so that it's empty.
+   */
+  reset(): void;
+}
+
+export interface Context {
+  message: ChatMessage;
+  nodes: NodeWithScore[];
+}
+
+/**
+ * A ContextGenerator is used to generate a context based on a message's text content
+ */
+export interface ContextGenerator {
+  generate(message: string, parentEvent?: Event): Promise<Context>;
+}

packages/core/src/env/index.edge-light.ts

@@ -0,0 +1,37 @@
+import { Sha256 } from "@aws-crypto/sha256-js";
+import { CompleteFileSystem, InMemoryFileSystem } from "../storage";
+
+export interface SHA256 {
+  update(data: string | Uint8Array): void;
+  // to base64
+  digest(): string;
+}
+
+export const EOL = "\n";
+
+export const defaultFS: CompleteFileSystem = new InMemoryFileSystem();
+
+export function ok(value: unknown, message?: string): asserts value {
+  if (!value) {
+    const error = Error(message);
+    error.name = "AssertionError";
+    error.message = message ?? "The expression evaluated to a falsy value.";
+    throw error;
+  }
+}
+
+export function createSHA256(): SHA256 {
+  const sha256 = new Sha256();
+  return {
+    update(data: string | Uint8Array): void {
+      sha256.update(data);
+    },
+    digest() {
+      return globalThis.btoa(sha256.digestSync().toString());
+    },
+  };
+}
+
+export function randomUUID(): string {
+  return crypto.randomUUID();
+}

packages/core/src/env/index.ts

@@ -0,0 +1,22 @@
+import { ok } from "node:assert";
+import { createHash, randomUUID } from "node:crypto";
+import fs from "node:fs/promises";
+import { EOL } from "node:os";
+import type { CompleteFileSystem } from "../storage";
+import type { SHA256 } from "./index.edge-light";
+
+export function createSHA256(): SHA256 {
+  const hash = createHash("sha256");
+  return {
+    update(data: string | Uint8Array): void {
+      hash.update(data);
+    },
+    digest() {
+      return hash.digest("base64");
+    },
+  };
+}
+
+export const defaultFS: CompleteFileSystem = fs;
+
+export { EOL, ok, randomUUID };

packages/core/src/extractors/MetadataExtractors.ts

@@ -0,0 +1,410 @@
+import { BaseNode, MetadataMode, TextNode } from "../Node";
+import { LLM } from "../llm";
+import {
+  defaultKeywordExtractorPromptTemplate,
+  defaultQuestionAnswerPromptTemplate,
+  defaultSummaryExtractorPromptTemplate,
+  defaultTitleCombinePromptTemplate,
+  defaultTitleExtractorPromptTemplate,
+} from "./prompts";
+import { BaseExtractor } from "./types";
+
+const STRIP_REGEX = /(\r\n|\n|\r)/gm;
+
+type ExtractKeyword = {
+  excerptKeywords: string;
+};
+
+/**
+ * Extract keywords from a list of nodes.
+ */
+export class KeywordExtractor extends BaseExtractor {
+  /**
+   * LLM instance.
+   * @type {LLM}
+   */
+  llm: LLM;
+
+  /**
+   * Number of keywords to extract.
+   * @type {number}
+   * @default 5
+   */
+  keywords: number = 5;
+
+  /**
+   * Constructor for the KeywordExtractor class.
+   * @param {LLM} llm LLM instance.
+   * @param {number} keywords Number of keywords to extract.
+   * @throws {Error} If keywords is less than 1.
+   */
+  constructor(llm: LLM, keywords: number = 5) {
+    if (keywords < 1) throw new Error("Keywords must be greater than 0");
+
+    super();
+    this.llm = llm;
+    this.keywords = keywords;
+  }
+
+  /**
+   *
+   * @param node Node to extract keywords from.
+   * @returns Keywords extracted from the node.
+   */
+  async extractKeywordsFromNodes(node: BaseNode): Promise<ExtractKeyword | {}> {
+    if (this.isTextNodeOnly && !(node instanceof TextNode)) {
+      return {};
+    }
+
+    const completion = await this.llm.complete({
+      prompt: defaultKeywordExtractorPromptTemplate({
+        contextStr: node.getContent(MetadataMode.ALL),
+        keywords: this.keywords,
+      }),
+    });
+
+    return {
+      excerptKeywords: completion.text,
+    };
+  }
+
+  /**
+   *
+   * @param nodes Nodes to extract keywords from.
+   * @returns Keywords extracted from the nodes.
+   */
+  async extract(nodes: BaseNode[]): Promise<Array<ExtractKeyword> | Array<{}>> {
+    const results = await Promise.all(
+      nodes.map((node) => this.extractKeywordsFromNodes(node)),
+    );
+    return results;
+  }
+}
+
+type ExtractTitle = {
+  documentTitle: string;
+};
+
+/**
+ * Extract title from a list of nodes.
+ */
+export class TitleExtractor extends BaseExtractor {
+  /**
+   * LLM instance.
+   * @type {LLM}
+   */
+  llm: LLM;
+
+  /**
+   * Can work for mixture of text and non-text nodes
+   * @type {boolean}
+   * @default false
+   */
+  isTextNodeOnly: boolean = false;
+
+  /**
+   * Number of nodes to extrct titles from.
+   * @type {number}
+   * @default 5
+   */
+  nodes: number = 5;
+
+  /**
+   * The prompt template to use for the title extractor.
+   * @type {string}
+   */
+  nodeTemplate: string;
+
+  /**
+   * The prompt template to merge title with..
+   * @type {string}
+   */
+  combineTemplate: string;
+
+  /**
+   * Constructor for the TitleExtractor class.
+   * @param {LLM} llm LLM instance.
+   * @param {number} nodes Number of nodes to extract titles from.
+   * @param {string} node_template The prompt template to use for the title extractor.
+   * @param {string} combine_template The prompt template to merge title with..
+   */
+  constructor(
+    llm: LLM,
+    nodes: number = 5,
+    node_template?: string,
+    combine_template?: string,
+  ) {
+    super();
+
+    this.llm = llm;
+    this.nodes = nodes;
+
+    this.nodeTemplate = node_template ?? defaultTitleExtractorPromptTemplate();
+    this.combineTemplate =
+      combine_template ?? defaultTitleCombinePromptTemplate();
+  }
+
+  /**
+   * Extract titles from a list of nodes.
+   * @param {BaseNode[]} nodes Nodes to extract titles from.
+   * @returns {Promise<BaseNode<ExtractTitle>[]>} Titles extracted from the nodes.
+   */
+  async extract(nodes: BaseNode[]): Promise<Array<ExtractTitle>> {
+    const nodesToExtractTitle: BaseNode[] = [];
+
+    for (let i = 0; i < this.nodes; i++) {
+      if (nodesToExtractTitle.length >= nodes.length) break;
+
+      if (this.isTextNodeOnly && !(nodes[i] instanceof TextNode)) continue;
+
+      nodesToExtractTitle.push(nodes[i]);
+    }
+
+    if (nodesToExtractTitle.length === 0) return [];
+
+    let titlesCandidates: string[] = [];
+    let title: string = "";
+
+    for (let i = 0; i < nodesToExtractTitle.length; i++) {
+      const completion = await this.llm.complete({
+        prompt: defaultTitleExtractorPromptTemplate({
+          contextStr: nodesToExtractTitle[i].getContent(MetadataMode.ALL),
+        }),
+      });
+
+      titlesCandidates.push(completion.text);
+    }
+
+    if (nodesToExtractTitle.length > 1) {
+      const combinedTitles = titlesCandidates.join(",");
+
+      const completion = await this.llm.complete({
+        prompt: defaultTitleCombinePromptTemplate({
+          contextStr: combinedTitles,
+        }),
+      });
+
+      title = completion.text;
+    }
+
+    if (nodesToExtractTitle.length === 1) {
+      title = titlesCandidates[0];
+    }
+
+    return nodes.map((_) => ({
+      documentTitle: title.trim().replace(STRIP_REGEX, ""),
+    }));
+  }
+}
+
+type ExtractQuestion = {
+  questionsThisExcerptCanAnswer: string;
+};
+
+/**
+ * Extract questions from a list of nodes.
+ */
+export class QuestionsAnsweredExtractor extends BaseExtractor {
+  /**
+   * LLM instance.
+   * @type {LLM}
+   */
+  llm: LLM;
+
+  /**
+   * Number of questions to generate.
+   * @type {number}
+   * @default 5
+   */
+  questions: number = 5;
+
+  /**
+   * The prompt template to use for the question extractor.
+   * @type {string}
+   */
+  promptTemplate: string;
+
+  /**
+   * Wheter to use metadata for embeddings only
+   * @type {boolean}
+   * @default false
+   */
+  embeddingOnly: boolean = false;
+
+  /**
+   * Constructor for the QuestionsAnsweredExtractor class.
+   * @param {LLM} llm LLM instance.
+   * @param {number} questions Number of questions to generate.
+   * @param {string} promptTemplate The prompt template to use for the question extractor.
+   * @param {boolean} embeddingOnly Wheter to use metadata for embeddings only.
+   */
+  constructor(
+    llm: LLM,
+    questions: number = 5,
+    promptTemplate?: string,
+    embeddingOnly: boolean = false,
+  ) {
+    if (questions < 1) throw new Error("Questions must be greater than 0");
+
+    super();
+
+    this.llm = llm;
+    this.questions = questions;
+    this.promptTemplate =
+      promptTemplate ??
+      defaultQuestionAnswerPromptTemplate({
+        numQuestions: questions,
+        contextStr: "",
+      });
+    this.embeddingOnly = embeddingOnly;
+  }
+
+  /**
+   * Extract answered questions from a node.
+   * @param {BaseNode} node Node to extract questions from.
+   * @returns {Promise<Array<ExtractQuestion> | Array<{}>>} Questions extracted from the node.
+   */
+  async extractQuestionsFromNode(
+    node: BaseNode,
+  ): Promise<ExtractQuestion | {}> {
+    if (this.isTextNodeOnly && !(node instanceof TextNode)) {
+      return {};
+    }
+
+    const contextStr = node.getContent(this.metadataMode);
+
+    const prompt = defaultQuestionAnswerPromptTemplate({
+      contextStr,
+      numQuestions: this.questions,
+    });
+
+    const questions = await this.llm.complete({
+      prompt,
+    });
+
+    return {
+      questionsThisExcerptCanAnswer: questions.text.replace(STRIP_REGEX, ""),
+    };
+  }
+
+  /**
+   * Extract answered questions from a list of nodes.
+   * @param {BaseNode[]} nodes Nodes to extract questions from.
+   * @returns {Promise<Array<ExtractQuestion> | Array<{}>>} Questions extracted from the nodes.
+   */
+  async extract(
+    nodes: BaseNode[],
+  ): Promise<Array<ExtractQuestion> | Array<{}>> {
+    const results = await Promise.all(
+      nodes.map((node) => this.extractQuestionsFromNode(node)),
+    );
+
+    return results;
+  }
+}
+
+type ExtractSummary = {
+  sectionSummary: string;
+  prevSectionSummary: string;
+  nextSectionSummary: string;
+};
+
+/**
+ * Extract summary from a list of nodes.
+ */
+export class SummaryExtractor extends BaseExtractor {
+  /**
+   * LLM instance.
+   * @type {LLM}
+   */
+  llm: LLM;
+
+  /**
+   * List of summaries to extract: 'self', 'prev', 'next'
+   * @type {string[]}
+   */
+  summaries: string[];
+
+  /**
+   * The prompt template to use for the summary extractor.
+   * @type {string}
+   */
+  promptTemplate: string;
+
+  private _selfSummary: boolean;
+  private _prevSummary: boolean;
+  private _nextSummary: boolean;
+
+  constructor(
+    llm: LLM,
+    summaries: string[] = ["self"],
+    promptTemplate?: string,
+  ) {
+    if (!summaries.some((s) => ["self", "prev", "next"].includes(s)))
+      throw new Error("Summaries must be one of 'self', 'prev', 'next'");
+
+    super();
+
+    this.llm = llm;
+    this.summaries = summaries;
+    this.promptTemplate =
+      promptTemplate ?? defaultSummaryExtractorPromptTemplate();
+
+    this._selfSummary = summaries.includes("self");
+    this._prevSummary = summaries.includes("prev");
+    this._nextSummary = summaries.includes("next");
+  }
+
+  /**
+   * Extract summary from a node.
+   * @param {BaseNode} node Node to extract summary from.
+   * @returns {Promise<string>} Summary extracted from the node.
+   */
+  async generateNodeSummary(node: BaseNode): Promise<string> {
+    if (this.isTextNodeOnly && !(node instanceof TextNode)) {
+      return "";
+    }
+
+    const contextStr = node.getContent(this.metadataMode);
+
+    const prompt = defaultSummaryExtractorPromptTemplate({
+      contextStr,
+    });
+
+    const summary = await this.llm.complete({
+      prompt,
+    });
+
+    return summary.text.replace(STRIP_REGEX, "");
+  }
+
+  /**
+   * Extract summaries from a list of nodes.
+   * @param {BaseNode[]} nodes Nodes to extract summaries from.
+   * @returns {Promise<Array<ExtractSummary> | Arry<{}>>} Summaries extracted from the nodes.
+   */
+  async extract(nodes: BaseNode[]): Promise<Array<ExtractSummary> | Array<{}>> {
+    if (!nodes.every((n) => n instanceof TextNode))
+      throw new Error("Only `TextNode` is allowed for `Summary` extractor");
+
+    const nodeSummaries = await Promise.all(
+      nodes.map((node) => this.generateNodeSummary(node)),
+    );
+
+    let metadataList: any[] = nodes.map(() => ({}));
+
+    for (let i = 0; i < nodes.length; i++) {
+      if (i > 0 && this._prevSummary && nodeSummaries[i - 1]) {
+        metadataList[i]["prevSectionSummary"] = nodeSummaries[i - 1];
+      }
+      if (i < nodes.length - 1 && this._nextSummary && nodeSummaries[i + 1]) {
+        metadataList[i]["nextSectionSummary"] = nodeSummaries[i + 1];
+      }
+      if (this._selfSummary && nodeSummaries[i]) {
+        metadataList[i]["sectionSummary"] = nodeSummaries[i];
+      }
+    }
+
+    return metadataList;
+  }
+}

packages/core/src/extractors/index.ts

@@ -0,0 +1,6 @@
+export {
+  KeywordExtractor,
+  QuestionsAnsweredExtractor,
+  SummaryExtractor,
+  TitleExtractor,
+} from "./MetadataExtractors";

packages/core/src/extractors/prompts.ts

@@ -0,0 +1,81 @@
+export interface DefaultPromptTemplate {
+  contextStr: string;
+}
+
+export interface DefaultKeywordExtractorPromptTemplate
+  extends DefaultPromptTemplate {
+  keywords: number;
+}
+
+export interface DefaultQuestionAnswerPromptTemplate
+  extends DefaultPromptTemplate {
+  numQuestions: number;
+}
+
+export interface DefaultNodeTextTemplate {
+  metadataStr: string;
+  content: string;
+}
+
+export const defaultKeywordExtractorPromptTemplate = ({
+  contextStr = "",
+  keywords = 5,
+}: DefaultKeywordExtractorPromptTemplate) => `
+  ${contextStr}. Give ${keywords} unique keywords for thiss
+  document. Format as comma separated. Keywords:
+`;
+
+export const defaultTitleExtractorPromptTemplate = (
+  { contextStr = "" }: DefaultPromptTemplate = {
+    contextStr: "",
+  },
+) => `
+  ${contextStr}. Give a title that summarizes all of the unique entities, titles or themes found in the context. Title:
+`;
+
+export const defaultTitleCombinePromptTemplate = (
+  { contextStr = "" }: DefaultPromptTemplate = {
+    contextStr: "",
+  },
+) => `
+  ${contextStr}. Based on the above candidate titles and content,s
+  what is the comprehensive title for this document? Title:
+`;
+
+export const defaultQuestionAnswerPromptTemplate = (
+  { contextStr = "", numQuestions = 5 }: DefaultQuestionAnswerPromptTemplate = {
+    contextStr: "",
+    numQuestions: 5,
+  },
+) => `
+  ${contextStr}. Given the contextual information,s
+  generate ${numQuestions} questions this context can provides
+  specific answers to which are unlikely to be found elsewhere.
+
+  Higher-level summaries of surrounding context may be provideds
+  as well. Try using these summaries to generate better questionss
+  that this context can answer.
+`;
+
+export const defaultSummaryExtractorPromptTemplate = (
+  { contextStr = "" }: DefaultPromptTemplate = {
+    contextStr: "",
+  },
+) => `
+  ${contextStr}. Summarize the key topics and entities of the section.s
+  Summary:
+`;
+
+export const defaultNodeTextTemplate = ({
+  metadataStr = "",
+  content = "",
+}: {
+  metadataStr?: string;
+  content?: string;
+} = {}) => `[Excerpt from document]
+${metadataStr}
+Excerpt:
+-----
+${content}
+-----
+`;

packages/core/src/extractors/types.ts

@@ -0,0 +1,73 @@
+import { BaseNode, MetadataMode, TextNode } from "../Node";
+import { TransformComponent } from "../ingestion";
+import { defaultNodeTextTemplate } from "./prompts";
+
+/*
+ * Abstract class for all extractors.
+ */
+export abstract class BaseExtractor implements TransformComponent {
+  isTextNodeOnly: boolean = true;
+  showProgress: boolean = true;
+  metadataMode: MetadataMode = MetadataMode.ALL;
+  disableTemplateRewrite: boolean = false;
+  inPlace: boolean = true;
+  numWorkers: number = 4;
+
+  abstract extract(nodes: BaseNode[]): Promise<Record<string, any>[]>;
+
+  async transform(nodes: BaseNode[], options?: any): Promise<BaseNode[]> {
+    return this.processNodes(
+      nodes,
+      options?.excludedEmbedMetadataKeys,
+      options?.excludedLlmMetadataKeys,
+    );
+  }
+
+  /**
+   *
+   * @param nodes Nodes to extract metadata from.
+   * @param excludedEmbedMetadataKeys Metadata keys to exclude from the embedding.
+   * @param excludedLlmMetadataKeys Metadata keys to exclude from the LLM.
+   * @returns Metadata extracted from the nodes.
+   */
+  async processNodes(
+    nodes: BaseNode[],
+    excludedEmbedMetadataKeys: string[] | undefined = undefined,
+    excludedLlmMetadataKeys: string[] | undefined = undefined,
+  ): Promise<BaseNode[]> {
+    let newNodes: BaseNode[];
+
+    if (this.inPlace) {
+      newNodes = nodes;
+    } else {
+      newNodes = nodes.slice();
+    }
+
+    let curMetadataList = await this.extract(newNodes);
+
+    for (let idx in newNodes) {
+      newNodes[idx].metadata = curMetadataList[idx];
+    }
+
+    for (let idx in newNodes) {
+      if (excludedEmbedMetadataKeys) {
+        newNodes[idx].excludedEmbedMetadataKeys.concat(
+          excludedEmbedMetadataKeys,
+        );
+      }
+      if (excludedLlmMetadataKeys) {
+        newNodes[idx].excludedLlmMetadataKeys.concat(excludedLlmMetadataKeys);
+      }
+      if (!this.disableTemplateRewrite) {
+        if (newNodes[idx] instanceof TextNode) {
+          newNodes[idx] = new TextNode({
+            ...newNodes[idx],
+            textTemplate: defaultNodeTextTemplate(),
+          });
+        }
+      }
+    }
+
+    return newNodes;
+  }
+}

packages/core/src/index.ts

@@ -1,4 +1,3 @@
-export * from "./ChatEngine";
 export * from "./ChatHistory";
 export * from "./GlobalsHelper";
 export * from "./Node";
@@ -15,19 +14,13 @@ export * from "./Tool";
 export * from "./callbacks/CallbackManager";
 export * from "./constants";
 export * from "./embeddings";
+export * from "./engines/chat";
+export * from "./extractors";
 export * from "./indices";
+export * from "./ingestion";
 export * from "./llm";
 export * from "./nodeParsers";
 export * from "./postprocessors";
-export * from "./readers/AssemblyAI";
-export * from "./readers/CSVReader";
-export * from "./readers/DocxReader";
-export * from "./readers/HTMLReader";
-export * from "./readers/MarkdownReader";
-export * from "./readers/NotionReader";
-export * from "./readers/PDFReader";
-export * from "./readers/SimpleDirectoryReader";
-export * from "./readers/SimpleMongoReader";
-export * from "./readers/base";
+export * from "./readers";
 export * from "./storage";
 export * from "./synthesizers";

packages/core/src/indices/BaseIndex.ts

@@ -1,8 +1,8 @@
-import { randomUUID } from "node:crypto";
 import { BaseNode, Document, jsonToNode } from "../Node";
 import { BaseQueryEngine } from "../QueryEngine";
 import { BaseRetriever } from "../Retriever";
 import { ServiceContext } from "../ServiceContext";
+import { randomUUID } from "../env";
 import { StorageContext } from "../storage/StorageContext";
 import { BaseDocumentStore } from "../storage/docStore/types";
 import { BaseIndexStore } from "../storage/indexStore/types";