Release 0.3.4 (#797)

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

apps/docs/CHANGELOG.md

@@ -1,5 +1,38 @@
 # docs
 
+## 0.0.12
+
+### Patch Changes
+
+- Updated dependencies [1dce275]
+- Updated dependencies [d10533e]
+- Updated dependencies [2008efe]
+- Updated dependencies [5e61934]
+- Updated dependencies [9e74a43]
+- Updated dependencies [ee719a1]
+  - llamaindex@0.3.4
+
+## 0.0.11
+
+### Patch Changes
+
+- Updated dependencies [e8c41c5]
+  - llamaindex@0.3.3
+
+## 0.0.10
+
+### Patch Changes
+
+- Updated dependencies [61103b6]
+  - llamaindex@0.3.2
+
+## 0.0.9
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - llamaindex@0.3.1
+
 ## 0.0.8
 
 ### Patch Changes

apps/docs/blog/2024-04-26-v0.3.0.md

@@ -72,12 +72,8 @@ export class MyAgent extends AgentRunner<MyLLM> {
   // create store is a function to create a store for each task, by default it only includes `messages` and `toolOutputs`
   createStore = AgentRunner.defaultCreateStore;
 
-  static taskHandler: TaskHandler<Anthropic> = async (step) => {
-    const { input } = step;
+  static taskHandler: TaskHandler<Anthropic> = async (step, enqueueOutput) => {
     const { llm, stream } = step.context;
-    if (input) {
-      step.context.store.messages = [...step.context.store.messages, input];
-    }
     // initialize the input
     const response = await llm.chat({
       stream,
@@ -90,27 +86,21 @@ export class MyAgent extends AgentRunner<MyLLM> {
     ];
     // your logic here to decide whether to continue the task
     const shouldContinue = Math.random(); /* <-- replace with your logic here */
+    enqueueOutput({
+      taskStep: step,
+      output: response,
+      isLast: !shouldContinue,
+    });
     if (shouldContinue) {
+      const content = await someHeavyFunctionCall();
       // if you want to continue the task, you can insert your new context for the next task step
       step.context.store.messages = [
         ...step.context.store.messages,
         {
-          content: "INSERT MY NEW DATA",
+          content,
           role: "user",
         },
       ];
-      return {
-        taskStep: step,
-        output: response,
-        isLast: false,
-      };
-    } else {
-      // if you want to end the task, you can return the response with `isLast: true`
-      return {
-        taskStep: step,
-        output: response,
-        isLast: true,
-      };
     }
   };
 }
@@ -263,6 +253,9 @@ const sumNumbers = FunctionTool.from<Input>(
 In addition to Node.js, LlamaIndexTS now offers enhanced support for Next.js, Deno, and Cloudflare Workers, making it
 more versatile across different platforms.
 
+For now, you can install llamaindex and directly import it into your existing Next.js, Deno or Cloudflare Worker project
+**without any extra configuration**.
+
 #### [Deno](https://deno.com/)
 
 You can use LlamaIndexTS in Deno by installation through JSR:

apps/docs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "docs",
-  "version": "0.0.8",
+  "version": "0.0.12",
   "private": true,
   "scripts": {
     "docusaurus": "docusaurus",

examples/.env.example

@@ -0,0 +1 @@
+DEBUG=llamaindex

examples/agent/openai-task.ts

@@ -0,0 +1,39 @@
+import { ChatResponseChunk, OpenAIAgent } from "llamaindex";
+import { ReadableStream } from "node:stream/web";
+import {
+  getCurrentIDTool,
+  getUserInfoTool,
+  getWeatherTool,
+} from "./utils/tools";
+
+async function main() {
+  // Create an OpenAIAgent with the function tools
+  const agent = new OpenAIAgent({
+    tools: [getCurrentIDTool, getUserInfoTool, getWeatherTool],
+  });
+
+  const task = await agent.createTask(
+    "What is my current address weather based on my profile?",
+    true,
+  );
+
+  for await (const stepOutput of task) {
+    const stream = stepOutput.output as ReadableStream<ChatResponseChunk>;
+    if (stepOutput.isLast) {
+      for await (const chunk of stream) {
+        process.stdout.write(chunk.delta);
+      }
+      process.stdout.write("\n");
+    } else {
+      // handing function call
+      console.log("handling function call...");
+      for await (const chunk of stream) {
+        console.log("debug:", JSON.stringify(chunk.raw));
+      }
+    }
+  }
+}
+
+void main().then(() => {
+  console.log("Done");
+});

examples/agent/openai.ts

@@ -53,7 +53,7 @@ async function main() {
     message: "How much is 5 + 5? then divide by 2",
   });
 
-  console.log(String(response));
+  console.log(response.response.message);
 }
 
 void main().then(() => {

examples/agent/react-task.ts

@@ -0,0 +1,40 @@
+import { ChatResponseChunk, ReActAgent } from "llamaindex";
+import { ReadableStream } from "node:stream/web";
+import {
+  getCurrentIDTool,
+  getUserInfoTool,
+  getWeatherTool,
+} from "./utils/tools";
+
+async function main() {
+  // Create an OpenAIAgent with the function tools
+  const agent = new ReActAgent({
+    tools: [getCurrentIDTool, getUserInfoTool, getWeatherTool],
+  });
+
+  const task = await agent.createTask(
+    "What is my current address weather based on my profile?",
+    true,
+  );
+
+  for await (const stepOutput of task) {
+    const stream = stepOutput.output as ReadableStream<ChatResponseChunk>;
+    if (stepOutput.isLast) {
+      for await (const chunk of stream) {
+        process.stdout.write(chunk.delta);
+      }
+      process.stdout.write("\n");
+    } else {
+      // handing function call
+      console.log("handling function call...");
+      for await (const chunk of stream) {
+        console.log("debug:", JSON.stringify(chunk.raw));
+      }
+    }
+    console.log("---");
+  }
+}
+
+void main().then(() => {
+  console.log("Done");
+});

examples/agent/step_wise_openai.ts

@@ -1,97 +0,0 @@
-import { FunctionTool, OpenAIAgent } from "llamaindex";
-import { ReadableStream } from "node:stream/web";
-
-// Define a function to sum two numbers
-function sumNumbers({ a, b }: { a: number; b: number }) {
-  return `${a + b}`;
-}
-
-// Define a function to divide two numbers
-function divideNumbers({ a, b }: { a: number; b: number }) {
-  return `${a / b}`;
-}
-
-// Define the parameters of the sum function as a JSON schema
-const sumJSON = {
-  type: "object",
-  properties: {
-    a: {
-      type: "number",
-      description: "The first number",
-    },
-    b: {
-      type: "number",
-      description: "The second number",
-    },
-  },
-  required: ["a", "b"],
-} as const;
-
-const divideJSON = {
-  type: "object",
-  properties: {
-    a: {
-      type: "number",
-      description: "The dividend",
-    },
-    b: {
-      type: "number",
-      description: "The divisor",
-    },
-  },
-  required: ["a", "b"],
-} as const;
-
-async function main() {
-  // Create a function tool from the sum function
-  const functionTool = new FunctionTool(sumNumbers, {
-    name: "sumNumbers",
-    description: "Use this function to sum two numbers",
-    parameters: sumJSON,
-  });
-
-  // Create a function tool from the divide function
-  const functionTool2 = new FunctionTool(divideNumbers, {
-    name: "divideNumbers",
-    description: "Use this function to divide two numbers",
-    parameters: divideJSON,
-  });
-
-  // Create an OpenAIAgent with the function tools
-  const agent = new OpenAIAgent({
-    tools: [functionTool, functionTool2],
-  });
-
-  // Create a task to sum and divide numbers
-  const task = await agent.createTask("How much is 5 + 5? then divide by 2");
-
-  let count = 0;
-
-  for await (const stepOutput of task) {
-    console.log(`Runnning step ${count++}`);
-    console.log(`======== OUTPUT ==========`);
-    const output = stepOutput.output;
-    if (output instanceof ReadableStream) {
-      for await (const chunk of output) {
-        process.stdout.write(chunk.delta);
-      }
-    } else {
-      console.log(output);
-    }
-    console.log(`==========================`);
-
-    if (stepOutput.isLast) {
-      if (stepOutput.output instanceof ReadableStream) {
-        for await (const chunk of stepOutput.output) {
-          process.stdout.write(chunk.delta);
-        }
-      } else {
-        console.log(stepOutput.output);
-      }
-    }
-  }
-}
-
-void main().then(() => {
-  console.log("Done");
-});

examples/agent/utils/tools.ts

@@ -0,0 +1,54 @@
+import { FunctionTool } from "llamaindex";
+
+export const getCurrentIDTool = FunctionTool.from(
+  () => {
+    console.log("Getting user id...");
+    return crypto.randomUUID();
+  },
+  {
+    name: "get_user_id",
+    description: "Get a random user id",
+  },
+);
+
+export const getUserInfoTool = FunctionTool.from(
+  ({ userId }: { userId: string }) => {
+    console.log("Getting user info...", userId);
+    return `Name: Alex; Address: 1234 Main St, CA; User ID: ${userId}`;
+  },
+  {
+    name: "get_user_info",
+    description: "Get user info",
+    parameters: {
+      type: "object",
+      properties: {
+        userId: {
+          type: "string",
+          description: "The user id",
+        },
+      },
+      required: ["userId"],
+    },
+  },
+);
+
+export const getWeatherTool = FunctionTool.from(
+  ({ address }: { address: string }) => {
+    console.log("Getting weather...", address);
+    return `${address} is in a sunny location!`;
+  },
+  {
+    name: "get_weather",
+    description: "Get the current weather for a location",
+    parameters: {
+      type: "object",
+      properties: {
+        address: {
+          type: "string",
+          description: "The address",
+        },
+      },
+      required: ["address"],
+    },
+  },
+);

examples/huggingface/chat.ts

@@ -0,0 +1,22 @@
+import { HuggingFaceInferenceAPI } from "llamaindex";
+
+(async () => {
+  if (!process.env.HUGGING_FACE_TOKEN) {
+    throw new Error("Please set the HUGGING_FACE_TOKEN environment variable.");
+  }
+  const hf = new HuggingFaceInferenceAPI({
+    accessToken: process.env.HUGGING_FACE_TOKEN,
+    model: "mistralai/Mixtral-8x7B-Instruct-v0.1",
+  });
+  const result = await hf.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);
+})();

packages/core/CHANGELOG.md

@@ -1,5 +1,34 @@
 # llamaindex
 
+## 0.3.4
+
+### Patch Changes
+
+- 1dce275: fix: export `StorageContext` on edge runtime
+- d10533e: feat: add hugging face llm
+- 2008efe: feat: add verbose mode to Agent
+- 5e61934: fix: remove clone object in `CallbackManager.dispatchEvent`
+- 9e74a43: feat: add top k to `asQueryEngine`
+- ee719a1: fix: streaming for ReAct Agent
+
+## 0.3.3
+
+### Patch Changes
+
+- e8c41c5: fix: wrong gemini streaming chat response
+
+## 0.3.2
+
+### Patch Changes
+
+- 61103b6: fix: streaming for `Agent.createTask` API
+
+## 0.3.1
+
+### Patch Changes
+
+- 46227f2: fix: build error on next.js nodejs runtime
+
 ## 0.3.0
 
 ### Minor Changes

packages/core/e2e/CHANGELOG.md

@@ -0,0 +1,7 @@
+# @llamaindex/core-e2e
+
+## 0.0.3
+
+### Patch Changes
+
+- 61103b6: fix: streaming for `Agent.createTask` API

packages/core/e2e/examples/cloudflare-worker-agent/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @llamaindex/cloudflare-worker-agent-test
 
+## 0.0.5
+
+### Patch Changes
+
+- Updated dependencies [1dce275]
+- Updated dependencies [d10533e]
+- Updated dependencies [2008efe]
+- Updated dependencies [5e61934]
+- Updated dependencies [9e74a43]
+- Updated dependencies [ee719a1]
+  - llamaindex@0.3.4
+
+## 0.0.4
+
+### Patch Changes
+
+- Updated dependencies [e8c41c5]
+  - llamaindex@0.3.3
+
+## 0.0.3
+
+### Patch Changes
+
+- Updated dependencies [61103b6]
+  - llamaindex@0.3.2
+
+## 0.0.2
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - llamaindex@0.3.1
+
 ## 0.0.1
 
 ### Patch Changes

packages/core/e2e/examples/cloudflare-worker-agent/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/cloudflare-worker-agent-test",
-  "version": "0.0.1",
+  "version": "0.0.5",
   "type": "module",
   "private": true,
   "scripts": {

packages/core/e2e/examples/nextjs-agent/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @llamaindex/next-agent-test
 
+## 0.1.5
+
+### Patch Changes
+
+- Updated dependencies [1dce275]
+- Updated dependencies [d10533e]
+- Updated dependencies [2008efe]
+- Updated dependencies [5e61934]
+- Updated dependencies [9e74a43]
+- Updated dependencies [ee719a1]
+  - llamaindex@0.3.4
+
+## 0.1.4
+
+### Patch Changes
+
+- Updated dependencies [e8c41c5]
+  - llamaindex@0.3.3
+
+## 0.1.3
+
+### Patch Changes
+
+- Updated dependencies [61103b6]
+  - llamaindex@0.3.2
+
+## 0.1.2
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - llamaindex@0.3.1
+
 ## 0.1.1
 
 ### Patch Changes

packages/core/e2e/examples/nextjs-agent/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/next-agent-test",
-  "version": "0.1.1",
+  "version": "0.1.5",
   "private": true,
   "scripts": {
     "dev": "next dev",

packages/core/e2e/examples/nextjs-agent/src/app/test/page.tsx

@@ -0,0 +1,5 @@
+import "llamaindex";
+
+export default function Page() {
+  return "hello world!";
+}

packages/core/e2e/examples/nextjs-edge-runtime/CHANGELOG.md

@@ -1,5 +1,12 @@
 # test-edge-runtime
 
+## 0.1.6
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - @llamaindex/edge@0.3.1
+
 ## 0.1.5
 
 ### Patch Changes

packages/core/e2e/examples/nextjs-edge-runtime/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/nextjs-edge-runtime-test",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "private": true,
   "scripts": {
     "dev": "next dev",

packages/core/e2e/examples/waku-query-engine/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @llamaindex/waku-query-engine-test
 
+## 0.0.5
+
+### Patch Changes
+
+- Updated dependencies [1dce275]
+- Updated dependencies [d10533e]
+- Updated dependencies [2008efe]
+- Updated dependencies [5e61934]
+- Updated dependencies [9e74a43]
+- Updated dependencies [ee719a1]
+  - llamaindex@0.3.4
+
+## 0.0.4
+
+### Patch Changes
+
+- Updated dependencies [e8c41c5]
+  - llamaindex@0.3.3
+
+## 0.0.3
+
+### Patch Changes
+
+- Updated dependencies [61103b6]
+  - llamaindex@0.3.2
+
+## 0.0.2
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - llamaindex@0.3.1
+
 ## 0.0.1
 
 ### Patch Changes

packages/core/e2e/examples/waku-query-engine/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/waku-query-engine-test",
-  "version": "0.0.1",
+  "version": "0.0.5",
   "type": "module",
   "private": true,
   "scripts": {

packages/core/e2e/node/react.e2e.ts

@@ -27,3 +27,23 @@ await test("react agent", async (t) => {
     ok(extractText(response.message.content).includes("72"));
   });
 });
+
+await test("react agent stream", async (t) => {
+  await mockLLMEvent(t, "react-agent-stream");
+  await t.test("get weather", async () => {
+    const agent = new ReActAgent({
+      tools: [getWeatherTool],
+    });
+
+    const stream = await agent.chat({
+      stream: true,
+      message: "What is the weather like in San Francisco?",
+    });
+
+    let content = "";
+    for await (const { response } of stream) {
+      content += response.delta;
+    }
+    ok(content.includes("72"));
+  });
+});

packages/core/e2e/node/snapshot/react-agent-stream.snap

@@ -0,0 +1,488 @@
+{
+  "llmEventStart": [
+    {
+      "id": "PRESERVE_0",
+      "messages": [
+        {
+          "role": "system",
+          "content": "You are designed to help with a variety of tasks, from answering questions to providing summaries to other types of analyses.\n\n## Tools\nYou have access to a wide variety of tools. You are responsible for using\nthe tools in any sequence you deem appropriate to complete the task at hand.\nThis may require breaking the task into subtasks and using different tools\nto complete each subtask.\n\nYou have access to the following tools:\n- getWeather: Get the weather for a city with schema: {\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\",\"description\":\"The city to get the weather for\"}},\"required\":[\"city\"]}\n\n## Output Format\nTo answer the question, please use the following format.\n\n\"\"\"\nThought: I need to use a tool to help me answer the question.\nAction: tool name (one of getWeather) if using a tool.\nAction Input: the input to the tool, in a JSON format representing the kwargs (e.g. {{\"input\": \"hello world\", \"num_beams\": 5}})\n\"\"\"\n\nPlease ALWAYS start with a Thought.\n\nPlease use a valid JSON format for the Action Input. Do NOT do this {{'input': 'hello world', 'num_beams': 5}}.\n\nIf this format is used, the user will respond in the following format:\n\n\"\"\"\"\nObservation: tool response\n\"\"\"\"\n\nYou should keep repeating the above format until you have enough information\nto answer the question without using any more tools. At that point, you MUST respond\nin the one of the following two formats:\n\n\"\"\"\"\nThought: I can answer without using any more tools.\nAnswer: [your answer here]\n\"\"\"\"\n\n\"\"\"\"\nThought: I cannot answer the question with the provided tools.\nAnswer: Sorry, I cannot answer your query.\n\"\"\"\"\n\n## Current Conversation\nBelow is the current conversation consisting of interleaving human and assistant messages."
+        },
+        {
+          "role": "user",
+          "content": "What is the weather like in San Francisco?"
+        }
+      ]
+    },
+    {
+      "id": "PRESERVE_1",
+      "messages": [
+        {
+          "role": "system",
+          "content": "You are designed to help with a variety of tasks, from answering questions to providing summaries to other types of analyses.\n\n## Tools\nYou have access to a wide variety of tools. You are responsible for using\nthe tools in any sequence you deem appropriate to complete the task at hand.\nThis may require breaking the task into subtasks and using different tools\nto complete each subtask.\n\nYou have access to the following tools:\n- getWeather: Get the weather for a city with schema: {\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\",\"description\":\"The city to get the weather for\"}},\"required\":[\"city\"]}\n\n## Output Format\nTo answer the question, please use the following format.\n\n\"\"\"\nThought: I need to use a tool to help me answer the question.\nAction: tool name (one of getWeather) if using a tool.\nAction Input: the input to the tool, in a JSON format representing the kwargs (e.g. {{\"input\": \"hello world\", \"num_beams\": 5}})\n\"\"\"\n\nPlease ALWAYS start with a Thought.\n\nPlease use a valid JSON format for the Action Input. Do NOT do this {{'input': 'hello world', 'num_beams': 5}}.\n\nIf this format is used, the user will respond in the following format:\n\n\"\"\"\"\nObservation: tool response\n\"\"\"\"\n\nYou should keep repeating the above format until you have enough information\nto answer the question without using any more tools. At that point, you MUST respond\nin the one of the following two formats:\n\n\"\"\"\"\nThought: I can answer without using any more tools.\nAnswer: [your answer here]\n\"\"\"\"\n\n\"\"\"\"\nThought: I cannot answer the question with the provided tools.\nAnswer: Sorry, I cannot answer your query.\n\"\"\"\"\n\n## Current Conversation\nBelow is the current conversation consisting of interleaving human and assistant messages."
+        },
+        {
+          "role": "user",
+          "content": "What is the weather like in San Francisco?"
+        },
+        {
+          "role": "assistant",
+          "content": "Thought: I need to use a tool to help me answer the question.\nAction: getWeather\nInput: {\n  city: San Francisco\n}"
+        },
+        {
+          "role": "user",
+          "content": "Observation: The weather in San Francisco is 72 degrees"
+        }
+      ]
+    }
+  ],
+  "llmEventEnd": [
+    {
+      "id": "PRESERVE_0",
+      "response": {
+        "raw": null,
+        "message": {
+          "content": "Thought: I need to use a tool to help me answer the question.\nAction: getWeather\nAction Input: {\"city\": \"San Francisco\"}",
+          "role": "assistant",
+          "options": {}
+        }
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "response": {
+        "raw": null,
+        "message": {
+          "content": "Thought: I can answer without using any more tools.\nAnswer: The weather in San Francisco is 72 degrees.",
+          "role": "assistant",
+          "options": {}
+        }
+      }
+    }
+  ],
+  "llmEventStream": [
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Thought"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ":"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " I"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " need"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " to"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " use"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " a"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " tool"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " to"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " help"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " me"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " answer"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " the"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " question"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ".\n"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Action"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ":"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " get"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Weather"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "\n"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Action"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " Input"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ":"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " {\""
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "city"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "\":"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " \""
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "San"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " Francisco"
+      }
+    },
+    {
+      "id": "PRESERVE_0",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "\"}"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Thought"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ":"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " I"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " can"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " answer"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " without"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " using"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " any"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " more"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " tools"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ".\n"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "Answer"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": ":"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " The"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " weather"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " in"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " San"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " Francisco"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " is"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " "
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "72"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": " degrees"
+      }
+    },
+    {
+      "id": "PRESERVE_1",
+      "chunk": {
+        "raw": null,
+        "options": {},
+        "delta": "."
+      }
+    }
+  ]
+}

packages/core/e2e/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@llamaindex/core-e2e",
   "private": true,
-  "version": "0.0.2",
+  "version": "0.0.3",
   "type": "module",
   "scripts": {
     "e2e": "node --import tsx --import ./mock-register.js --test ./node/*.e2e.ts",

packages/core/jsr.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/core",
-  "version": "0.2.3",
+  "version": "0.3.4",
   "exports": "./src/index.ts",
   "imports": {
     "@llamaindex/env": "jsr:@llamaindex/env@0.0.6"

packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llamaindex",
-  "version": "0.3.0",
+  "version": "0.3.4",
   "expectedMinorVersion": "3",
   "license": "MIT",
   "type": "module",
@@ -10,6 +10,7 @@
     "@datastax/astra-db-ts": "^1.0.1",
     "@google/generative-ai": "^0.8.0",
     "@grpc/grpc-js": "^1.10.6",
+    "@huggingface/inference": "^2.6.7",
     "@llamaindex/cloud": "0.0.5",
     "@llamaindex/env": "workspace:*",
     "@mistralai/mistralai": "^0.1.3",

packages/core/src/Settings.ts

@@ -55,9 +55,9 @@ class GlobalSettings implements Config {
   get debug() {
     const debug = getEnv("DEBUG");
     return (
-      getEnv("NODE_ENV") === "development" &&
-      Boolean(debug) &&
-      debug?.includes("llamaindex")
+      (Boolean(debug) && debug?.includes("llamaindex")) ||
+      debug === "*" ||
+      debug === "true"
     );
   }
 

packages/core/src/agent/anthropic.ts

@@ -49,6 +49,7 @@ export class AnthropicAgent extends AgentRunner<Anthropic> {
         "tools" in params
           ? params.tools
           : params.toolRetriever.retrieve.bind(params.toolRetriever),
+      verbose: params.verbose ?? false,
     });
   }
 
@@ -67,12 +68,8 @@ export class AnthropicAgent extends AgentRunner<Anthropic> {
     return super.chat(params);
   }
 
-  static taskHandler: TaskHandler<Anthropic> = async (step) => {
-    const { input } = step;
+  static taskHandler: TaskHandler<Anthropic> = async (step, enqueueOutput) => {
     const { llm, getTools, stream } = step.context;
-    if (input) {
-      step.context.store.messages = [...step.context.store.messages, input];
-    }
     const lastMessage = step.context.store.messages.at(-1)!.content;
     const tools = await getTools(lastMessage);
     if (stream === true) {
@@ -88,37 +85,36 @@ export class AnthropicAgent extends AgentRunner<Anthropic> {
       response.message,
     ];
     const options = response.message.options ?? {};
+    enqueueOutput({
+      taskStep: step,
+      output: response,
+      isLast: !("toolCall" in options),
+    });
     if ("toolCall" in options) {
       const { toolCall } = options;
       const targetTool = tools.find(
         (tool) => tool.metadata.name === toolCall.name,
       );
-      const toolOutput = await callTool(targetTool, toolCall);
+      const toolOutput = await callTool(
+        targetTool,
+        toolCall,
+        step.context.logger,
+      );
       step.context.store.toolOutputs.push(toolOutput);
-      return {
-        taskStep: step,
-        output: {
-          raw: response.raw,
-          message: {
-            content: stringifyJSONToMessageContent(toolOutput.output),
-            role: "user",
-            options: {
-              toolResult: {
-                result: toolOutput.output,
-                isError: toolOutput.isError,
-                id: toolCall.id,
-              },
+      step.context.store.messages = [
+        ...step.context.store.messages,
+        {
+          content: stringifyJSONToMessageContent(toolOutput.output),
+          role: "user",
+          options: {
+            toolResult: {
+              result: toolOutput.output,
+              isError: toolOutput.isError,
+              id: toolCall.id,
             },
           },
         },
-        isLast: false,
-      };
-    } else {
-      return {
-        taskStep: step,
-        output: response,
-        isLast: true,
-      };
+      ];
     }
   };
 }

packages/core/src/agent/base.ts

@@ -4,12 +4,14 @@ import {
   pipeline,
   randomUUID,
 } from "@llamaindex/env";
+import { Settings } from "../Settings.js";
 import {
   type ChatEngine,
   type ChatEngineParamsNonStreaming,
   type ChatEngineParamsStreaming,
 } from "../engines/chat/index.js";
 import { wrapEventCaller } from "../internal/context/EventCaller.js";
+import { consoleLogger, emptyLogger } from "../internal/logger.js";
 import { getCallbackManager } from "../internal/settings/CallbackManager.js";
 import { isAsyncIterable } from "../internal/utils.js";
 import type {
@@ -27,14 +29,10 @@ import type {
   TaskStep,
   TaskStepOutput,
 } from "./types.js";
-import { consumeAsyncIterable } from "./utils.js";
 
 export const MAX_TOOL_CALLS = 10;
 
-/**
- * @internal
- */
-export async function* createTaskImpl<
+export function createTaskOutputStream<
   Model extends LLM,
   Store extends object = {},
   AdditionalMessageOptions extends object = Model extends LLM<
@@ -46,65 +44,67 @@ export async function* createTaskImpl<
 >(
   handler: TaskHandler<Model, Store, AdditionalMessageOptions>,
   context: AgentTaskContext<Model, Store, AdditionalMessageOptions>,
-  _input: ChatMessage<AdditionalMessageOptions>,
-): AsyncGenerator<TaskStepOutput<Model, Store, AdditionalMessageOptions>> {
-  let isFirst = true;
-  let isDone = false;
-  let input: ChatMessage<AdditionalMessageOptions> | null = _input;
-  let prevStep: TaskStep<Model, Store, AdditionalMessageOptions> | null = null;
-  while (!isDone) {
-    const step: TaskStep<Model, Store, AdditionalMessageOptions> = {
-      id: randomUUID(),
-      input,
-      context,
-      prevStep,
-      nextSteps: new Set(),
-    };
-    if (prevStep) {
-      prevStep.nextSteps.add(step);
-    }
-    const prevToolCallCount = step.context.toolCallCount;
-    if (!step.context.shouldContinue(step)) {
-      throw new Error("Tool call count exceeded limit");
-    }
-    if (isFirst) {
+): ReadableStream<TaskStepOutput<Model, Store, AdditionalMessageOptions>> {
+  const steps: TaskStep<Model, Store, AdditionalMessageOptions>[] = [];
+  return new ReadableStream<
+    TaskStepOutput<Model, Store, AdditionalMessageOptions>
+  >({
+    pull: async (controller) => {
+      const step: TaskStep<Model, Store, AdditionalMessageOptions> = {
+        id: randomUUID(),
+        context,
+        prevStep: null,
+        nextSteps: new Set(),
+      };
+      if (steps.length > 0) {
+        step.prevStep = steps[steps.length - 1];
+      }
+      const taskOutputs: TaskStepOutput<
+        Model,
+        Store,
+        AdditionalMessageOptions
+      >[] = [];
+      steps.push(step);
+      const enqueueOutput = (
+        output: TaskStepOutput<Model, Store, AdditionalMessageOptions>,
+      ) => {
+        context.logger.log("Enqueueing output for step(id, %s).", step.id);
+        taskOutputs.push(output);
+        controller.enqueue(output);
+      };
       getCallbackManager().dispatchEvent("agent-start", {
         payload: {
           startStep: step,
         },
       });
-      isFirst = false;
-    }
-    const taskOutput = await handler(step);
-    const { isLast, output, taskStep } = taskOutput;
-    // do not consume last output
-    if (!isLast) {
-      if (output) {
-        input = isAsyncIterable(output)
-          ? await consumeAsyncIterable(output)
-          : output.message;
-      } else {
-        input = null;
-      }
-    }
-    context = {
-      ...taskStep.context,
-      store: {
-        ...taskStep.context.store,
-      },
-      toolCallCount: prevToolCallCount + 1,
-    };
-    if (isLast) {
-      isDone = true;
-      getCallbackManager().dispatchEvent("agent-end", {
-        payload: {
-          endStep: step,
+
+      context.logger.log("Starting step(id, %s).", step.id);
+      await handler(step, enqueueOutput);
+      context.logger.log("Finished step(id, %s).", step.id);
+      // fixme: support multi-thread when there are multiple outputs
+      // todo: for now we pretend there is only one task output
+      const { isLast, taskStep } = taskOutputs[0];
+      context = {
+        ...taskStep.context,
+        store: {
+          ...taskStep.context.store,
         },
-      });
-    }
-    prevStep = taskStep;
-    yield taskOutput;
-  }
+        toolCallCount: 1,
+      };
+      if (isLast) {
+        context.logger.log(
+          "Final step(id, %s) reached, closing task.",
+          step.id,
+        );
+        getCallbackManager().dispatchEvent("agent-end", {
+          payload: {
+            endStep: step,
+          },
+        });
+        controller.close();
+      }
+    },
+  });
 }
 
 export type AgentStreamChatResponse<Options extends object> = {
@@ -134,6 +134,7 @@ export type AgentRunnerParams<
   tools:
     | BaseToolWithCall[]
     | ((query: MessageContent) => Promise<BaseToolWithCall[]>);
+  verbose: boolean;
 };
 
 export type AgentParamsBase<
@@ -148,6 +149,7 @@ export type AgentParamsBase<
   llm?: AI;
   chatHistory?: ChatMessage<AdditionalMessageOptions>[];
   systemPrompt?: MessageContent;
+  verbose?: boolean;
 };
 
 /**
@@ -170,15 +172,16 @@ export abstract class AgentWorker<
     query: string,
     context: AgentTaskContext<AI, Store, AdditionalMessageOptions>,
   ): ReadableStream<TaskStepOutput<AI, Store, AdditionalMessageOptions>> {
-    const taskGenerator = createTaskImpl(this.taskHandler, context, {
+    context.store.messages.push({
       role: "user",
       content: query,
     });
+    const taskOutputStream = createTaskOutputStream(this.taskHandler, context);
     return new ReadableStream<
       TaskStepOutput<AI, Store, AdditionalMessageOptions>
     >({
       start: async (controller) => {
-        for await (const stepOutput of taskGenerator) {
+        for await (const stepOutput of taskOutputStream) {
           this.#taskSet.add(stepOutput.taskStep);
           controller.enqueue(stepOutput);
           if (stepOutput.isLast) {
@@ -226,6 +229,7 @@ export abstract class AgentRunner<
   readonly #systemPrompt: MessageContent | null = null;
   #chatHistory: ChatMessage<AdditionalMessageOptions>[];
   readonly #runner: AgentWorker<AI, Store, AdditionalMessageOptions>;
+  readonly #verbose: boolean;
 
   // create extra store
   abstract createStore(): Store;
@@ -237,14 +241,15 @@ export abstract class AgentRunner<
   protected constructor(
     params: AgentRunnerParams<AI, Store, AdditionalMessageOptions>,
   ) {
-    const { llm, chatHistory, runner, tools } = params;
+    const { llm, chatHistory, systemPrompt, runner, tools, verbose } = params;
     this.#llm = llm;
     this.#chatHistory = chatHistory;
     this.#runner = runner;
-    if (params.systemPrompt) {
-      this.#systemPrompt = params.systemPrompt;
+    if (systemPrompt) {
+      this.#systemPrompt = systemPrompt;
     }
     this.#tools = tools;
+    this.#verbose = verbose;
   }
 
   get llm() {
@@ -255,6 +260,10 @@ export abstract class AgentRunner<
     return this.#chatHistory;
   }
 
+  get verbose(): boolean {
+    return Settings.debug || this.#verbose;
+  }
+
   public reset(): void {
     this.#chatHistory = [];
   }
@@ -278,8 +287,11 @@ export abstract class AgentRunner<
     return task.context.toolCallCount < MAX_TOOL_CALLS;
   }
 
-  // fixme: this shouldn't be async
-  async createTask(message: MessageContent, stream: boolean = false) {
+  createTask(
+    message: MessageContent,
+    stream: boolean = false,
+    verbose: boolean | undefined = undefined,
+  ) {
     const initialMessages = [...this.#chatHistory];
     if (this.#systemPrompt !== null) {
       const systemPrompt = this.#systemPrompt;
@@ -304,6 +316,13 @@ export abstract class AgentRunner<
         toolOutputs: [] as ToolOutput[],
       },
       shouldContinue: AgentRunner.shouldContinue,
+      logger:
+        // disable verbose if explicitly set to false
+        verbose === false
+          ? emptyLogger
+          : verbose || this.verbose
+            ? consoleLogger
+            : emptyLogger,
     });
   }
 
@@ -320,7 +339,7 @@ export abstract class AgentRunner<
     | AgentChatResponse<AdditionalMessageOptions>
     | ReadableStream<AgentStreamChatResponse<AdditionalMessageOptions>>
   > {
-    const task = await this.createTask(params.message, !!params.stream);
+    const task = this.createTask(params.message, !!params.stream);
     const stepOutput = await pipeline(
       task,
       async (

packages/core/src/agent/openai.ts

@@ -46,17 +46,14 @@ export class OpenAIAgent extends AgentRunner<OpenAI> {
         "tools" in params
           ? params.tools
           : params.toolRetriever.retrieve.bind(params.toolRetriever),
+      verbose: params.verbose ?? false,
     });
   }
 
   createStore = AgentRunner.defaultCreateStore;
 
-  static taskHandler: TaskHandler<OpenAI> = async (step) => {
-    const { input } = step;
+  static taskHandler: TaskHandler<OpenAI> = async (step, enqueueOutput) => {
     const { llm, stream, getTools } = step.context;
-    if (input) {
-      step.context.store.messages = [...step.context.store.messages, input];
-    }
     const lastMessage = step.context.store.messages.at(-1)!.content;
     const tools = await getTools(lastMessage);
     const response = await llm.chat({
@@ -71,37 +68,36 @@ export class OpenAIAgent extends AgentRunner<OpenAI> {
         response.message,
       ];
       const options = response.message.options ?? {};
+      enqueueOutput({
+        taskStep: step,
+        output: response,
+        isLast: !("toolCall" in options),
+      });
       if ("toolCall" in options) {
         const { toolCall } = options;
         const targetTool = tools.find(
           (tool) => tool.metadata.name === toolCall.name,
         );
-        const toolOutput = await callTool(targetTool, toolCall);
+        const toolOutput = await callTool(
+          targetTool,
+          toolCall,
+          step.context.logger,
+        );
         step.context.store.toolOutputs.push(toolOutput);
-        return {
-          taskStep: step,
-          output: {
-            raw: response.raw,
-            message: {
-              content: stringifyJSONToMessageContent(toolOutput.output),
-              role: "user",
-              options: {
-                toolResult: {
-                  result: toolOutput.output,
-                  isError: toolOutput.isError,
-                  id: toolCall.id,
-                },
+        step.context.store.messages = [
+          ...step.context.store.messages,
+          {
+            role: "user" as const,
+            content: stringifyJSONToMessageContent(toolOutput.output),
+            options: {
+              toolResult: {
+                result: toolOutput.output,
+                isError: toolOutput.isError,
+                id: toolCall.id,
               },
             },
           },
-          isLast: false,
-        };
-      } else {
-        return {
-          taskStep: step,
-          output: response,
-          isLast: true,
-        };
+        ];
       }
     } else {
       const responseChunkStream = new ReadableStream<
@@ -126,6 +122,11 @@ export class OpenAIAgent extends AgentRunner<OpenAI> {
       // check if first chunk has tool calls, if so, this is a function call
       // otherwise, it's a regular message
       const hasToolCall = !!(value.options && "toolCall" in value.options);
+      enqueueOutput({
+        taskStep: step,
+        output: finalStream,
+        isLast: !hasToolCall,
+      });
 
       if (hasToolCall) {
         // you need to consume the response to get the full toolCalls
@@ -158,7 +159,11 @@ export class OpenAIAgent extends AgentRunner<OpenAI> {
               },
             },
           ];
-          const toolOutput = await callTool(targetTool, toolCall);
+          const toolOutput = await callTool(
+            targetTool,
+            toolCall,
+            step.context.logger,
+          );
           step.context.store.messages = [
             ...step.context.store.messages,
             {
@@ -175,17 +180,6 @@ export class OpenAIAgent extends AgentRunner<OpenAI> {
           ];
           step.context.store.toolOutputs.push(toolOutput);
         }
-        return {
-          taskStep: step,
-          output: null,
-          isLast: false,
-        };
-      } else {
-        return {
-          taskStep: step,
-          output: finalStream,
-          isLast: true,
-        };
       }
     }
   };

packages/core/src/agent/react.ts

@@ -1,5 +1,4 @@
-import { pipeline, randomUUID } from "@llamaindex/env";
-import { Settings } from "../Settings.js";
+import { randomUUID, ReadableStream } from "@llamaindex/env";
 import { getReACTAgentSystemHeader } from "../internal/prompt/react.js";
 import {
   isAsyncIterable,
@@ -13,6 +12,7 @@ import {
 } from "../llm/index.js";
 import { extractText } from "../llm/utils.js";
 import { ObjectRetriever } from "../objects/index.js";
+import { Settings } from "../Settings.js";
 import type {
   BaseTool,
   BaseToolWithCall,
@@ -60,7 +60,7 @@ type ActionReason = BaseReason & {
 type ResponseReason = BaseReason & {
   type: "response";
   thought: string;
-  response: ChatResponse | AsyncIterable<ChatResponseChunk>;
+  response: ChatResponse;
 };
 
 type Reason = ObservationReason | ActionReason | ResponseReason;
@@ -74,16 +74,9 @@ function reasonFormatter(reason: Reason): string | Promise<string> {
         reason.input,
       )}`;
     case "response": {
-      if (isAsyncIterable(reason.response)) {
-        return consumeAsyncIterable(reason.response).then(
-          (message) =>
-            `Thought: ${reason.thought}\nAnswer: ${extractText(message.content)}`,
-        );
-      } else {
-        return `Thought: ${reason.thought}\nAnswer: ${extractText(
-          reason.response.message.content,
-        )}`;
-      }
+      return `Thought: ${reason.thought}\nAnswer: ${extractText(
+        reason.response.message.content,
+      )}`;
     }
   }
 }
@@ -146,35 +139,52 @@ function actionInputParser(jsonStr: string): JSONObject {
 
 type ReACTOutputParser = <Options extends object>(
   output: ChatResponse<Options> | AsyncIterable<ChatResponseChunk<Options>>,
+  onResolveType: (
+    type: "action" | "thought" | "answer",
+    response:
+      | ChatResponse<Options>
+      | ReadableStream<ChatResponseChunk<Options>>,
+  ) => void,
 ) => Promise<Reason>;
 
 const reACTOutputParser: ReACTOutputParser = async (
   output,
+  onResolveType,
 ): Promise<Reason> => {
   let reason: Reason | null = null;
 
   if (isAsyncIterable(output)) {
     const [peakStream, finalStream] = createReadableStream(output).tee();
-    const type = await pipeline(peakStream, async (iter) => {
-      let content = "";
-      for await (const chunk of iter) {
-        content += chunk.delta;
-        if (content.includes("Action:")) {
-          return "action";
-        } else if (content.includes("Answer:")) {
-          return "answer";
-        } else if (content.includes("Thought:")) {
-          return "thought";
-        }
+    const reader = peakStream.getReader();
+    let type: "action" | "thought" | "answer" | null = null;
+    let content = "";
+    do {
+      const { done, value } = await reader.read();
+      if (done) {
+        break;
       }
-    });
+      content += value.delta;
+      if (content.includes("Action:")) {
+        type = "action";
+      } else if (content.includes("Answer:")) {
+        type = "answer";
+      }
+    } while (true);
+    if (type === null) {
+      // `Thought:` is always present at the beginning of the output.
+      type = "thought";
+    }
+    reader.releaseLock();
+    if (!type) {
+      throw new Error("Could not determine type of output");
+    }
+    onResolveType(type, finalStream);
     // step 2: do the parsing from content
     switch (type) {
       case "action": {
         // have to consume the stream to get the full content
-        const response = await consumeAsyncIterable(finalStream);
-        const { content } = response;
-        const [thought, action, input] = extractToolUse(content);
+        const response = await consumeAsyncIterable(peakStream, content);
+        const [thought, action, input] = extractToolUse(response.content);
         const jsonStr = extractJsonStr(input);
         let json: JSONObject;
         try {
@@ -192,18 +202,20 @@ const reACTOutputParser: ReACTOutputParser = async (
       }
       case "thought": {
         const thought = "(Implicit) I can answer without any more tools!";
+        const response = await consumeAsyncIterable(peakStream, content);
         reason = {
           type: "response",
           thought,
-          // bypass the response, because here we don't need to do anything with it
-          response: finalStream,
+          response: {
+            raw: peakStream,
+            message: response,
+          },
         };
         break;
       }
       case "answer": {
-        const response = await consumeAsyncIterable(finalStream);
-        const { content } = response;
-        const [thought, answer] = extractFinalResponse(content);
+        const response = await consumeAsyncIterable(peakStream, content);
+        const [thought, answer] = extractFinalResponse(response.content);
         reason = {
           type: "response",
           thought,
@@ -227,7 +239,9 @@ const reACTOutputParser: ReACTOutputParser = async (
       ? "answer"
       : content.includes("Action:")
         ? "action"
-        : "thought";
+        : // `Thought:` is always present at the beginning of the output.
+          "thought";
+    onResolveType(type, output);
 
     // step 2: do the parsing from content
     switch (type) {
@@ -340,6 +354,7 @@ export class ReActAgent extends AgentRunner<LLM, ReACTAgentStore> {
         "tools" in params
           ? params.tools
           : params.toolRetriever.retrieve.bind(params.toolRetriever),
+      verbose: params.verbose ?? false,
     });
   }
 
@@ -349,12 +364,11 @@ export class ReActAgent extends AgentRunner<LLM, ReACTAgentStore> {
     };
   }
 
-  static taskHandler: TaskHandler<LLM, ReACTAgentStore> = async (step) => {
+  static taskHandler: TaskHandler<LLM, ReACTAgentStore> = async (
+    step,
+    enqueueOutput,
+  ) => {
     const { llm, stream, getTools } = step.context;
-    const input = step.input;
-    if (input) {
-      step.context.store.messages.push(input);
-    }
     const lastMessage = step.context.store.messages.at(-1)!.content;
     const tools = await getTools(lastMessage);
     const messages = await chatFormatter(
@@ -367,35 +381,33 @@ export class ReActAgent extends AgentRunner<LLM, ReACTAgentStore> {
       stream,
       messages,
     });
-    const reason = await reACTOutputParser(response);
-    step.context.store.reasons = [...step.context.store.reasons, reason];
-    if (reason.type === "response") {
-      return {
-        isLast: true,
-        output: response,
+    const reason = await reACTOutputParser(response, (type, response) => {
+      enqueueOutput({
         taskStep: step,
-      };
-    } else {
-      if (reason.type === "action") {
-        const tool = tools.find((tool) => tool.metadata.name === reason.action);
-        const toolOutput = await callTool(tool, {
+        output: response,
+        isLast: type !== "action",
+      });
+    });
+    step.context.logger.log("current reason: %O", reason);
+    step.context.store.reasons = [...step.context.store.reasons, reason];
+    if (reason.type === "action") {
+      const tool = tools.find((tool) => tool.metadata.name === reason.action);
+      const toolOutput = await callTool(
+        tool,
+        {
           id: randomUUID(),
           input: reason.input,
           name: reason.action,
-        });
-        step.context.store.reasons = [
-          ...step.context.store.reasons,
-          {
-            type: "observation",
-            observation: toolOutput.output,
-          },
-        ];
-      }
-      return {
-        isLast: false,
-        output: null,
-        taskStep: step,
-      };
+        },
+        step.context.logger,
+      );
+      step.context.store.reasons = [
+        ...step.context.store.reasons,
+        {
+          type: "observation",
+          observation: toolOutput.output,
+        },
+      ];
     }
   };
 }

packages/core/src/agent/types.ts

@@ -1,4 +1,5 @@
 import { ReadableStream } from "@llamaindex/env";
+import type { Logger } from "../internal/logger.js";
 import type { BaseEvent } from "../internal/type.js";
 import type {
   ChatMessage,
@@ -32,6 +33,7 @@ export type AgentTaskContext<
     toolOutputs: ToolOutput[];
     messages: ChatMessage<AdditionalMessageOptions>[];
   } & Store;
+  logger: Readonly<Logger>;
 };
 
 export type TaskStep<
@@ -45,7 +47,6 @@ export type TaskStep<
     : never,
 > = {
   id: UUID;
-  input: ChatMessage<AdditionalMessageOptions> | null;
   context: AgentTaskContext<Model, Store, AdditionalMessageOptions>;
 
   // linked list
@@ -62,22 +63,14 @@ export type TaskStepOutput<
   >
     ? AdditionalMessageOptions
     : never,
-> =
-  | {
-      taskStep: TaskStep<Model, Store, AdditionalMessageOptions>;
-      output:
-        | null
-        | ChatResponse<AdditionalMessageOptions>
-        | ReadableStream<ChatResponseChunk<AdditionalMessageOptions>>;
-      isLast: false;
-    }
-  | {
-      taskStep: TaskStep<Model, Store, AdditionalMessageOptions>;
-      output:
-        | ChatResponse<AdditionalMessageOptions>
-        | ReadableStream<ChatResponseChunk<AdditionalMessageOptions>>;
-      isLast: true;
-    };
+> = {
+  taskStep: TaskStep<Model, Store, AdditionalMessageOptions>;
+  // output shows the response to the user
+  output:
+    | ChatResponse<AdditionalMessageOptions>
+    | ReadableStream<ChatResponseChunk<AdditionalMessageOptions>>;
+  isLast: boolean;
+};
 
 export type TaskHandler<
   Model extends LLM,
@@ -90,7 +83,10 @@ export type TaskHandler<
     : never,
 > = (
   step: TaskStep<Model, Store, AdditionalMessageOptions>,
-) => Promise<TaskStepOutput<Model, Store, AdditionalMessageOptions>>;
+  enqueueOutput: (
+    taskOutput: TaskStepOutput<Model, Store, AdditionalMessageOptions>,
+  ) => void,
+) => Promise<void>;
 
 export type AgentStartEvent = BaseEvent<{
   startStep: TaskStep;

packages/core/src/agent/utils.ts

@@ -1,4 +1,5 @@
 import { ReadableStream } from "@llamaindex/env";
+import type { Logger } from "../internal/logger.js";
 import { getCallbackManager } from "../internal/settings/CallbackManager.js";
 import { isAsyncIterable, prettifyError } from "../internal/utils.js";
 import type {
@@ -13,12 +14,14 @@ import type { BaseTool, JSONObject, JSONValue, ToolOutput } from "../types.js";
 export async function callTool(
   tool: BaseTool | undefined,
   toolCall: ToolCall | PartialToolCall,
+  logger: Logger,
 ): Promise<ToolOutput> {
   const input: JSONObject =
     typeof toolCall.input === "string"
       ? JSON.parse(toolCall.input)
       : toolCall.input;
   if (!tool) {
+    logger.error(`Tool ${toolCall.name} does not exist.`);
     const output = `Tool ${toolCall.name} does not exist.`;
     return {
       tool,
@@ -30,6 +33,9 @@ export async function callTool(
   const call = tool.call;
   let output: JSONValue;
   if (!call) {
+    logger.error(
+      `Tool ${tool.metadata.name} (remote:${toolCall.name}) does not have a implementation.`,
+    );
     output = `Tool ${tool.metadata.name} (remote:${toolCall.name}) does not have a implementation.`;
     return {
       tool,
@@ -45,6 +51,10 @@ export async function callTool(
       },
     });
     output = await call.call(tool, input);
+    logger.log(
+      `Tool ${tool.metadata.name} (remote:${toolCall.name}) succeeded.`,
+    );
+    logger.log(`Output: ${JSON.stringify(output)}`);
     const toolOutput: ToolOutput = {
       tool,
       input,
@@ -60,6 +70,9 @@ export async function callTool(
     return toolOutput;
   } catch (e) {
     output = prettifyError(e);
+    logger.error(
+      `Tool ${tool.metadata.name} (remote:${toolCall.name}) failed: ${output}`,
+    );
   }
   return {
     tool,
@@ -71,16 +84,19 @@ export async function callTool(
 
 export async function consumeAsyncIterable<Options extends object>(
   input: ChatMessage<Options>,
+  previousContent?: string,
 ): Promise<ChatMessage<Options>>;
 export async function consumeAsyncIterable<Options extends object>(
   input: AsyncIterable<ChatResponseChunk<Options>>,
+  previousContent?: string,
 ): Promise<TextChatMessage<Options>>;
 export async function consumeAsyncIterable<Options extends object>(
   input: ChatMessage<Options> | AsyncIterable<ChatResponseChunk<Options>>,
+  previousContent: string = "",
 ): Promise<ChatMessage<Options>> {
   if (isAsyncIterable(input)) {
     const result: ChatMessage<Options> = {
-      content: "",
+      content: previousContent,
       // only assistant will give streaming response
       role: "assistant",
       options: {} as Options,

packages/core/src/callbacks/CallbackManager.ts

@@ -212,10 +212,13 @@ export class CallbackManager implements CallbackManagerMethods {
     if (!handlers) {
       return;
     }
-    const clone = structuredClone(detail);
     queueMicrotask(() => {
       handlers.forEach((handler) =>
-        handler(LlamaIndexCustomEvent.fromEvent(event, clone)),
+        handler(
+          LlamaIndexCustomEvent.fromEvent(event, {
+            ...detail,
+          }),
+        ),
       );
     });
   }

packages/core/src/embeddings/ClipEmbedding.ts

@@ -3,7 +3,10 @@ import type { ImageType } from "../Node.js";
 import { MultiModalEmbedding } from "./MultiModalEmbedding.js";
 
 async function readImage(input: ImageType) {
-  const { RawImage } = await import("@xenova/transformers");
+  const { RawImage } = await import(
+    /* webpackIgnore: true */
+    "@xenova/transformers"
+  );
   if (input instanceof Blob) {
     return await RawImage.fromBlob(input);
   } else if (_.isString(input) || input instanceof URL) {
@@ -29,7 +32,10 @@ export class ClipEmbedding extends MultiModalEmbedding {
 
   async getTokenizer() {
     if (!this.tokenizer) {
-      const { AutoTokenizer } = await import("@xenova/transformers");
+      const { AutoTokenizer } = await import(
+        /* webpackIgnore: true */
+        "@xenova/transformers"
+      );
       this.tokenizer = await AutoTokenizer.from_pretrained(this.modelType);
     }
     return this.tokenizer;
@@ -37,7 +43,10 @@ export class ClipEmbedding extends MultiModalEmbedding {
 
   async getProcessor() {
     if (!this.processor) {
-      const { AutoProcessor } = await import("@xenova/transformers");
+      const { AutoProcessor } = await import(
+        /* webpackIgnore: true */
+        "@xenova/transformers"
+      );
       this.processor = await AutoProcessor.from_pretrained(this.modelType);
     }
     return this.processor;
@@ -46,6 +55,7 @@ export class ClipEmbedding extends MultiModalEmbedding {
   async getVisionModel() {
     if (!this.visionModel) {
       const { CLIPVisionModelWithProjection } = await import(
+        /* webpackIgnore: true */
         "@xenova/transformers"
       );
       this.visionModel = await CLIPVisionModelWithProjection.from_pretrained(
@@ -59,6 +69,7 @@ export class ClipEmbedding extends MultiModalEmbedding {
   async getTextModel() {
     if (!this.textModel) {
       const { CLIPTextModelWithProjection } = await import(
+        /* webpackIgnore: true */
         "@xenova/transformers"
       );
       this.textModel = await CLIPTextModelWithProjection.from_pretrained(

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

@@ -13,6 +13,11 @@ export interface ChatEngineParamsBase {
    * Optional chat history if you want to customize the chat history.
    */
   chatHistory?: ChatMessage[] | ChatHistory;
+  /**
+   * Optional flag to enable verbose mode.
+   * @default false
+   */
+  verbose?: boolean;
 }
 
 export interface ChatEngineParamsStreaming extends ChatEngineParamsBase {

packages/core/src/index.edge.ts

@@ -27,6 +27,7 @@ export * from "./objects/index.js";
 export * from "./postprocessors/index.js";
 export * from "./prompts/index.js";
 export * from "./selectors/index.js";
+export * from "./storage/StorageContext.js";
 export * from "./synthesizers/index.js";
 export * from "./tools/index.js";
 export * from "./types.js";

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

@@ -279,18 +279,29 @@ export class VectorStoreIndex extends BaseIndex<IndexDict> {
     return new VectorIndexRetriever({ index: this, ...options });
   }
 
+  /**
+   * Create a RetrieverQueryEngine.
+   * similarityTopK is only used if no existing retriever is provided.
+   */
   asQueryEngine(options?: {
     retriever?: BaseRetriever;
     responseSynthesizer?: BaseSynthesizer;
     preFilters?: MetadataFilters;
     nodePostprocessors?: BaseNodePostprocessor[];
+    similarityTopK?: number;
   }): QueryEngine & RetrieverQueryEngine {
-    const { retriever, responseSynthesizer } = options ?? {};
-    return new RetrieverQueryEngine(
-      retriever ?? this.asRetriever(),
+    const {
+      retriever,
       responseSynthesizer,
-      options?.preFilters,
-      options?.nodePostprocessors,
+      preFilters,
+      nodePostprocessors,
+      similarityTopK,
+    } = options ?? {};
+    return new RetrieverQueryEngine(
+      retriever ?? this.asRetriever({ similarityTopK }),
+      responseSynthesizer,
+      preFilters,
+      nodePostprocessors,
     );
   }
 

packages/core/src/internal/logger.ts

@@ -0,0 +1,17 @@
+export type Logger = {
+  log: (...args: unknown[]) => void;
+  error: (...args: unknown[]) => void;
+  warn: (...args: unknown[]) => void;
+};
+
+export const emptyLogger: Logger = Object.freeze({
+  log: () => {},
+  error: () => {},
+  warn: () => {},
+});
+
+export const consoleLogger: Logger = Object.freeze({
+  log: console.log.bind(console),
+  error: console.error.bind(console),
+  warn: console.warn.bind(console),
+});

packages/core/src/llm/gemini.ts

@@ -302,12 +302,10 @@ export class Gemini extends ToolCallLLM<GeminiAdditionalChatOptions> {
   ): GeminiChatStreamResponse {
     const { chat, messageContent } = this.prepareChat(params);
     const result = await chat.sendMessageStream(messageContent);
-    return streamConverter(result.stream, (response) => {
-      return {
-        text: response.text(),
-        raw: response,
-      };
-    });
+    yield* streamConverter(result.stream, (response) => ({
+      delta: response.text(),
+      raw: response,
+    }));
   }
 
   chat(params: GeminiChatParamsStreaming): Promise<GeminiChatStreamResponse>;

packages/core/src/llm/huggingface.ts

@@ -0,0 +1,141 @@
+import {
+  HfInference,
+  type Options as HfInferenceOptions,
+} from "@huggingface/inference";
+import { BaseLLM } from "./base.js";
+import type {
+  ChatMessage,
+  ChatResponse,
+  ChatResponseChunk,
+  LLMChatParamsNonStreaming,
+  LLMChatParamsStreaming,
+  LLMMetadata,
+  ToolCallLLMMessageOptions,
+} from "./types.js";
+import { streamConverter, wrapLLMEvent } from "./utils.js";
+
+const DEFAULT_PARAMS = {
+  temperature: 0.1,
+  topP: 1,
+  maxTokens: undefined,
+  contextWindow: 3900,
+};
+export type HFConfig = Partial<typeof DEFAULT_PARAMS> &
+  HfInferenceOptions & {
+    model: string;
+    accessToken: string;
+    endpoint?: string;
+  };
+
+/**
+    Wrapper on the Hugging Face's Inference API.
+    API Docs: https://huggingface.co/docs/huggingface.js/inference/README
+    List of tasks with models: huggingface.co/api/tasks 
+    
+    Note that Conversational API is not yet supported by the Inference API.
+    They recommend using the text generation API instead.
+    See: https://github.com/huggingface/huggingface.js/issues/586#issuecomment-2024059308
+ */
+export class HuggingFaceInferenceAPI extends BaseLLM {
+  model: string;
+  temperature: number;
+  topP: number;
+  maxTokens?: number;
+  contextWindow: number;
+  hf: HfInference;
+
+  constructor(init: HFConfig) {
+    super();
+    const {
+      model,
+      temperature,
+      topP,
+      maxTokens,
+      contextWindow,
+      accessToken,
+      endpoint,
+      ...hfInferenceOpts
+    } = init;
+    this.hf = new HfInference(accessToken, hfInferenceOpts);
+    this.model = model;
+    this.temperature = temperature ?? DEFAULT_PARAMS.temperature;
+    this.topP = topP ?? DEFAULT_PARAMS.topP;
+    this.maxTokens = maxTokens ?? DEFAULT_PARAMS.maxTokens;
+    this.contextWindow = contextWindow ?? DEFAULT_PARAMS.contextWindow;
+    if (endpoint) this.hf.endpoint(endpoint);
+  }
+
+  get metadata(): LLMMetadata {
+    return {
+      model: this.model,
+      temperature: this.temperature,
+      topP: this.topP,
+      maxTokens: this.maxTokens,
+      contextWindow: this.contextWindow,
+      tokenizer: undefined,
+    };
+  }
+
+  chat(
+    params: LLMChatParamsStreaming,
+  ): Promise<AsyncIterable<ChatResponseChunk>>;
+  chat(params: LLMChatParamsNonStreaming): Promise<ChatResponse>;
+  @wrapLLMEvent
+  async chat(
+    params: LLMChatParamsStreaming | LLMChatParamsNonStreaming,
+  ): Promise<AsyncIterable<ChatResponseChunk> | ChatResponse<object>> {
+    if (params.stream) return this.streamChat(params);
+    return this.nonStreamChat(params);
+  }
+
+  private messagesToPrompt(messages: ChatMessage<ToolCallLLMMessageOptions>[]) {
+    let prompt = "";
+    for (const message of messages) {
+      if (message.role === "system") {
+        prompt += `<|system|>\n${message.content}</s>\n`;
+      } else if (message.role === "user") {
+        prompt += `<|user|>\n${message.content}</s>\n`;
+      } else if (message.role === "assistant") {
+        prompt += `<|assistant|>\n${message.content}</s>\n`;
+      }
+    }
+    // ensure we start with a system prompt, insert blank if needed
+    if (!prompt.startsWith("<|system|>\n")) {
+      prompt = "<|system|>\n</s>\n" + prompt;
+    }
+    // add final assistant prompt
+    prompt = prompt + "<|assistant|>\n";
+    return prompt;
+  }
+
+  protected async nonStreamChat(
+    params: LLMChatParamsNonStreaming,
+  ): Promise<ChatResponse> {
+    const res = await this.hf.textGeneration({
+      model: this.model,
+      inputs: this.messagesToPrompt(params.messages),
+      parameters: this.metadata,
+    });
+    return {
+      raw: res,
+      message: {
+        content: res.generated_text,
+        role: "assistant",
+      },
+    };
+  }
+
+  protected async *streamChat(
+    params: LLMChatParamsStreaming,
+  ): AsyncIterable<ChatResponseChunk> {
+    const stream = this.hf.textGenerationStream({
+      model: this.model,
+      inputs: this.messagesToPrompt(params.messages),
+      parameters: this.metadata,
+    });
+    yield* streamConverter(stream, (chunk) => ({
+      delta: chunk.token.text,
+      raw: chunk,
+    }));
+  }
+}

packages/core/src/llm/index.ts

@@ -7,6 +7,7 @@ export {
 export { FireworksLLM } from "./fireworks.js";
 export { GEMINI_MODEL, Gemini } from "./gemini.js";
 export { Groq } from "./groq.js";
+export { HuggingFaceInferenceAPI } from "./huggingface.js";
 export {
   ALL_AVAILABLE_MISTRAL_MODELS,
   MistralAI,

packages/core/src/objects/base.ts

@@ -1,7 +1,7 @@
 import type { BaseNode, Metadata } from "../Node.js";
 import { TextNode } from "../Node.js";
 import type { BaseRetriever } from "../Retriever.js";
-import type { VectorStoreIndex } from "../indices/index.js";
+import type { VectorStoreIndex } from "../indices/vectorStore/index.js";
 import type { MessageContent } from "../llm/index.js";
 import { extractText } from "../llm/utils.js";
 import type { BaseTool } from "../types.js";

packages/edge/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/edge",
-  "version": "0.3.0",
+  "version": "0.3.4",
   "license": "MIT",
   "type": "module",
   "dependencies": {
@@ -9,6 +9,7 @@
     "@datastax/astra-db-ts": "^1.0.1",
     "@google/generative-ai": "^0.8.0",
     "@grpc/grpc-js": "^1.10.6",
+    "@huggingface/inference": "^2.6.7",
     "@llamaindex/cloud": "0.0.5",
     "@llamaindex/env": "workspace:*",
     "@mistralai/mistralai": "^0.1.3",

packages/env/jsr.json

@@ -1,6 +1,6 @@
 {
   "name": "@llamaindex/env",
-  "version": "0.0.6",
+  "version": "0.1.0",
   "exports": {
     ".": "./src/index.ts",
     "./type": "./src/type.ts"

packages/experimental/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @llamaindex/experimental
 
+## 0.0.21
+
+### Patch Changes
+
+- Updated dependencies [1dce275]
+- Updated dependencies [d10533e]
+- Updated dependencies [2008efe]
+- Updated dependencies [5e61934]
+- Updated dependencies [9e74a43]
+- Updated dependencies [ee719a1]
+  - llamaindex@0.3.4
+
+## 0.0.20
+
+### Patch Changes
+
+- Updated dependencies [e8c41c5]
+  - llamaindex@0.3.3
+
+## 0.0.19
+
+### Patch Changes
+
+- Updated dependencies [61103b6]
+  - llamaindex@0.3.2
+
+## 0.0.18
+
+### Patch Changes
+
+- Updated dependencies [46227f2]
+  - llamaindex@0.3.1
+
 ## 0.0.17
 
 ### Patch Changes

packages/experimental/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@llamaindex/experimental",
   "description": "Experimental package for LlamaIndexTS",
-  "version": "0.0.17",
+  "version": "0.0.21",
   "type": "module",
   "types": "dist/type/index.d.ts",
   "main": "dist/cjs/index.js",

scripts/check-minor-version.js

@@ -26,6 +26,23 @@ if (minorVersion !== expectedMinorVersion) {
   process.exit(1);
 }
 
+const packages = ["core", "env"];
+for (const pkg of packages) {
+  const packageJson = JSON.parse(
+    fs.readFileSync(`./packages/${pkg}/package.json`, "utf8"),
+  );
+  const jsrJson = JSON.parse(
+    fs.readFileSync(`./packages/${pkg}/jsr.json`, "utf8"),
+  );
+
+  jsrJson.version = packageJson.version;
+
+  fs.writeFileSync(
+    `./packages/${pkg}/jsr.json`,
+    JSON.stringify(jsrJson, null, 2) + "\n",
+  );
+}
+
 console.log("Current expected minor version is: " + expectedMinorVersion);
 console.log("Minor version is: " + minorVersion);
 console.log("Good to go!");