From 98ba1e71bd28a8a8019f03d298501986a6f727cb Mon Sep 17 00:00:00 2001
From: Erik <erik.balfe@protonmail.com>
Date: Sat, 26 Oct 2024 18:23:10 +0400
Subject: [PATCH] feat: implement context-aware agent functionality. (#1394)
Co-authored-by: Alex Yang <himself65@outlook.com>
---
.changeset/tough-oranges-itch.md | 6 ++
.../docs/examples/context_aware_agent.mdx | 63 ++++++++++++++
.../docs/docs/guides/agents/4_agentic_rag.mdx | 87 ++++++++-----------
examples/contextAwareAgent.js | 55 ++++++++++++
packages/core/src/agent/index.ts | 7 +-
packages/llamaindex/src/agent/anthropic.ts | 6 ++
.../llamaindex/src/agent/contextAwareMixin.ts | 85 ++++++++++++++++++
packages/llamaindex/src/agent/index.ts | 12 +--
packages/llamaindex/src/agent/openai.ts | 7 ++
9 files changed, 273 insertions(+), 55 deletions(-)
create mode 100644 .changeset/tough-oranges-itch.md
create mode 100644 apps/docs/docs/examples/context_aware_agent.mdx
create mode 100644 examples/contextAwareAgent.js
create mode 100644 packages/llamaindex/src/agent/contextAwareMixin.ts
create mode 100644 packages/llamaindex/src/agent/openai.ts
diff --git a/.changeset/tough-oranges-itch.md b/.changeset/tough-oranges-itch.md
new file mode 100644
index 000000000..a3dc7acbc
--- /dev/null
+++ b/.changeset/tough-oranges-itch.md
@@ -0,0 +1,6 @@
+---
+"llamaindex": minor
+"@llamaindex/core": minor
+---
+
+fea:t implement context-aware agent
diff --git a/apps/docs/docs/examples/context_aware_agent.mdx b/apps/docs/docs/examples/context_aware_agent.mdx
new file mode 100644
index 000000000..2a5fe4f68
--- /dev/null
+++ b/apps/docs/docs/examples/context_aware_agent.mdx
@@ -0,0 +1,63 @@
+---
+sidebar_position: 14
+---
+
+# Context-Aware Agent
+
+The Context-Aware Agent enhances the capabilities of standard LLM agents by incorporating relevant context from a retriever for each query. This allows the agent to provide more informed and specific responses based on the available information.
+
+## Usage
+
+Here's a simple example of how to use the Context-Aware Agent:
+
+```typescript
+import {
+ Document,
+ VectorStoreIndex,
+ OpenAIContextAwareAgent,
+ OpenAI,
+} from "llamaindex";
+
+async function createContextAwareAgent() {
+ // Create and index some documents
+ const documents = [
+ new Document({
+ text: "LlamaIndex is a data framework for LLM applications.",
+ id_: "doc1",
+ }),
+ new Document({
+ text: "The Eiffel Tower is located in Paris, France.",
+ id_: "doc2",
+ }),
+ ];
+
+ const index = await VectorStoreIndex.fromDocuments(documents);
+ const retriever = index.asRetriever({ similarityTopK: 1 });
+
+ // Create the Context-Aware Agent
+ const agent = new OpenAIContextAwareAgent({
+ llm: new OpenAI({ model: "gpt-3.5-turbo" }),
+ contextRetriever: retriever,
+ });
+
+ // Use the agent to answer queries
+ const response = await agent.chat({
+ message: "What is LlamaIndex used for?",
+ });
+
+ console.log("Agent Response:", response.response);
+}
+
+createContextAwareAgent().catch(console.error);
+```
+
+In this example, the Context-Aware Agent uses the retriever to fetch relevant context for each query, allowing it to provide more accurate and informed responses based on the indexed documents.
+
+## Key Components
+
+- `contextRetriever`: A retriever (e.g., from a VectorStoreIndex) that fetches relevant documents or passages for each query.
+
+## Available Context-Aware Agents
+
+- `OpenAIContextAwareAgent`: A context-aware agent using OpenAI's models.
+- `AnthropicContextAwareAgent`: A context-aware agent using Anthropic's models.
diff --git a/apps/docs/docs/guides/agents/4_agentic_rag.mdx b/apps/docs/docs/guides/agents/4_agentic_rag.mdx
index d4dfce695..d4a9f41a4 100644
--- a/apps/docs/docs/guides/agents/4_agentic_rag.mdx
+++ b/apps/docs/docs/guides/agents/4_agentic_rag.mdx
@@ -4,22 +4,23 @@ While an agent that can perform math is nifty (LLMs are usually not very good at
To learn more about RAG, we recommend this [introduction](https://docs.llamaindex.ai/en/stable/getting_started/concepts/) from our Python docs. We'll assume you know the basics:
-- You need to parse your source data into chunks of text
-- You need to encode that text as numbers, called embeddings
-- You need to search your embeddings for the most relevant chunks of text
-- You feed your relevant chunks and a query to an LLM to answer a question
+- Parse your source data into chunks of text.
+- Encode that text as numbers, called embeddings.
+- Search your embeddings for the most relevant chunks of text.
+- Use the relevant chunks along with a query to ask an LLM to generate an answer.
We're going to start with the same agent we [built in step 1](https://github.com/run-llama/ts-agents/blob/main/1_agent/agent.ts), but make a few changes. You can find the finished version [in the repository](https://github.com/run-llama/ts-agents/blob/main/2_agentic_rag/agent.ts).
### New dependencies
-We'll be bringing in `SimpleDirectoryReader`, `HuggingFaceEmbedding`, `VectorStoreIndex`, and `QueryEngineTool` from LlamaIndex.TS, as well as the dependencies we previously used.
+We'll be bringing in `SimpleDirectoryReader`, `HuggingFaceEmbedding`, `VectorStoreIndex`, and `QueryEngineTool`, `OpenAIContextAwareAgent` from LlamaIndex.TS, as well as the dependencies we previously used.
```javascript
import {
OpenAI,
FunctionTool,
OpenAIAgent,
+ OpenAIContextAwareAgent,
Settings,
SimpleDirectoryReader,
HuggingFaceEmbedding,
@@ -41,7 +42,7 @@ Settings.embedModel = new HuggingFaceEmbedding({
### Load data using SimpleDirectoryReader
-SimpleDirectoryReader is a flexible tool that can read a variety of file formats. We're going to point it at our data directory, which contains just the single PDF file, and get it to return a set of documents.
+`SimpleDirectoryReader` is a flexible tool that can read various file formats. We will point it at our data directory, which contains a single PDF file, and retrieve a set of documents.
```javascript
const reader = new SimpleDirectoryReader();
@@ -50,7 +51,7 @@ const documents = await reader.loadData("../data");
### Index our data
-Now we turn our text into embeddings. The `VectorStoreIndex` class takes care of this for us when we use the `fromDocuments` method (it uses the embedding model we defined in `Settings` earlier).
+We will convert our text into embeddings using the `VectorStoreIndex` class through the `fromDocuments` method, which utilizes the embedding model defined earlier in `Settings`.
```javascript
const index = await VectorStoreIndex.fromDocuments(documents);
@@ -72,21 +73,35 @@ By default LlamaIndex will retrieve just the 2 most relevant chunks of text. Thi
retriever.similarityTopK = 10;
```
-### Create a query engine
+### Approach 1: Create a Context-Aware Agent
-And our final step in creating a RAG pipeline is to create a query engine that will use the retriever to find the most relevant chunks of text, and then use the LLM to answer the question.
+With the retriever ready, you can create a **context-aware agent**.
```javascript
-const queryEngine = await index.asQueryEngine({
- retriever,
+const agent = new OpenAIContextAwareAgent({
+ contextRetriever: retriever,
});
+
+// Example query to the context-aware agent
+let response = await agent.chat({
+ message: `What's the budget of San Francisco in 2023-2024?`,
+});
+
+console.log(response);
```
-### Define the query engine as a tool
+**Expected Output:**
+
+```md
+The total budget for the City and County of San Francisco for the fiscal year 2023-2024 is $14.6 billion. This represents a $611.8 million, or 4.4 percent, increase over the previous fiscal year's budget. The budget covers various expenditures across different departments and services, including significant allocations to public works, transportation, commerce, public protection, and health services.
+```
-Just as before we created a `FunctionTool`, we're going to create a `QueryEngineTool` that uses our `queryEngine`.
+### Approach 2: Using QueryEngineTool (Alternative Approach)
+
+If you prefer more flexibility and don't mind additional complexity, you can create a `QueryEngineTool`. This approach allows you to define the query logic, providing a more tailored way to interact with the data, but note that it introduces a delay due to the extra tool call.
```javascript
+const queryEngine = await index.asQueryEngine({ retriever });
const tools = [
new QueryEngineTool({
queryEngine: queryEngine,
@@ -96,28 +111,18 @@ const tools = [
},
}),
];
-```
-
-As before, we've created an array of tools with just one tool in it. The metadata is slightly different: we don't need to define our parameters, we just give the tool a name and a natural-language description.
-### Create the agent as before
-
-Creating the agent and asking a question is exactly the same as before, but we'll ask a different question.
-
-```javascript
-// create the agent
+// Create an agent using the tools array
const agent = new OpenAIAgent({ tools });
-let response = await agent.chat({
+let toolResponse = await agent.chat({
message: "What's the budget of San Francisco in 2023-2024?",
});
-console.log(response);
+console.log(toolResponse);
```
-Once again we'll run `npx tsx agent.ts` and see what we get:
-
-**_Output_**
+**Expected Output:**
```javascript
{
@@ -138,28 +143,12 @@ Once again we'll run `npx tsx agent.ts` and see what we get:
}
```
-```javascript
-{
- response: {
- raw: {
- id: 'chatcmpl-9KxUkwizVCYCmxwFQcZFSHrInzNFU',
- object: 'chat.completion',
- created: 1714782286,
- model: 'gpt-4-turbo-2024-04-09',
- choices: [Array],
- usage: [Object],
- system_fingerprint: 'fp_ea6eb70039'
- },
- message: {
- content: "The total budget for the City and County of San Francisco for the fiscal year 2023-2024 is $14.6 billion. This represents a $611.8 million, or 4.4 percent, increase over the previous fiscal year's budget. The budget covers various expenditures across different departments and services, including significant allocations to public works, transportation, commerce, public protection, and health services.",
- role: 'assistant',
- options: {}
- }
- },
- sources: [Getter]
-}
-```
-
Once again we see a `toolResult`. You can see the query the LLM decided to send to the query engine ("total budget"), and the output the engine returned. In `response.message` you see that the LLM has returned the output from the tool almost verbatim, although it trimmed out the bit about 2024-2025 since we didn't ask about that year.
+### Comparison of Approaches
+
+The `OpenAIContextAwareAgent` approach simplifies the setup by allowing you to directly link the retriever to the agent, making it straightforward to access relevant context for your queries. This is ideal for situations where you want easy integration with existing data sources, like a context chat engine.
+
+On the other hand, using the `QueryEngineTool` offers more flexibility and power. This method allows for customization in how queries are constructed and executed, enabling you to query data from various storages and process them in different ways. However, this added flexibility comes with increased complexity and response time due to the separate tool call and queryEngine generating tool output by LLM that is then passed to the agent.
+
So now we have an agent that can index complicated documents and answer questions about them. Let's [combine our math agent and our RAG agent](rag_and_tools)!
diff --git a/examples/contextAwareAgent.js b/examples/contextAwareAgent.js
new file mode 100644
index 000000000..f63073e47
--- /dev/null
+++ b/examples/contextAwareAgent.js
@@ -0,0 +1,55 @@
+import {
+ Document,
+ OpenAI,
+ OpenAIContextAwareAgent,
+ VectorStoreIndex,
+} from "llamaindex";
+
+import dotenv from "dotenv";
+dotenv.config();
+
+async function createTestContextAwareAgent() {
+ // Create test documents
+ const testDocument1 = new Document({
+ text: "LlamaIndex is a data framework for LLM applications to ingest, structure, and access private or domain-specific data.",
+ id_: "doc1",
+ });
+
+ const testDocument2 = new Document({
+ text: "The Eiffel Tower is a wrought-iron lattice tower on the Champ de Mars in Paris, France. It is named after the engineer Gustave Eiffel, whose company designed and built the tower.",
+ id_: "doc2",
+ });
+
+ const testDocument3 = new Document({
+ text: "Photosynthesis is the process by which green plants and some other organisms use sunlight to synthesize foods with the help of chlorophyll pigments.",
+ id_: "doc3",
+ });
+
+ // Create a test index
+ const testIndex = await VectorStoreIndex.fromDocuments([
+ testDocument1,
+ testDocument2,
+ testDocument3,
+ ]);
+
+ // Create a retriever from the index to get only 1 relevant document
+ const testRetriever = testIndex.asRetriever({
+ similarityTopK: 1,
+ });
+
+ // Create an OpenAI Context-Aware Agent with the retriever
+ const contextAwareAgent = new OpenAIContextAwareAgent({
+ llm: new OpenAI({ model: "gpt-4o-mini" }),
+ tools: [],
+ contextRetriever: testRetriever,
+ });
+
+ // Test the agent with a query that should trigger relevant document retrieval
+ const response = await contextAwareAgent.chat({
+ message: "What is LlamaIndex used for?",
+ });
+
+ console.log("Context-aware Agent Response:", response.response);
+}
+
+createTestContextAwareAgent().catch(console.error);
diff --git a/packages/core/src/agent/index.ts b/packages/core/src/agent/index.ts
index 1599230a5..b02590f31 100644
--- a/packages/core/src/agent/index.ts
+++ b/packages/core/src/agent/index.ts
@@ -1,6 +1,11 @@
export { AgentRunner, AgentWorker, type AgentParamsBase } from "./base.js";
export { LLMAgent, LLMAgentWorker, type LLMAgentParams } from "./llm.js";
-export type { AgentEndEvent, AgentStartEvent, TaskHandler } from "./types.js";
+export type {
+ AgentEndEvent,
+ AgentStartEvent,
+ TaskHandler,
+ TaskStep,
+} from "./types.js";
export {
callTool,
consumeAsyncIterable,
diff --git a/packages/llamaindex/src/agent/anthropic.ts b/packages/llamaindex/src/agent/anthropic.ts
index e240a27e2..75f1f34b8 100644
--- a/packages/llamaindex/src/agent/anthropic.ts
+++ b/packages/llamaindex/src/agent/anthropic.ts
@@ -1 +1,7 @@
+import { AnthropicAgent } from "@llamaindex/anthropic";
+import { withContextAwareness } from "./contextAwareMixin.js";
+
+export const AnthropicContextAwareAgent = withContextAwareness(AnthropicAgent);
+export type { ContextAwareConfig } from "./contextAwareMixin.js";
+
export * from "@llamaindex/anthropic";
diff --git a/packages/llamaindex/src/agent/contextAwareMixin.ts b/packages/llamaindex/src/agent/contextAwareMixin.ts
new file mode 100644
index 000000000..0553f79ee
--- /dev/null
+++ b/packages/llamaindex/src/agent/contextAwareMixin.ts
@@ -0,0 +1,85 @@
+import {
+ AnthropicAgent,
+ type AnthropicAgentParams,
+} from "@llamaindex/anthropic";
+import type {
+ NonStreamingChatEngineParams,
+ StreamingChatEngineParams,
+} from "@llamaindex/core/chat-engine";
+import type { MessageContent } from "@llamaindex/core/llms";
+import type { BaseRetriever } from "@llamaindex/core/retriever";
+import { EngineResponse, MetadataMode } from "@llamaindex/core/schema";
+import { OpenAIAgent, type OpenAIAgentParams } from "@llamaindex/openai";
+
+export interface ContextAwareConfig {
+ contextRetriever: BaseRetriever;
+}
+
+export interface ContextAwareState {
+ contextRetriever: BaseRetriever;
+ retrievedContext: string | null;
+}
+
+export type SupportedAgent = typeof OpenAIAgent | typeof AnthropicAgent;
+export type AgentParams<T> = T extends typeof OpenAIAgent
+ ? OpenAIAgentParams
+ : T extends typeof AnthropicAgent
+ ? AnthropicAgentParams
+ : never;
+
+/**
+ * ContextAwareAgentRunner enhances the base AgentRunner with the ability to retrieve and inject relevant context
+ * for each query. This allows the agent to access and utilize appropriate information from a given index or retriever,
+ * providing more informed and context-specific responses to user queries.
+ */
+export function withContextAwareness<T extends SupportedAgent>(Base: T) {
+ return class ContextAwareAgent extends Base {
+ public readonly contextRetriever: BaseRetriever;
+ public retrievedContext: string | null = null;
+ public declare chatHistory: T extends typeof OpenAIAgent
+ ? OpenAIAgent["chatHistory"]
+ : T extends typeof AnthropicAgent
+ ? AnthropicAgent["chatHistory"]
+ : never;
+
+ constructor(params: AgentParams<T> & ContextAwareConfig) {
+ super(params);
+ this.contextRetriever = params.contextRetriever;
+ }
+
+ async retrieveContext(query: MessageContent): Promise<string> {
+ const nodes = await this.contextRetriever.retrieve({ query });
+ return nodes
+ .map((node) => node.node.getContent(MetadataMode.NONE))
+ .join("\n");
+ }
+
+ async injectContext(context: string): Promise<void> {
+ const systemMessage = this.chatHistory.find(
+ (msg) => msg.role === "system",
+ );
+ if (systemMessage) {
+ systemMessage.content = `${context}\n\n${systemMessage.content}`;
+ } else {
+ this.chatHistory.unshift({ role: "system", content: context });
+ }
+ }
+
+ async chat(params: NonStreamingChatEngineParams): Promise<EngineResponse>;
+ async chat(
+ params: StreamingChatEngineParams,
+ ): Promise<ReadableStream<EngineResponse>>;
+ async chat(
+ params: NonStreamingChatEngineParams | StreamingChatEngineParams,
+ ): Promise<EngineResponse | ReadableStream<EngineResponse>> {
+ const context = await this.retrieveContext(params.message);
+ await this.injectContext(context);
+
+ if ("stream" in params && params.stream === true) {
+ return super.chat(params);
+ } else {
+ return super.chat(params as NonStreamingChatEngineParams);
+ }
+ }
+ };
+}
diff --git a/packages/llamaindex/src/agent/index.ts b/packages/llamaindex/src/agent/index.ts
index 3473f0c5f..8d9790f26 100644
--- a/packages/llamaindex/src/agent/index.ts
+++ b/packages/llamaindex/src/agent/index.ts
@@ -1,14 +1,16 @@
export * from "@llamaindex/core/agent";
-export {
- OpenAIAgent,
- OpenAIAgentWorker,
- type OpenAIAgentParams,
-} from "@llamaindex/openai";
export {
AnthropicAgent,
AnthropicAgentWorker,
+ AnthropicContextAwareAgent,
type AnthropicAgentParams,
} from "./anthropic.js";
+export {
+ OpenAIAgent,
+ OpenAIAgentWorker,
+ OpenAIContextAwareAgent,
+ type OpenAIAgentParams,
+} from "./openai.js";
export {
ReACTAgentWorker,
ReActAgent,
diff --git a/packages/llamaindex/src/agent/openai.ts b/packages/llamaindex/src/agent/openai.ts
new file mode 100644
index 000000000..ded98b654
--- /dev/null
+++ b/packages/llamaindex/src/agent/openai.ts
@@ -0,0 +1,7 @@
+import { OpenAIAgent } from "@llamaindex/openai";
+import { withContextAwareness } from "./contextAwareMixin.js";
+
+export const OpenAIContextAwareAgent = withContextAwareness(OpenAIAgent);
+export type { ContextAwareConfig } from "./contextAwareMixin.js";
+
+export * from "@llamaindex/openai";
--
GitLab