feat(agent): Implement URL summarization agent

This commit is contained in:
Willie Zutz 2025-06-29 13:59:52 -06:00
parent d66300e78e
commit f88c650704
6 changed files with 347 additions and 2 deletions

View file

@ -66,4 +66,12 @@ export const AgentState = Annotation.Root({
reducer: (x, y) => y ?? x,
default: () => 'webSearch',
}),
urlsToSummarize: Annotation<string[]>({
reducer: (x, y) => y ?? x,
default: () => [],
}),
summarizationIntent: Annotation<string>({
reducer: (x, y) => y ?? x,
default: () => '',
}),
});

View file

@ -80,6 +80,38 @@ export class AnalyzerAgent {
state.originalQuery = state.query;
}
// Check for URLs first - if found and not yet processed, route to URL summarization
if (!state.urlsToSummarize || state.urlsToSummarize.length === 0) {
const urlRegex = /https?:\/\/[^\s]+/gi;
const urls = [...new Set(state.query.match(urlRegex) || [])];
if (urls.length > 0) {
console.log('URLs detected in initial query, routing to URL summarization');
console.log(`URLs found: ${urls.join(', ')}`);
// Emit URL detection event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URLS_DETECTED_ROUTING',
message: `Detected ${urls.length} URL(s) in query - processing content first`,
details: {
query: state.query,
urls: urls,
},
},
});
return new Command({
goto: 'url_summarization',
update: {
urlsToSummarize: urls,
summarizationIntent: `Process the content from the provided URLs to help answer: ${state.query}`,
},
});
}
}
// Skip full analysis if this is the first run.
//if (state.fullAnalysisAttempts > 0) {
// Emit initial analysis event
@ -108,7 +140,7 @@ export class AnalyzerAgent {
context: state.relevantDocuments
.map(
(doc, index) =>
`<source${index + 1}>${doc?.metadata?.title ? `<title>${doc?.metadata?.title}</title>` : ''}<content>${doc.pageContent}</content></source${index + 1}>`,
`<source${index + 1}>${doc?.metadata?.title ? `<title>${doc?.metadata?.title}</title>` : ''}${doc?.metadata.url ? `<url>${doc?.metadata?.url}</url>` : ''}<content>${doc.pageContent}</content></source${index + 1}>`,
)
.join('\n\n'),
date: formatDateForLLM(new Date()),

View file

@ -5,3 +5,4 @@ export { SynthesizerAgent } from './synthesizerAgent';
export { TaskManagerAgent } from './taskManagerAgent';
export { FileSearchAgent } from './fileSearchAgent';
export { ContentRouterAgent } from './contentRouterAgent';
export { URLSummarizationAgent } from './urlSummarizationAgent';

View file

@ -0,0 +1,288 @@
import { BaseChatModel } from '@langchain/core/language_models/chat_models';
import { AIMessage } from '@langchain/core/messages';
import { Command, END } from '@langchain/langgraph';
import { EventEmitter } from 'events';
import { Document } from 'langchain/document';
import { AgentState } from './agentState';
import { getWebContent } from '../utils/documents';
import { removeThinkingBlocks } from '../utils/contentUtils';
import { setTemperature } from '../utils/modelUtils';
export class URLSummarizationAgent {
private llm: BaseChatModel;
private emitter: EventEmitter;
private systemInstructions: string;
private signal: AbortSignal;
constructor(
llm: BaseChatModel,
emitter: EventEmitter,
systemInstructions: string,
signal: AbortSignal,
) {
this.llm = llm;
this.emitter = emitter;
this.systemInstructions = systemInstructions;
this.signal = signal;
}
/**
* URL processing agent node
*/
async execute(state: typeof AgentState.State): Promise<Command> {
try {
setTemperature(this.llm, 0); // Set temperature to 0 for deterministic output
// Use pre-analyzed URLs from ContentRouterAgent
const urlsToProcess = state.urlsToSummarize || [];
const summarizationIntent = state.summarizationIntent || 'process content to help answer the user query';
if (urlsToProcess.length === 0) {
console.log('No URLs found for processing, routing back to content router');
return new Command({
goto: 'content_router',
update: {
messages: [new AIMessage('No URLs found for processing, routing to content router')],
},
});
}
console.log(`URL processing detected. URLs: ${urlsToProcess.join(', ')}`);
console.log(`Processing intent: ${summarizationIntent}`);
// Emit URL detection event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_PROCESSING_DETECTED',
message: `Processing ${urlsToProcess.length} URL(s) to extract content for analysis`,
details: {
query: state.query,
urls: urlsToProcess,
intent: summarizationIntent,
},
},
});
const documents: Document[] = [];
// Process each URL
for (const url of urlsToProcess) {
if (this.signal.aborted) {
console.warn('URL summarization operation aborted by signal');
break;
}
try {
// Emit URL processing event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'PROCESSING_URL',
message: `Retrieving and processing content from: ${url}`,
details: {
query: state.query,
sourceUrl: url,
intent: summarizationIntent,
},
},
});
// Fetch full content using the enhanced web content retrieval
const webContent = await getWebContent(url, true);
if (!webContent || !webContent.pageContent) {
console.warn(`No content retrieved from URL: ${url}`);
// Emit URL processing failure event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_PROCESSING_FAILED',
message: `Failed to retrieve content from: ${url}`,
details: {
query: state.query,
sourceUrl: url,
reason: 'No content retrieved',
},
},
});
continue;
}
const contentLength = webContent.pageContent.length;
let finalContent: string;
let processingType: string;
// If content is short (< 4000 chars), use it directly; otherwise summarize
if (contentLength < 4000) {
finalContent = webContent.pageContent;
processingType = 'url-direct-content';
console.log(`Content is short (${contentLength} chars), using directly without summarization`);
// Emit direct content usage event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_DIRECT_CONTENT',
message: `Content is short (${contentLength} chars), using directly from: ${url}`,
details: {
query: state.query,
sourceUrl: url,
sourceTitle: webContent.metadata.title || 'Web Page',
contentLength: contentLength,
intent: summarizationIntent,
},
},
});
} else {
// Content is long, summarize using LLM
console.log(`Content is long (${contentLength} chars), generating summary`);
const systemPrompt = this.systemInstructions
? `${this.systemInstructions}\n\n`
: '';
const summarizationPrompt = `${systemPrompt}You are a web content processor. Extract and summarize ONLY the information from the provided web page content that is relevant to the user's query.
# Critical Instructions
- Output ONLY a summary of the web page content provided below
- Focus on information that relates to or helps answer the user's query
- Do NOT add pleasantries, greetings, or conversational elements
- Do NOT mention missing URLs, other pages, or content not provided
- Do NOT ask follow-up questions or suggest additional actions
- Do NOT add commentary about the user's request or query
- Present the information in a clear, well-structured format with key facts and details
- Include all relevant details that could help answer the user's question
# User's Query: ${state.query}
# Content Title: ${webContent.metadata.title || 'Web Page'}
# Content URL: ${url}
# Web Page Content to Summarize:
${webContent.pageContent}
Provide a comprehensive summary of the above web page content, focusing on information relevant to the user's query:`;
const result = await this.llm.invoke(summarizationPrompt, {
signal: this.signal,
});
finalContent = removeThinkingBlocks(result.content as string);
processingType = 'url-content-extraction';
}
if (finalContent && finalContent.trim().length > 0) {
const document = new Document({
pageContent: finalContent,
metadata: {
title: webContent.metadata.title || 'URL Content',
url: url,
source: url,
processingType: processingType,
processingIntent: summarizationIntent,
originalContentLength: contentLength,
},
});
documents.push(document);
// Emit successful URL processing event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_CONTENT_EXTRACTED',
message: `Successfully processed content from: ${url}`,
details: {
query: state.query,
sourceUrl: url,
sourceTitle: webContent.metadata.title || 'Web Page',
contentLength: finalContent.length,
originalContentLength: contentLength,
processingType: processingType,
intent: summarizationIntent,
},
},
});
console.log(
`Successfully processed content from ${url} (${finalContent.length} characters, ${processingType})`,
);
} else {
console.warn(`No valid content generated for URL: ${url}`);
}
} catch (error) {
console.error(`Error processing URL ${url}:`, error);
// Emit URL processing error event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_PROCESSING_ERROR',
message: `Error processing URL: ${url}`,
details: {
query: state.query,
sourceUrl: url,
error: error instanceof Error ? error.message : 'Unknown error',
},
},
});
}
}
if (documents.length === 0) {
const errorMessage = `No content could be retrieved or summarized from the provided URL(s): ${urlsToProcess.join(', ')}`;
console.error(errorMessage);
return new Command({
goto: 'analyzer',
update: {
messages: [new AIMessage(errorMessage)],
},
});
}
// Emit completion event
this.emitter.emit('agent_action', {
type: 'agent_action',
data: {
action: 'URL_PROCESSING_COMPLETED',
message: `Successfully processed ${documents.length} URL(s) and extracted content`,
details: {
query: state.query,
processedUrls: urlsToProcess.length,
successfulExtractions: documents.length,
intent: summarizationIntent,
},
},
});
const responseMessage = `URL processing completed. Successfully processed ${documents.length} out of ${urlsToProcess.length} URLs.`;
console.log(responseMessage);
return new Command({
goto: 'analyzer', // Route to analyzer to continue with normal workflow after URL processing
update: {
messages: [new AIMessage(responseMessage)],
relevantDocuments: documents,
},
});
} catch (error) {
console.error('URL summarization error:', error);
const errorMessage = new AIMessage(
`URL summarization failed: ${error instanceof Error ? error.message : 'Unknown error'}`,
);
return new Command({
goto: END,
update: {
messages: [errorMessage],
},
});
} finally {
setTemperature(this.llm, undefined); // Reset temperature to default
}
}
}

View file

@ -28,6 +28,7 @@ Consider the following when evaluating sufficiency:
- If the user is requesting to avoid web searches respond with \`good_content\`
- If the user is asking you to be creative, such as writing a story, poem, or creative content respond with \`good_content\` unless the context is clearly insufficient
- If file documents contain complete information for file-specific queries respond with \`good_content\`
- If the user is requesting specific web content and there is a source that corresponds to that request in the context, it can be considered sufficient even if the content is not exhaustive or looks like errors respond with \`good_content\`
## Step 2: If content is insufficient, determine the type of missing information

View file

@ -21,6 +21,7 @@ import {
TaskManagerAgent,
FileSearchAgent,
ContentRouterAgent,
URLSummarizationAgent,
} from '../agents';
/**
@ -37,6 +38,7 @@ export class AgentSearch {
private synthesizerAgent: SynthesizerAgent;
private fileSearchAgent: FileSearchAgent;
private contentRouterAgent: ContentRouterAgent;
private urlSummarizationAgent: URLSummarizationAgent;
private emitter: EventEmitter;
private focusMode: string;
@ -95,6 +97,12 @@ export class AgentSearch {
systemInstructions,
signal,
);
this.urlSummarizationAgent = new URLSummarizationAgent(
llm,
emitter,
systemInstructions,
signal,
);
}
/**
@ -102,6 +110,13 @@ export class AgentSearch {
*/
private createWorkflow() {
const workflow = new StateGraph(AgentState)
.addNode(
'url_summarization',
this.urlSummarizationAgent.execute.bind(this.urlSummarizationAgent),
{
ends: ['task_manager', 'analyzer'],
},
)
.addNode(
'task_manager',
this.taskManagerAgent.execute.bind(this.taskManagerAgent),
@ -134,7 +149,7 @@ export class AgentSearch {
'analyzer',
this.analyzerAgent.execute.bind(this.analyzerAgent),
{
ends: ['task_manager', 'synthesizer'],
ends: ['url_summarization', 'task_manager', 'synthesizer'],
},
)
.addNode(