Telegram News Article RAG Chatbot Setup Guide

What This Agent Does

This intelligent workflow transforms your Telegram into a powerful news analysis and research assistant. When users share news articles via URL, the bot automatically scrapes the content, analyzes it with AI, and stores it in a vector database for future reference. When users ask questions without URLs, the bot searches through previously stored articles and provides contextual answers based on your accumulated knowledge base.

Key benefits include:

• Instant article summaries delivered directly to Telegram, saving hours of manual reading
• Intelligent Q&A capabilities that reference your stored article database for consistent, informed responses
• Automated content management that organizes and indexes articles without manual effort
• 24/7 availability for your team to access news analysis and ask research questions anytime

This agent is perfect for news teams, research departments, competitive intelligence units, and knowledge workers who need to process and reference multiple articles efficiently.

Who Is It For

This workflow serves multiple user personas:

• News Researchers & Journalists: Quickly summarize and cross-reference multiple news sources
• Competitive Intelligence Teams: Monitor industry news and maintain searchable archives of competitor activities
• Knowledge Workers: Build a personal knowledge base of articles relevant to your domain
• Content Curators: Automatically analyze and categorize incoming news for team distribution
• Business Analysts: Extract insights from news articles and correlate them with historical data

No coding experience required—just API credentials and basic configuration knowledge.

Required Integrations

Telegram

Why it's needed: Telegram serves as your user interface, allowing team members to submit articles and ask questions through a familiar messaging platform.

Setup steps:

1. Open Telegram and search for @BotFather
2. Send the command /newbot and follow the prompts
3. Choose a name for your bot (e.g., "NewsAnalyzerBot")
4. Choose a username (must end with "bot")
5. Copy the API Token provided by BotFather (format: 123456789:ABCdefGHIjklmnoPQRstuvWXYZabcdefg)
6. In TaskAGI, navigate to Integrations → Telegram
7. Paste your API Token in the Bot Token field
8. Set your Webhook URL (TaskAGI will provide this)
9. Click Verify Connection to confirm setup

Configuration in TaskAGI:

• Store the Bot Token securely in your integration settings
• Enable webhook mode for real-time message processing
• Test by sending a message to your bot—you should receive a confirmation

FireCrawl

Why it's needed: FireCrawl extracts clean, readable content from web articles, handling complex page layouts and removing ads/navigation clutter.

Setup steps:

1. Visit firecrawl.dev and sign up for an account
2. Navigate to your Dashboard → API Keys
3. Click Create New API Key and copy it
4. In TaskAGI, go to Integrations → FireCrawl
5. Paste your API Key in the API Key field
6. Select Scraping Mode: Choose markdown for best article readability
7. Enable Wait for Dynamic Content if articles use JavaScript rendering
8. Click Test Connection to verify

Configuration in TaskAGI:

• Set timeout to 30 seconds for reliable scraping
• Enable Extract Metadata to capture article titles, dates, and authors
• Configure Output Format as markdown for clean text processing

OpenAI

Why it's needed: OpenAI's GPT-4 powers article analysis and intelligent question answering, providing sophisticated natural language understanding.

Setup steps:

1. Go to platform.openai.com and log in
2. Click your profile icon → API Keys
3. Select Create new secret key
4. Copy the key immediately (you won't see it again)
5. In TaskAGI, navigate to Integrations → OpenAI
6. Paste your API Key in the Secret Key field
7. Set Model to gpt-4o (the model used in this workflow)
8. Set Temperature to 0.7 for balanced creativity and consistency
9. Set Max Tokens to 1500 for comprehensive responses

Configuration in TaskAGI:

• Verify your OpenAI account has billing enabled and sufficient credits
• Test with a simple prompt to confirm API access
• Monitor token usage in your OpenAI dashboard to manage costs

Pinecone

Why it's needed: Pinecone is a vector database that stores article embeddings, enabling semantic search to find related articles based on meaning rather than keywords.

Setup steps:

1. Sign up at pinecone.io and create an account
2. Create a new Index with these settings:
   • Name: news-articles (or your preferred name)
   • Dimension: 1024 (for multilingual-e5-large embeddings)
   • Metric: cosine (best for semantic similarity)
   • Pod Type: starter (sufficient for most use cases)
3. Wait for the index to initialize (2-3 minutes)
4. Navigate to API Keys in your Pinecone dashboard
5. Copy your API Key and Environment (e.g., us-east-1-aws)
6. In TaskAGI, go to Integrations → Pinecone
7. Paste your API Key and Environment
8. Enter your Index Name (news-articles)
9. Click Verify Connection

Configuration in TaskAGI:

• Set Embedding Model to multilingual-e5-large (supports 100+ languages)
• Configure Batch Size to 10 for efficient vector uploads
• Enable Metadata Storage to preserve article URLs and titles

Configuration Steps

Node 1: Telegram Message Received (Trigger)

This is your workflow's entry point. No configuration needed—it automatically activates when someone sends a message to your bot.

Node 2: Check if Message Contains URL

Purpose: Routes the workflow based on whether the message includes a URL.

Configuration:

• Condition: message.text contains "http"
• True Path: Processes article (Nodes 3-7)
• False Path: Handles questions (Nodes 9-13)

Node 3: Extract URL from Message

Purpose: Isolates the URL from the message text.

Configuration:

• Input: message.text
• Regex Pattern: https?://[^\s]+
• Output Variable: extracted_url

Example: If user sends "Check this article: https://example.com/news", output is https://example.com/news

Node 4: Scrape Article Content

Purpose: Downloads and cleans article content using FireCrawl.

Configuration:

• URL Input: extracted_url
• Output Format: markdown
• Extract Metadata: Enabled
• Output Variables: article_content, article_title, article_metadata

Node 5: Summarize and Analyze Article

Purpose: Uses GPT-4 to analyze the article and extract key insights.

Configuration:

• Model: gpt-4o
• Temperature: 0.7
• Max Tokens: 1500
• Prompt:

Analyze this news article and provide a structured summary including:
1. Main headline and key facts
2. Impact and significance
3. Key stakeholders mentioned
4. Related topics or industries affected

Article content:
{article_content}

• Output Variable: article_summary

Node 6: Generate Embedding for Article

Purpose: Converts article content into a vector for semantic search.

Configuration:

• Input Text: article_content
• Model: multilingual-e5-large
• Output Variable: article_embedding

Node 7: Store Article in Vector Database

Purpose: Saves the article and its embedding to Pinecone for future retrieval.

Configuration:

• Index Name: news-articles
• Vector: article_embedding
• Metadata:

  {
    "title": "article_title",
    "url": "extracted_url",
    "summary": "article_summary",
    "timestamp": "current_date"
  }
  

• ID: Generate unique ID from URL hash

Node 8: Send Article Summary to Telegram

Purpose: Delivers the analysis back to the user.

Configuration:

• Chat ID: message.chat_id
• Message Text:

📰 **Article Summary**

{article_title}

{article_summary}

🔗 Source: {extracted_url}

Node 9: Handle Question (No URL)

Purpose: Prepares the user's question for semantic search.

Configuration:

• Input: message.text
• Output Variable: user_question

Node 10: Generate Query Embedding

Purpose: Converts the question into a vector for searching similar articles.

Configuration:

• Input Text: user_question
• Model: multilingual-e5-large
• Output Variable: query_embedding

Node 11: Search Related Articles

Purpose: Finds the most relevant stored articles using semantic similarity.

Configuration:

• Index Name: news-articles
• Query Vector: query_embedding
• Top K Results: 3 (retrieve 3 most relevant articles)
• Output Variable: search_results

Node 12: Prepare Context for Answer

Purpose: Formats retrieved articles into a context string for the AI.

Configuration:

• Input: search_results
• Format:

Based on these related articles:
{search_results[0].metadata.title} - {search_results[0].metadata.summary}
{search_results[1].metadata.title} - {search_results[1].metadata.summary}
{search_results[2].metadata.title} - {search_results[2].metadata.summary}

• Output Variable: context

Node 13: Generate AI Answer

Purpose: Creates an informed response using retrieved articles as context.

Configuration:

• Model: gpt-4o
• Temperature: 0.7
• Max Tokens: 1000
• Prompt:

You are a helpful assistant that answers questions based on news articles. 
Use the following context to answer the user's question accurately.

Context:
{context}

User Question: {user_question}

Provide a comprehensive answer citing the relevant articles.

• Output Variable: ai_answer

Node 14: Send Answer to Telegram

Purpose: Returns the AI-generated answer to the user.

Configuration:

• Chat ID: message.chat_id
• Message Text:

💡 **Answer Based on Your Articles**

{ai_answer}

_This answer was generated from {search_results.length} related articles in your knowledge base._

Testing Your Agent

Step 1: Verify All Integrations

Before testing the workflow:

1. Go to Integrations Dashboard
2. Click Test Connection on each integration:
   • ✅ Telegram (should show "Bot connected")
   • ✅ FireCrawl (should show "API accessible")
   • ✅ OpenAI (should show "API key valid")
   • ✅ Pinecone (should show "Index ready")

Step 2: Test Article Processing

Test Case 1: Submit a News Article

1. Send a message to your Telegram bot with a real news URL:

   Check this: https://www.bbc.com/news/world
   

2. Verify at each step:

   • ✅ Node 2: Message received (check logs)
   • ✅ Node 3: URL extracted correctly
   • ✅ Node 4: Article content scraped (check for article text)
   • ✅ Node 5: Summary generated (should be 2-3 paragraphs)
   • ✅ Node 6: Embedding created (vector dimension: 1024)
   • ✅ Node 7: Stored in Pinecone (check index size increased)
   • ✅ Node 8: Summary message received in Telegram

Expected result: Within 15-20 seconds, you receive a formatted summary with the article title, key points, and source link.

Step 3: Test Question Answering

Test Case 2: Ask a Question

1. Send a question to your bot:

   What recent news is there about technology?
   

2. Verify at each step:

   • ✅ Node 9: Question captured
   • ✅ Node 10: Query embedding generated
   • ✅ Node 11: Search returns 3 articles (check Pinecone logs)
   • ✅ Node 12: Context formatted correctly
   • ✅ Node 13: AI answer generated (should reference articles)
   • ✅ Node 14: Answer received in Telegram

Expected result: Within 10-15 seconds, you receive an answer that references the articles you previously submitted.

Step 4: Monitor Performance

Success Indicators:

• ✅ All messages processed within 20 seconds
• ✅ No error messages in workflow logs
• ✅ Pinecone index grows with each article submitted
• ✅ Answers become more relevant as you add more articles
• ✅ Zero failed API calls in integration logs

Troubleshooting:

• If articles aren't stored: Verify Pinecone index name matches configuration
• If answers are irrelevant: Add more articles to build a stronger knowledge base
• If timeouts occur: Increase timeout settings in FireCrawl configuration

Your Telegram News Article RAG Chatbot is now ready to revolutionize how you process and reference news!