klbr/n8nworkflows.xyz
1
0
Fork 0
mirror of https://github.com/khoaliber/n8nworkflows.xyz.git synced 2026-04-28 00:29:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

creation

Browse Source
This commit is contained in:
nusquama
2025-11-12 18:50:25 +01:00
parent 0a7eafed5e
commit e371350cb2
1 changed files with 425 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
workflows/Multi-Language Telegram RAG Chatbot with Supervisor AI & Automated Google Drive Pipeline-6158/readme-6158.md
+425
View File
@@ -0,0 +1,425 @@
Multi-Language Telegram RAG Chatbot with Supervisor AI & Automated Google Drive Pipeline
https://n8nworkflows.xyz/workflows/multi-language-telegram-rag-chatbot-with-supervisor-ai---automated-google-drive-pipeline-6158
# Multi-Language Telegram RAG Chatbot with Supervisor AI & Automated Google Drive Pipeline
### 1. Workflow Overview
This workflow implements a **Multi-Language Telegram Retrieval-Augmented Generation (RAG) Chatbot** combined with a **Supervisor AI** and an **Automated Google Drive document processing pipeline**. It is designed for managing documents and chat interactions in multiple languages, enriching chatbot responses with contextual data stored in vector databases, and maintaining synchronized Google Sheets and Drive files.
**Primary use cases include:**
- Automated ingestion and indexing of documents from Google Drive (Word, PDF, Google Docs).
- Vector embedding and storage of document content in Supabase for fast semantic search.
- Multi-language chat interactions via Telegram with AI-powered translation and supervision.
- Supervisory AI agent managing context and routing queries to specialized AI agents (Product, Academy, News).
- Periodic scheduled document updates and cleaning of old data.
- Integration with Google Sheets for managing links and manual document metadata.
**Logical blocks:**
- **1.1 Trigger & Input Reception:** Telegram messages and Google Drive file triggers, plus manual and scheduled triggers.
- **1.2 Document Processing Pipeline:** Downloading, converting, extracting text from files, splitting, embedding, and vector store updating.
- **1.3 Data Management:** Google Sheets data retrieval and update, Supabase vector store maintenance, and cleaning old data.
- **1.4 AI Agents & Chatbot Logic:** OpenAI chat models, Langchain agents for content formatting, translation, supervision, and specialized domain agents.
- **1.5 Output & Response:** Message formatting, translation output, and sending responses back to Telegram.
- **1.6 Control Flow & Utilities:** Wait nodes, conditionals (If/Switch), split batches, and merges to manage workflow logic.
---
### 2. Block-by-Block Analysis
#### 1.1 Trigger & Input Reception
**Overview:**
This block handles the reception of input via Telegram, Google Drive triggers, and manual or scheduled triggers, initiating the workflow processes accordingly.
**Nodes Involved:**
- Telegram Trigger
- Google Drive File Created
- Google Drive File Updated
- Schedule Trigger1
- When clicking ‘Test workflow’
- When Executed Supervisor
**Node Details:**
- **Telegram Trigger**
- Type: Trigger node for Telegram webhook messages.
- Config: Listens for incoming Telegram chat messages.
- Inputs: External via webhook. Outputs: To message preprocessing nodes.
- Failures: Telegram API downtime, webhook misconfiguration.
- **Google Drive File Created / Updated**
- Type: Trigger for Google Drive file creation and update events.
- Config: Monitors specified Drive folders for new/changed files.
- Inputs: External webhook triggers. Outputs: To file processing batch nodes.
- Failures: Google Drive API permission errors, webhook failures.
- **Schedule Trigger1**
- Type: Scheduled trigger node.
- Config: Runs periodically to initiate batch processing of manual documents and website links.
- Inputs: None. Outputs: To Google Sheets retrieval nodes.
- **When clicking ‘Test workflow’**
- Type: Manual trigger node for testing workflow execution.
- Config: For manual start during design or debug.
- Outputs: To Google Sheets Website Links retrieval block.
- **When Executed Supervisor**
- Type: Execute Workflow Trigger (sub-workflow call).
- Config: Triggers Supervisor AI agent workflow for chat memory and context management.
- Inputs: Internal triggers from conditions or higher-level logic.
---
#### 1.2 Document Processing Pipeline
**Overview:**
Processes documents originating from Google Drive or manual input: downloading files, converting Word to Google Docs, extracting text from various formats, splitting text, generating embeddings, and updating vector stores.
**Nodes Involved:**
- Download File / Download File1
- Switch File Type / Switch File Type1
- Extract From Document / Extract From Document1
- Extract From PDF / Extract From PDF1
- Convert Word to Google Docs
- Google Drive - Delete File / Google Drive - Delete File1 / Google Drive - Delete File (second)
- Set File Data / Set File Data1 / Set File Data3 / Set File Data4
- Loop Over Files / Loop Over Files1 / Loop Over Files3 / Loop Over Files4
- Delete Old Doc Rows / Delete Old Doc Rows1 / Delete Old Doc Rows3 / Delete Old Doc Rows4
- Default Data Loader / Default Data Loader2
- Character Text Splitter / Character Text Splitter1
- Embeddings OpenAI / Embeddings OpenAI1 / Embeddings OpenAI2 / Embeddings OpenAI3 / Embeddings OpenAI4 / Embeddings OpenAI5
- Insert Content into Supabase Vectorstore / Insert Content Into Supabase Vectorstore
- Google Sheets Get Document
**Node Details:**
- **Download File / Download File1**
- Type: Google Drive node to download files.
- Config: Downloads files based on IDs passed from triggers or batch loops.
- Inputs: File ID from prior node. Outputs: File binary for extraction.
- Failures: Access permission errors, file not found, large file size timeout.
- **Switch File Type / Switch File Type1**
- Type: Switch node to split logic based on file type (e.g., docx, pdf).
- Config: Routes files to appropriate extraction nodes.
- Inputs: File metadata. Outputs: To extract nodes accordingly.
- Edge Cases: Unrecognized file types, corrupted files.
- **Extract From Document / Extract From Document1**
- Type: Extract text from Google Docs, Word, or similar document formats.
- Outputs: Extracted plain text for further processing.
- Failures: Unsupported formats, extraction errors.
- **Extract From PDF / Extract From PDF1**
- Type: Extract text from PDF files.
- Outputs: Extracted text for embedding.
- Failures: Scanned PDFs without OCR, extraction inaccuracies.
- **Convert Word to Google Docs**
- Type: HTTP Request node calling Google Docs API or similar service.
- Config: Converts Word files to Google Docs format for easier text extraction.
- Failures: API rate limits, conversion errors.
- **Google Drive - Delete File(s)**
- Type: Deletes files from Google Drive after processing or cleanup.
- Config: Runs on error continue mode to avoid workflow stops.
- **Set File Data (multiple)**
- Type: Set node to configure or modify variables and data fields.
- Used for: Preparing metadata or tracking file states during loops.
- **Loop Over Files (multiple)**
- Type: SplitInBatches node to process multiple files in batches for scalability and API quota control.
- **Delete Old Doc Rows (multiple)**
- Type: Supabase node to delete old document metadata or vectorstore rows to maintain freshness.
- **Default Data Loader / Default Data Loader2**
- Type: Langchain document loader node to prepare text chunks for embedding.
- **Character Text Splitter / Character Text Splitter1**
- Type: Langchain text splitter node to chunk long texts into manageable pieces.
- **Embeddings OpenAI (multiple)**
- Type: OpenAI embeddings node generating vector representations of text chunks.
- Requires: OpenAI API credentials.
- **Insert Content into Supabase Vectorstore (multiple)**
- Type: Inserts embeddings and associated metadata into Supabase vector database for semantic search.
- **Google Sheets Get Document**
- Retrieves metadata or document links from Google Sheets to drive processing.
---
#### 1.3 Data Management
**Overview:**
Manages Google Sheets data for website links and manual documents, updates links, and cleans old data from Sheets and Supabase. Also handles aggregation of processed data and updates to Sheets.
**Nodes Involved:**
- Google Sheets Website Links
- Google Sheets Website Links - Update
- Google Sheets - Manual Documents
- Google Sheets - Manual Documents - Get Rows
- Google Sheets Manual Documents - Delete Rows
- Google Sheets Website Links - Delete Rows
- Aggregate
- Filter Links
- Edit Fields Settings
- Edit Fields Telegram Message1
- Edit Fields Translator
- Edit Fields Output
**Node Details:**
- **Google Sheets Nodes**
- Types: Read (Get Rows), Update, and Delete rows in Google Sheets.
- Config: Manage two main sheets - Website Links and Manual Documents.
- Failures: API quota exceeded, permission errors, sheet not found.
- **Aggregate**
- Type: Aggregates data from multiple inputs into combined output for batch updates.
- **Filter Links**
- Type: Filter node to remove or select specific rows based on conditions.
- **Edit Fields (multiple)**
- Type: Set nodes to edit or enrich data fields for consistency before passing to downstream nodes.
---
#### 1.4 AI Agents & Chatbot Logic
**Overview:**
This block orchestrates multiple AI agents including OpenAI Chat Models, Langchain agents for translation, supervision, content formatting, and domain-specific RAG agents (Product, Academy, News). It manages chat memory and routing of queries to appropriate AI tools.
**Nodes Involved:**
- OpenAI Chat Model (multiple instances)
- Content Formatter
- Structured Output Parser / Structured Output Parser1 / Structured Output Parser2
- Translator1
- Output First Translator
- Output Translator
- Supervisor AI Agent1
- RAG AI Agent - Product
- RAG AI Agent - Academy
- RAG AI Agent - News
- Postgres Chat Memory1
- Supabase Vector Store / Supabase Vector Store1 / Supabase Vector Store2 / Supabase Vector Store3
- News AI Agent
- Academy AI Agent
- Product AI Agent
- Telegram Waiting
- If Question
- If English
- If
**Node Details:**
- **OpenAI Chat Model Nodes**
- Type: Langchain OpenAI chat completion nodes.
- Config: Used for generating AI responses, translation, and supervision.
- Inputs: Formatted user queries, context from vector stores or memory.
- Outputs: AI-generated text or structured data.
- Failures: API rate limits, invalid prompt formatting, connectivity issues.
- **Langchain Agents (Content Formatter, Translator, Supervisors, RAG AI Agents)**
- Type: Langchain agent nodes orchestrating multi-step AI workflows and tool usage.
- Role: Format content, translate between languages, supervise query processing, fetch domain-specific knowledge from vector stores.
- Inputs: Chat inputs, vector search results, memory context.
- Outputs: Structured responses or translated messages.
- **Structured Output Parser Nodes**
- Type: Parses AI output into structured JSON or predefined formats for further processing.
- **Postgres Chat Memory1**
- Type: Stores chat history and context in a PostgreSQL memory backend.
- Role: Maintains conversation state for more coherent AI responses.
- **Supabase Vector Store Nodes**
- Type: Connects to Supabase vector database for semantic similarity search.
- Role: Retrieves relevant document embeddings to augment AI responses.
- **If Question / If English / If**
- Type: Conditional nodes to branch logic based on message content or language detection.
- **Telegram Waiting**
- Type: Telegram node sending intermediate messages or typing indicators to users.
---
#### 1.5 Output & Response
**Overview:**
Formats final AI-generated replies, optionally translates them, merges outputs, and sends the resulting messages back to the Telegram user.
**Nodes Involved:**
- Edit Fields Output
- Merge Output
- Send a text message
**Node Details:**
- **Edit Fields Output**
- Type: Set node to prepare final message fields before sending.
- **Merge Output**
- Type: Merge node combines multiple translation or AI agent outputs into a single message.
- **Send a text message**
- Type: Telegram node to send the final chat message to the user.
- Config: Uses Telegram credentials and chat ID from incoming message.
- Failures: Telegram API errors, invalid chat ID.
---
#### 1.6 Control Flow & Utilities
**Overview:**
Contains utility nodes managing delays, batch processing, merging datasets, and sticky notes for documentation.
**Nodes Involved:**
- Wait / Wait1 / Wait2 / Wait4 / Wait5
- SplitInBatches (Loop Over Links, Loop Over Files, Loop Over Items)
- Merge (Merge Document Output, Merge Output)
- Sticky Notes (multiple)
**Node Details:**
- **Wait Nodes**
- Type: Pauses workflow execution, often used to space API calls or wait for asynchronous processes.
- **SplitInBatches Nodes**
- Type: Splits large datasets into smaller batches for processing to avoid API limits.
- **Merge Nodes**
- Type: Combine multiple data branches back into a single stream.
- **Sticky Notes**
- Used for inline documentation and developer notes. No functional role.
---
### 3. Summary Table
| Node Name | Node Type | Functional Role | Input Node(s) | Output Node(s) | Sticky Note |
|----------------------------------|--------------------------------------|---------------------------------------------|------------------------------------|------------------------------------|----------------------------------|
| When clicking ‘Test workflow’ | Manual Trigger | Manual start for testing workflow | | Edit Fields Settings | |
| Edit Fields Settings | Set | Prepare initial settings/data | When clicking ‘Test workflow’ | Google Sheets Website Links | |
| Google Sheets Website Links | Google Sheets | Retrieve list of website links | Edit Fields Settings | Filter Links | |
| Filter Links | Filter | Filter website links as per criteria | Google Sheets Website Links | Loop Over Links | |
| Loop Over Links | SplitInBatches | Process website links in batches | Filter Links | HTTP Request - Crawl4AI | |
| HTTP Request - Crawl4AI | HTTP Request | Crawl links for AI content | Loop Over Links | Edit Fields | |
| Edit Fields | Set | Format and prepare crawled data | HTTP Request - Crawl4AI | Content Formatter | |
| Content Formatter | Langchain Agent | Format AI content from crawled data | Edit Fields | If Document ID | |
| If Document ID | If | Check if document ID exists | Content Formatter | Get Text From Doc / Google Docs - Create | |
| Get Text From Doc | Google Docs | Retrieve text from existing Google Docs | If Document ID | Update Text To Doc | |
| Update Text To Doc | Google Docs | Update existing Google Docs with new text | Get Text From Doc | Merge Document Output | |
| Google Docs - Create | Google Docs | Create new Google Docs document | If Document ID (false path) | Save Text To Doc | |
| Save Text To Doc | Google Docs | Save extracted text into Google Docs | Google Docs - Create | Merge Document Output | |
| Merge Document Output | Merge | Combine document update/create outputs | Update Text To Doc / Save Text To Doc | Google Sheets Website Links - Update | |
| Google Sheets Website Links - Update | Google Sheets | Update website links metadata | Merge Document Output | Wait | |
| Wait | Wait | Delay between batches | Google Sheets Website Links - Update | Loop Over Links | |
| Loop Over Files | SplitInBatches | Process files from Google Drive in batches | Google Drive File Created/Updated | Set File Data | |
| Set File Data | Set | Prepare file metadata for processing | Loop Over Files | Google Sheets Get Document | |
| Google Sheets Get Document | Google Sheets | Retrieve document metadata from Sheets | Set File Data | Delete Old Doc Rows | |
| Delete Old Doc Rows | Supabase | Clean old document vectorstore rows | Google Sheets Get Document | Download File | |
| Download File | Google Drive | Download files from Drive | Delete Old Doc Rows | Switch File Type | |
| Switch File Type | Switch | Route files based on type (doc/pdf) | Download File | Extract From Document / Extract From PDF | |
| Extract From Document | Extract From File | Extract text from docs | Switch File Type | Insert Content into Supabase Vectorstore | |
| Extract From PDF | Extract From File | Extract text from PDFs | Switch File Type | Insert Content into Supabase Vectorstore | |
| Insert Content into Supabase Vectorstore | Langchain VectorStore Supabase | Store embeddings and metadata | Extract From Document / Extract From PDF | Aggregate | |
| Aggregate | Aggregate | Aggregate vectorstore insert outputs | Insert Content into Supabase Vectorstore | Google Sheets Manual Documents | |
| Google Sheets - Manual Documents | Google Sheets | Manage manual document metadata | Aggregate | Wait2 | |
| Wait1 | Wait | Delay for batch processing | Insert Content into Supabase Vectorstore | Loop Over Files | |
| Loop Over Files1 | SplitInBatches | Process second batch of files | Wait1 | Set File Data1 | |
| Set File Data1 | Set | Prepare metadata for second batch files | Loop Over Files1 | Delete Old Doc Rows1 | |
| Delete Old Doc Rows1 | Supabase | Clean old rows for second batch | Set File Data1 | Download File1 | |
| Download File1 | Google Drive | Download files for second batch | Delete Old Doc Rows1 | Switch File Type1 | |
| Switch File Type1 | Switch | Route second batch files by type | Download File1 | Extract From Document1 / Extract From PDF1 | |
| Extract From Document1 | Extract From File | Extract text from document (second batch) | Switch File Type1 | Insert Content Into Supabase Vectorstore | |
| Extract From PDF1 | Extract From File | Extract text from PDF (second batch) | Switch File Type1 | Insert Content Into Supabase Vectorstore | |
| Insert Content Into Supabase Vectorstore | Langchain VectorStore Supabase | Insert embeddings (second batch) | Extract From Document1 / Extract From PDF1 | Aggregate | |
| Wait2 | Wait | Delay before processing manual documents | Google Sheets - Manual Documents | Loop Over Files1 | |
| Loop Over Items | SplitInBatches | Process items from sheets in batches | Set File Get Data / Set File Get Data1 | If / Loop Over Files1 | |
| Set File Get Data | Set | Prepare data fields for batch processing | Google Drive File Created1 / Google Drive File Updated1 | Loop Over Items | |
| Edit Fields Telegram Message1 | Set | Prepare Telegram message data | Telegram Trigger | Translator1 | |
| Translator1 | Langchain Agent | Translate incoming message | Edit Fields Telegram Message1 | Edit Fields Translator | |
| Edit Fields Translator | Set | Prepare translated fields | Translator1 | Output First Translator | |
| Output First Translator | Langchain Agent | Initial output translation | Edit Fields Translator | Telegram Waiting | |
| Telegram Waiting | Telegram | Send typing indicator or waiting message | Output First Translator | If Question | |
| If Question | If | Check if message is a question | Telegram Waiting | Supervisor AI Agent1 | |
| Supervisor AI Agent1 | Langchain Agent | Supervise AI interactions, manage chat memory | If Question | If English | |
| If English | If | Check if response language is English | Supervisor AI Agent1 | Edit Fields Output / Output Translator | |
| Edit Fields Output | Set | Prepare final output fields | If English | Merge Output | |
| Output Translator | Langchain Agent | Translate response if needed | If English (false path) | Merge Output | |
| Merge Output | Merge | Merge translated and original outputs | Edit Fields Output / Output Translator | Send a text message | |
| Send a text message | Telegram | Send final chatbot response to user | Merge Output | | |
| Postgres Chat Memory1 | Langchain Memory Postgres | Store and retrieve chat history | | Supervisor AI Agent1 | |
| Supabase Vector Store (various) | Langchain VectorStore Supabase | Connect to different vector stores for Product, Academy, News agents | | AI agents | |
| RAG AI Agent - Product | Langchain Agent | Product domain knowledge retrieval | OpenAI Chat Model4 | Supervisor AI Agent1 | |
| RAG AI Agent - Academy | Langchain Agent | Academy domain knowledge retrieval | OpenAI Chat Model6 | Supervisor AI Agent1 | |
| RAG AI Agent - News | Langchain Agent | News domain knowledge retrieval | OpenAI Chat Model7 | Supervisor AI Agent1 | |
| News AI Agent | Langchain ToolWorkflow | Specialized news AI agent workflow | Supervisor AI Agent1 | | |
| Academy AI Agent | Langchain ToolWorkflow | Specialized academy AI agent workflow | Supervisor AI Agent1 | | |
| Product AI Agent | Langchain ToolWorkflow | Specialized product AI agent workflow | Supervisor AI Agent1 | | |
---
### 4. Reproducing the Workflow from Scratch
1. **Set up credentials:**
- Configure Google Drive, Google Sheets, Telegram Bot, OpenAI, and Supabase credentials in n8n.
2. **Triggers:**
- Create a **Telegram Trigger** node to listen to incoming messages.
- Add **Google Drive File Created** and **Google Drive File Updated** triggers for monitoring files.
- Add a **Schedule Trigger** node to run periodic batch jobs.
- Add a **Manual Trigger** node for manual testing.
3. **Initial Data Retrieval:**
- Add **Google Sheets** nodes to retrieve lists of website links and manual documents.
- Use **Set** nodes to prepare and edit fields from trigger data.
4. **Batch Processing:**
- Add **SplitInBatches** nodes to process files and links in manageable chunks.
5. **File Download and Type Routing:**
- Add **Google Drive Download File** nodes for downloading files.
- Add **Switch** nodes to route files based on type (docx, pdf, etc.).
6. **File Conversion and Text Extraction:**
- For Word files, add an **HTTP Request** node for conversion to Google Docs.
- Add **Extract From File** nodes for document and PDF text extraction.
7. **Text Processing:**
- Add Langchain **Text Splitter** nodes to divide large text into chunks.
- Add Langchain **Default Data Loader** nodes to prepare chunks for embedding.
8. **Embeddings and Vector Store:**
- Add **OpenAI Embeddings** nodes to generate vector embeddings.
- Add **Supabase Vector Store** nodes to insert embeddings and metadata.
9. **Data Cleaning:**
- Add **Supabase** nodes to delete old document rows before inserting new ones.
- Add **Google Sheets Delete Rows** nodes to clean outdated records.
10. **AI Agents:**
- Add multiple **OpenAI Chat Model** nodes for chat completions.
- Add Langchain **Agent** nodes for:
- Content formatting
- Translation
- Supervisor AI
- Domain-specific RAG agents (Product, Academy, News)
- Add **Structured Output Parser** nodes to parse AI outputs.
11. **Chat Memory:**
- Add **Postgres Chat Memory** node to maintain conversation context.
12. **Message Preparation and Routing:**
- Add **Set** nodes to format messages for Telegram.
- Add **If** nodes to check message language and question presence.
- Add **Merge** nodes to combine multiple AI outputs.
13. **Sending Responses:**
- Add **Telegram** node to send final text messages to users.
14. **Control Flow:**
- Insert **Wait** nodes as needed to manage API rate limits and asynchronous processes.
15. **Testing and Debugging:**
- Use **Manual Trigger** to start workflow runs for testing.
- Add **Sticky Notes** for documentation and clarification.
---
### 5. General Notes & Resources
| Note Content | Context or Link |
|-------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|
| Workflow is a Market Template called "AIAutomationPro Ultimate RAG Chatbot v1" | Found in n8n Marketplace under AI Automation templates. |
| Uses Langchain nodes for advanced AI orchestration with OpenAI integration | Requires n8n Langchain nodes and OpenAI API access. |
| Integrates with Supabase vector store for semantic search | Supabase configuration is mandatory for vector storage and retrieval. |
| Multi-language support with translators and output parsers | Enables seamless conversation in various languages, useful for global chatbot applications. |
| Telegram API webhook requires proper setup and bot token | Telegram bot must be created and webhook URL configured in Telegram Botfather. |
| Google Drive and Sheets API require OAuth2 credentials | Ensure OAuth credentials allow file read/write and spreadsheet access. |
| Workflow includes error handling with "continue on fail" for file deletions | Prevents workflow halts due to non-critical errors during cleanup steps. |
| Scheduled triggers enable regular updates and cleaning of document indexes | Helps maintain freshness and relevancy of indexed data for chatbot responses. |
| Workflow is designed for scalability with batch processing and delays | Avoids hitting API rate limits and manages large document volumes efficiently. |
---
This document provides a thorough, structured reference for the workflow “Multi-Language Telegram RAG Chatbot with Supervisor AI & Automated Google Drive Pipeline,” enabling advanced users and AI agents to understand, reproduce, and extend the system effectively.