AI Assistant Agent: RAG Input Flow¶

This workflow automates the ingestion of documents from Google Drive into a vector database for AI-powered search and retrieval. It processes various file formats, extracts metadata, chunks content appropriately, and stores everything in Supabase with proper embeddings for semantic search capabilities.

Purpose¶

No business context provided yet — add a context.md to enrich this documentation.

How It Works¶

Trigger Activation: The workflow starts via webhook, manual trigger, or scheduled execution
Configuration Setup: Defines Google Drive folder ID and admin chat ID for notifications
Folder Discovery: Lists all subfolders in the specified Google Drive directory
Human Approval: Sends Telegram messages to admin for folder selection approval
File Processing: Downloads files from approved folders, supporting DOCX, PDF, Markdown, PowerPoint, and JSON formats
Content Extraction: Extracts text content using appropriate parsers for each file type
Document Chunking: Splits large documents into manageable chunks with overlap for better retrieval
Metadata Extraction: Uses AI to extract themes and keywords from document content
Vector Storage: Generates embeddings and stores documents in Supabase vector database
Cleanup Operations: Optionally deletes old documents with human confirmation
Notifications: Sends completion status via Telegram

Mermaid Diagram¶

graph TD
    A[Webhook/Manual Trigger] --> B[Configuration]
    B --> C[List Subfolders]
    C --> D[Choose Folder]
    D --> E{Approval?}
    E -->|Yes| F[List Files]
    E -->|No| G[Loop Over Items1]
    F --> H[Mapping]
    H --> I[File Id List]
    H --> J[Collection Name]
    H --> K[Wait for delete flow]
    I --> L[Merge1]
    J --> L
    L --> M[Confirm Delete Vectors]
    M --> N{Delete Approved?}
    N -->|Yes| O[Delete Old Documents]
    N -->|No| P[Send Declined Message]
    O --> Q[Start Upsert]
    P --> Q
    Q --> K
    K --> R[Loop Over Items]
    R --> S[Download File From Google Drive]
    S --> T[Switch]
    T -->|DOCX| U[Google Docs]
    T -->|PDF| V[Extract from File]
    T -->|JSON_FIN| W[Extract from JSON]
    T -->|Invalid| X[Send Invalid Filetype Message]
    U --> Y[Edit Fields]
    V --> Y
    Y --> Z[Split Out1]
    Z --> AA[Chunking]
    AA --> BB[Extract Meta Data]
    BB --> CC[Merge]
    CC --> DD[Data Loader]
    DD --> EE[Supabase Vector Store]
    EE --> FF[Wait]
    FF --> R
    W --> GG[Split Out Codes]
    W --> HH[Split Out Categories]
    W --> II[Intents]
    GG --> JJ[Upsert Codes]
    HH --> KK[Upsert Categories]
    II --> LL[Split out intents]
    LL --> MM[Upsert Intents]
    JJ --> NN[Merge2]
    KK --> NN
    MM --> NN
    NN --> R

Trigger¶

Webhook: Accessible at /webhook/upsert endpoint for external integrations
Manual Trigger: For testing and manual execution
Schedule Trigger: Daily at 12:00 PM (currently disabled)

Nodes Used¶

Node Type	Purpose
Webhook	Receives external HTTP requests to start the workflow
Manual Trigger	Allows manual workflow execution for testing
Schedule Trigger	Automated daily execution (disabled)
Google Drive	Lists folders/files and downloads documents
Google Docs	Extracts content from Google Docs format
Extract from File	Processes PDF and other binary file formats
Extract from JSON	Handles JSON financial data files
Switch	Routes files based on type/extension
Telegram	Sends notifications and approval requests to admin
Code	Custom chunking logic for document splitting
Information Extractor	AI-powered metadata extraction
Supabase Vector Store	Stores documents with embeddings
OpenAI Chat Model	Powers AI extraction and processing
Embeddings OpenAI	Generates vector embeddings
Split In Batches	Processes items in controlled batches
Merge	Combines data from multiple sources
Set	Data transformation and field mapping
If/Switch	Conditional logic and routing

External Services & Credentials Required¶

Google Drive OAuth2: For accessing and downloading files
Google Docs OAuth2: For reading Google Docs content
OpenAI API: For embeddings and AI-powered metadata extraction
Supabase: Vector database for document storage
Telegram Bot: For admin notifications and approvals
PostgreSQL: For structured data storage (budget codes, categories, intents)

Environment Variables¶

No specific environment variables are documented. Configuration is handled through the Configuration node with hardcoded values for: - folder_id: Google Drive folder ID - admin_chat_id: Telegram chat ID for notifications

Data Flow¶

Input: - Google Drive folder containing documents (DOCX, PDF, MD, PPTX, JSON) - Admin approval decisions via Telegram

Processing: - Document content extraction and chunking - AI-generated metadata (themes, keywords) - Vector embeddings generation

Output: - Documents stored in Supabase vector database with metadata - Financial codes/categories stored in PostgreSQL - Status notifications via Telegram

Error Handling¶

File Download Errors: Continues processing other files if individual downloads fail
Invalid File Types: Sends notification to admin about unsupported files
Processing Errors: Continues with remaining items in batch
Human Approval Timeout: 15-minute limit for approval decisions
Deletion Confirmation: Double approval required for destructive operations

Known Limitations¶

Hardcoded configuration values require manual updates
Limited to specific file formats (DOCX, PDF, MD, PPTX, JSON)
Requires manual approval for each folder processing
PowerPoint conversion functionality is disabled
Financial JSON processing is specific to budget code structure

No related workflows mentioned in the current documentation.

Setup Instructions¶

Import Workflow: Import the JSON workflow definition into your n8n instance
Configure Credentials:
- Set up Google Drive OAuth2 connection
- Configure Google Docs OAuth2 access
- Add OpenAI API key for embeddings and chat
- Set up Supabase connection with vector store enabled
- Create Telegram bot and get API credentials
- Configure PostgreSQL connection for structured data
Update Configuration Node:
- Set folder_id to your Google Drive folder ID
- Update admin_chat_id with your Telegram chat ID
Database Setup:
- Ensure Supabase has documents table configured for vector storage
- Create PostgreSQL tables: budget_codes, budget_categories, budget_intents
Test Workflow:
- Use manual trigger to test the complete flow
- Verify Telegram notifications are working
- Confirm documents are properly stored in vector database
Production Deployment:
- Enable webhook endpoint for external integrations
- Configure schedule trigger if automated processing is needed
- Set up monitoring for error notifications