Intelligent chatbot with custom knowledge base

Who's it for

Businesses, developers, and organizations who need a customizable AI chatbot for internal documentation access, customer support, e-commerce assistance, or any use case requiring intelligent conversation with access to specific knowledge bases.

What it does

This workflow creates a fully customizable AI chatbot that can be deployed on any platform supporting webhook triggers (websites, Slack, Teams, etc.). The chatbot accesses a personalized knowledge base stored in Supabase and can perform advanced actions like sending emails, scheduling appointments, or updating databases beyond simple conversation.

How it works

The workflow combines several powerful components:

Webhook Trigger: Accepts messages from any platform that supports webhooks
AI Agent: Processes user queries with customizable personality and instructions
Vector Database: Searches relevant information from your Supabase knowledge base
Memory System: Maintains conversation history for context and traceability
Action Tools: Performs additional tasks like email sending or calendar booking

Technical architecture

Chat trigger connects directly to AI Agent
Language model, memory, and vector store all connect as tools/components to the AI Agent
Embeddings connect specifically to the Supabase Vector Store for similarity search

Requirements

Supabase account and project
AI model API key (any LLM provider of your choice)
OpenAI API key (for embeddings - this is covered in Cole Medin's tutorial)
n8n built-in PostgreSQL access (for conversation memory)
Platform-specific webhook configuration (optional)

How to set up

Step 1: Configure your trigger

The template uses n8n's default chat trigger
For external platforms: Replace with webhook trigger and configure your platform's webhook URL
Supported platforms: Any service with webhook capabilities (websites, Slack, Teams, Discord, etc.)

Step 2: Set up your knowledge base

For creating and managing your vector database, follow this comprehensive guide:

Watch Cole Medin's tutorial on document vectorization
This video shows how to build a complete knowledge base on Supabase
The tutorial covers document processing, embedding creation, and database optimization
Important: The video explains the OpenAI embeddings configuration required for vector search

Step 3: Configure the AI agent

Define your prompt: Customize the agent's personality and role

Example: "You are the virtual assistant for example.com. Help users by answering their questions about our products and services."

Select your language model: Choose any AI provider you prefer (OpenAI, Anthropic, Google, etc.)
Set behavior parameters: Define response style, tone, and limitations

Step 4: Connect Supabase Vector Store

Add the "Supabase Vector Store" tool to your agent
Configure your Supabase project credentials
Mode: Set to "retrieve-as-tool" for automatic agent integration
Tool Description: Customize description (default: "Database") to describe your knowledge base
Table configuration:

Specify the table containing your knowledge base (example shows "growth_ai_documents")
Ensure your table name matches your actual knowledge base structure
Multiple tables: You can connect several tables for organized data structure

The agent will automatically decide when to search the knowledge base based on user queries

Step 5: Set up conversation memory (recommended)

Use "Postgres Chat Memory" with n8n's built-in PostgreSQL credentials
Configure table name: Choose a name for your chat history table (will be auto-created)
Context Window Length: Set to 20 messages by default (adjustable based on your needs)
Benefits:

Conversation traceability and analytics
Context retention across messages
Unique conversation IDs for user sessions
Stored in n8n's database, not Supabase

How to customize the workflow

Basic conversation features

Response style: Modify prompts to change personality and tone
Knowledge scope: Update Supabase tables to expand or focus the knowledge base
Language support: Configure for multiple languages
Response length: Set limits for concise or detailed answers
Memory retention: Adjust context window length for longer or shorter conversation memory

Advanced action capabilities

The chatbot can be extended with additional tools for:

Email automation: Send support emails when users request assistance
Calendar integration: Book appointments directly in Google Calendar
Database updates: Modify Airtable or other databases based on user interactions
API integrations: Connect to external services and systems
File handling: Process and analyze uploaded documents

Platform-specific deployments

Website integration

Replace chat trigger with webhook trigger
Configure your website's chat widget to send messages to the n8n webhook URL
Handle response formatting for your specific chat interface

Slack/Teams deployment

Set up webhook trigger with Slack/Teams webhook URL
Configure response formatting for platform-specific message structures
Add platform-specific features (mentions, channels, etc.)

E-commerce integration

Connect to product databases
Add order tracking capabilities
Integrate with payment systems
Configure support ticket creation

Results interpretation

Conversation management

Chat history: All conversations stored in n8n's PostgreSQL database with unique IDs
Context tracking: Agent maintains conversation flow and references previous messages
Analytics potential: Historical data available for analysis and improvement

Knowledge retrieval

Semantic search: Vector database returns most relevant information based on meaning, not just keywords
Automatic decision: Agent automatically determines when to search the knowledge base
Source tracking: Ability to trace answers back to source documents
Accuracy improvement: Continuously refine knowledge base based on user queries

Use cases

Internal applications

Developer documentation: Quick access to technical guides and APIs
HR support: Employee handbook and policy questions
IT helpdesk: Troubleshooting guides and system information
Training assistant: Learning materials and procedure guidance

External customer service

E-commerce support: Product information and order assistance
Technical support: User manuals and troubleshooting
Sales assistance: Product recommendations and pricing
FAQ automation: Common questions and instant responses

Specialized implementations

Lead qualification: Gather customer information and schedule sales calls
Appointment booking: Healthcare, consulting, or service appointments
Order processing: Take orders and update inventory systems
Multi-language support: Global customer service with language detection

Workflow limitations

Knowledge base dependency: Quality depends on source documentation and embedding setup
Memory storage: Requires active n8n PostgreSQL connection for conversation history
Platform restrictions: Some platforms may have webhook limitations
Response time: Vector search may add slight delay to responses
Token limits: Large context windows may increase API costs
Embedding costs: OpenAI embeddings required for vector search functionality