Build an n8n + Peliqan RAG + Text-to-SQL Business Data Chatbot

Input: multi-source business data (CRM/Notion/ERP) → Process: Peliqan sync + Supabase vector store + AI Agent routing → Output: natural-language query answers

Who

Teams needing an AI chatbot in n8n that can query internal business data

You get

Users ask in natural language; AI Agent automatically chooses RAG (knowledge retrieval) or Text-to-SQL (data analytics)

You need

n8n · supabase · peliqan

Steps

1. Prepare Peliqan account and data sources

   Sign up for a free trial at peliqan.io. Add one or more business data sources in Peliqan (e.g. Hubspot, Pipedrive, Notion). Peliqan syncs data to its built-in warehouse via one-click ELT.

2. Configure n8n credentials and self-hosted environment

   Configure three credential types in n8n: openAiApi (OpenAI API Key), supabaseApi (Supabase project URL and service key), peliqanApi (API Key from Peliqan Settings). **Self-hosted n8n only**: set environment variable `export N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=True` on the server/container running n8n, otherwise the Peliqan node cannot be used as an AI Agent tool.

   Setup guidance · Source text。Environment variable for n8n self-hosted server

   Copy prompt

   export N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=True

3. Run data sync workflow to populate vector store

   Manually execute the 'When clicking Execute workflow' branch: Peliqan node 'Get table data' pulls from a specified table → Default Data Loader joins title and content into documents → Recursive Character Text Splitter (chunkSize=500, chunkOverlap=100) splits text → Embeddings OpenAI generates vectors → inserts into Supabase Vector Store (table 'documents', metadata includes source_url and title). Change the tableId in 'Get table data' if using a different source table.

   Execution prompt · Source text。Default Data Loader node jsonData expression field

   Copy prompt

   =# {{ $json.title + '\n' + $json.content }}

4. Configure AI Agent system message and data model

   Open the AI Agent node's System Message. Keep the role definition and tool usage instructions. Replace the example schema under 'This is the data model of tables & columns' with your actual Peliqan source tables and columns. The source example includes chargebee.customers, chargebee.credit_notes, chargebee.invoices, etc., but your environment needs the real schema.

   Execution prompt · Source text。AI Agent node System Message field; note source is excerpt, replace with full data model in production

   Copy prompt

   You are an assistant answering questions on internal business data.
   
   You can use the Vector Store to find info, which contains all pages of Notion (intranet) with internal knowledge.
   
   You can also run SQL queries for quantitive questions on tables. To execute an SQL query: send the query to Peliqan tool "Execute SQL" in a key SQL_Query__View_. Get the results from key "detail".
   
   This is the data model of tables & columns that can be used in SQL queries:
   
   Columns in table chargebee.customers:
   
   customer_id
   customer_first_name
   customer_last_name
   customer_email
   customer_phone
   
   Columns in table chargebee.credit_notes:
   
   credit_note_id
   credit_note_customer_id
   credit_note_subscription_id
   credit_note_reference_invoice_id
   credit_note_type
   
   Columns in table chargebee.invoices:
   
   invoice_id
   invoice_customer_id
   invoice_subscription_id
   
   (etc. truncated)

5. Activate workflow and start chatting

   Save and activate the workflow. Send messages via the 'When chat message received' trigger public URL or n8n's built-in Chat interface. The AI Agent routes by question type: knowledge questions call Supabase Vector Store (RAG), analytics questions call Peliqan Execute SQL tool (Text-to-SQL), or a combination. Simple Memory maintains conversation context.

After importing this template into n8n, configure Peliqan, Supabase, and OpenAI credentials in order, set the self-hosted environment variable, replace the data model schema in the AI Agent, then activate the workflow to start testing.