HumationHuman + Automation
Back to Catalog

Automated customer support system with Gemini AI, RAG & security guardrails

RamSRamS
237 views
2/3/2026
GitHubReleasesAutomationAPI IntegrationCRM
Official Page

Description

This workflow acts as an autonomous Tier 2 Customer Support Agent. It doesn't just answer questions; it manages the entire lifecycle of a support ticket—from triage to resolution with Guardrails to deal with prompt injections, PII information blocking, etc. enabling such threats are blocked and logged in Airtable.

Unlike standard auto-responders, this system uses a "Master Orchestrator" architecture to coordinate specialized sub-agents. It creates a safe, human-like support experience by combining RAG (Knowledge Base retrieval) with a safety-first state machine.

How it works

The workflow operates on a strict "Hub and Spoke" model managed by a Master Orchestrator:

  1. Security Guardrails (The Gatekeeper) Before the AI even sees the message, a hard-coded security layer scans for Prompt Injection attacks, Profanity, and PII. If a threat is detected, the workflow locks down, logs the incident to Airtable, and stops execution immediately.

  2. Orchestration & Triage Once the message passes safety checks, the Master Orchestrator takes over. Its first action is to call the Ticket Analyser Agent.

  3. Analysis & Scoring The Ticket Analyser classifies the issue (e.g., "Technical," "Billing") and scores the customer's sentiment. It returns a priority_score to the Master Orchestrator.

  4. The Decision Logic (Circuit Breaker) The Master Orchestrator evaluates the score:

Escalation: If the customer is "Furious" or the score is high, it bypasses AI drafting and immediately alerts a human manager via Slack.

Resolution Path: If the request is standard, it proceeds to the next steps.

  1. Knowledge Retrieval (RAG) The Orchestrator calls the Knowledge Worker Agent. This agent searches your Supabase vector store to find specific, verified company policies or troubleshooting steps relevant to the user's issue.

  2. Resolution Drafting Armed with the analysis and the retrieved facts, the Orchestrator calls the Resolution Agent. This agent synthesizes a polite, professional email draft.

  3. Final Execution The Master Orchestrator reviews the final draft and sends the email via Gmail.

Set up

This is multi-agent system. Please follow these steps to configure the environment:

⚠️ IMPORTANT: This template contains the Main Orchestrator AND the Sub-Agents in a single view. You must separate them for the system to function:

Separate the Agents: Copy the nodes for each sub-agent (Ticket Analyser, Knowledge Worker, Resolution Agent) into their own new workflows.

Link the Tools: In the Main Orchestrator workflow, open the "Call [Agent Name]" tool nodes and update the Workflow ID to point to the new workflows you just created.

Configure Credentials: You will need credentials for Gmail (or your preferred email provider), Slack, Airtable, Supabase (for the vector store), and Google Gemini (or OpenAI).

Initialize the Knowledge Base:

Open the "One time Document Loader" section in the workflow.

Upload your policy document (PDF/Text) to the "Upload your file here" node.

Run this branch once to vectorize your documents into Supabase.

Setup Airtable: Create a simple table with columns for Sender Email, Incident Type, and Flagged Content to log security threats caught by the guardrails.

Customize the Trigger: Update the Gmail Trigger node to watch for your specific support alias (e.g., support@yourdomain.com) and ensure it only picks up "Unread" emails.

Adjust the Escalation Sensitivity: In the Orchestrator Agent node, you can tweak the "Phase 2" logic to change what triggers a human hand-off (currently set to priority_score >= 0.9).

Good to go!

image.png

Automated Customer Support System with Gemini AI, RAG & Security Guardrails

This n8n workflow automates customer support by leveraging Google Gemini AI, Retrieval Augmented Generation (RAG) with Supabase, and security guardrails to ensure safe and relevant responses. It can be triggered by incoming emails or form submissions, providing a robust and intelligent solution for handling customer inquiries.

What it does

This workflow streamlines customer support interactions through the following steps:

  1. Triggers on new input: The workflow can be initiated either by a new email received via IMAP/Gmail or by a submission to an n8n form.
  2. Prepares input for AI: It processes the incoming email subject and body, or form data, to extract the customer's query.
  3. Applies Security Guardrails: The customer's query is first passed through a Guardrails node to detect and prevent prompt injections, ensuring the AI agent operates securely and within defined boundaries.
  4. Retrieves relevant information (RAG): The AI Agent utilizes a Supabase Vector Store and an OpenAI Embeddings model to search for relevant information from a knowledge base. This Retrieval Augmented Generation (RAG) approach ensures that the AI's responses are informed by your specific data.
  5. Generates AI response: The AI Agent, powered by the Google Gemini Chat Model, processes the customer's query and the retrieved context to generate a comprehensive and helpful response.
  6. Stores conversation history: A Simple Memory buffer is used to maintain conversational context, allowing for more coherent and continuous interactions.
  7. Executes external actions (optional): The AI Agent can be configured to call other n8n workflows as tools, enabling it to perform actions like updating records in Airtable based on the customer's request.
  8. Sends AI-generated reply: The workflow can then send the AI-generated response back to the customer via email or display it as a form submission confirmation.

Prerequisites/Requirements

To use this workflow, you will need:

  • n8n Instance: A running n8n instance.
  • Email Account: Access to an IMAP-enabled email account or a Gmail account for the Email Trigger.
  • Google Gemini API Key: For the Google Gemini Chat Model.
  • OpenAI API Key: For the OpenAI Embeddings model.
  • Supabase Account: Configured with a vector store for your knowledge base.
  • Airtable Account (Optional): If you intend to use the "Call n8n Workflow Tool" to interact with Airtable.
  • Credentials: Appropriate n8n credentials configured for all connected services (IMAP/Gmail, Google Gemini, OpenAI, Supabase, Airtable).

Setup/Usage

  1. Import the workflow: Download the JSON provided and import it into your n8n instance.
  2. Configure Triggers:
    • Email Trigger (IMAP/Gmail): Set up your IMAP or Gmail credentials and specify the mailbox and rules for new emails to trigger the workflow.
    • n8n Form Trigger: Configure the form fields as needed.
  3. Configure AI Components:
    • Guardrails: Review and adjust the guardrail settings to match your security requirements.
    • Embeddings OpenAI: Provide your OpenAI API key.
    • Supabase Vector Store: Configure your Supabase credentials, project URL, and API key. Ensure your Supabase database is populated with the relevant knowledge base data.
    • Google Gemini Chat Model: Provide your Google Gemini API key.
    • Simple Memory: This node usually requires no specific configuration but ensures it's connected.
  4. Configure Tools (Optional):
    • Call n8n Workflow Tool: If you want the AI to interact with other systems (e.g., Airtable), configure this tool to point to a separate n8n workflow that performs the desired action.
    • Airtable Node: If using the Airtable tool, configure the Airtable credentials, base, and table.
  5. Activate the workflow: Once all credentials and configurations are set, activate the workflow.

The workflow is now ready to automatically process customer inquiries, provide intelligent responses, and potentially perform actions based on the conversation, all while maintaining security and leveraging your knowledge base.

Related Templates

Track competitor SEO keywords with Decodo + GPT-4.1-mini + Google Sheets

This workflow automates competitor keyword research using OpenAI LLM and Decodo for intelligent web scraping. Who this is for SEO specialists, content strategists, and growth marketers who want to automate keyword research and competitive intelligence. Marketing analysts managing multiple clients or websites who need consistent SEO tracking without manual data pulls. Agencies or automation engineers using Google Sheets as an SEO data dashboard for keyword monitoring and reporting. What problem this workflow solves Tracking competitor keywords manually is slow and inconsistent. Most SEO tools provide limited API access or lack contextual keyword analysis. This workflow solves that by: Automatically scraping any competitor’s webpage with Decodo. Using OpenAI GPT-4.1-mini to interpret keyword intent, density, and semantic focus. Storing structured keyword insights directly in Google Sheets for ongoing tracking and trend analysis. What this workflow does Trigger — Manually start the workflow or schedule it to run periodically. Input Setup — Define the website URL and target country (e.g., https://dev.to, france). Data Scraping (Decodo) — Fetch competitor web content and metadata. Keyword Analysis (OpenAI GPT-4.1-mini) Extract primary and secondary keywords. Identify focus topics and semantic entities. Generate a keyword density summary and SEO strength score. Recommend optimization and internal linking opportunities. Data Structuring — Clean and convert GPT output into JSON format. Data Storage (Google Sheets) — Append structured keyword data to a Google Sheet for long-term tracking. Setup Prerequisites If you are new to Decode, please signup on this link visit.decodo.com n8n account with workflow editor access Decodo API credentials OpenAI API key Google Sheets account connected via OAuth2 Make sure to install the Decodo Community node. Create a Google Sheet Add columns for: primarykeywords, seostrengthscore, keyworddensity_summary, etc. Share with your n8n Google account. Connect Credentials Add credentials for: Decodo API credentials - You need to register, login and obtain the Basic Authentication Token via Decodo Dashboard OpenAI API (for GPT-4o-mini) Google Sheets OAuth2 Configure Input Fields Edit the “Set Input Fields” node to set your target site and region. Run the Workflow Click Execute Workflow in n8n. View structured results in your connected Google Sheet. How to customize this workflow Track Multiple Competitors → Use a Google Sheet or CSV list of URLs; loop through them using the Split In Batches node. Add Language Detection → Add a Gemini or GPT node before keyword analysis to detect content language and adjust prompts. Enhance the SEO Report → Expand the GPT prompt to include backlink insights, metadata optimization, or readability checks. Integrate Visualization → Connect your Google Sheet to Looker Studio for SEO performance dashboards. Schedule Auto-Runs → Use the Cron Node to run weekly or monthly for competitor keyword refreshes. Summary This workflow automates competitor keyword research using: Decodo for intelligent web scraping OpenAI GPT-4.1-mini for keyword and SEO analysis Google Sheets for live tracking and reporting It’s a complete AI-powered SEO intelligence pipeline ideal for teams that want actionable insights on keyword gaps, optimization opportunities, and content focus trends, without relying on expensive SEO SaaS tools.

Ranjan DailataBy Ranjan Dailata
161

Generate song lyrics and music from text prompts using OpenAI and Fal.ai Minimax

Spark your creativity instantly in any chat—turn a simple prompt like "heartbreak ballad" into original, full-length lyrics and a professional AI-generated music track, all without leaving your conversation. 📋 What This Template Does This chat-triggered workflow harnesses AI to generate detailed, genre-matched song lyrics (at least 600 characters) from user messages, then queues them for music synthesis via Fal.ai's minimax-music model. It polls asynchronously until the track is ready, delivering lyrics and audio URL back in chat. Crafts original, structured lyrics with verses, choruses, and bridges using OpenAI Submits to Fal.ai for melody, instrumentation, and vocals aligned to the style Handles long-running generations with smart looping and status checks Returns complete song package (lyrics + audio link) for seamless sharing 🔧 Prerequisites n8n account (self-hosted or cloud with chat integration enabled) OpenAI account with API access for GPT models Fal.ai account for AI music generation 🔑 Required Credentials OpenAI API Setup Go to platform.openai.com → API keys (sidebar) Click "Create new secret key" → Name it (e.g., "n8n Songwriter") Copy the key and add to n8n as "OpenAI API" credential type Test by sending a simple chat completion request Fal.ai HTTP Header Auth Setup Sign up at fal.ai → Dashboard → API Keys Generate a new API key → Copy it In n8n, create "HTTP Header Auth" credential: Name="Fal.ai", Header Name="Authorization", Header Value="Key [Your API Key]" Test with a simple GET to their queue endpoint (e.g., /status) ⚙️ Configuration Steps Import the workflow JSON into your n8n instance Assign OpenAI API credentials to the "OpenAI Chat Model" node Assign Fal.ai HTTP Header Auth to the "Generate Music Track", "Check Generation Status", and "Fetch Final Result" nodes Activate the workflow—chat trigger will appear in your n8n chat interface Test by messaging: "Create an upbeat pop song about road trips" 🎯 Use Cases Content Creators: YouTubers generating custom jingles for videos on the fly, streamlining production from idea to audio export Educators: Music teachers using chat prompts to create era-specific folk tunes for classroom discussions, fostering interactive learning Gift Personalization: Friends crafting anniversary R&B tracks from shared memories via quick chats, delivering emotional audio surprises Artist Brainstorming: Songwriters prototyping hip-hop beats in real-time during sessions, accelerating collaboration and iteration ⚠️ Troubleshooting Invalid JSON from AI Agent: Ensure the system prompt stresses valid JSON; test the agent standalone with a sample query Music Generation Fails (401/403): Verify Fal.ai API key has minimax-music access; check usage quotas in dashboard Status Polling Loops Indefinitely: Bump wait time to 45-60s for complex tracks; inspect fal.ai queue logs for bottlenecks Lyrics Under 600 Characters: Tweak agent prompt to enforce fuller structures like [V1][C][V2][B][C]; verify output length in executions

Daniel NkenchoBy Daniel Nkencho
601

Automate invoice processing with OCR, GPT-4 & Salesforce opportunity creation

PDF Invoice Extractor (AI) End-to-end pipeline: Watch Drive ➜ Download PDF ➜ OCR text ➜ AI normalize to JSON ➜ Upsert Buyer (Account) ➜ Create Opportunity ➜ Map Products ➜ Create OLI via Composite API ➜ Archive to OneDrive. --- Node by node (what it does & key setup) 1) Google Drive Trigger Purpose: Fire when a new file appears in a specific Google Drive folder. Key settings: Event: fileCreated Folder ID: google drive folder id Polling: everyMinute Creds: googleDriveOAuth2Api Output: Metadata { id, name, ... } for the new file. --- 2) Download File From Google Purpose: Get the file binary for processing and archiving. Key settings: Operation: download File ID: ={{ $json.id }} Creds: googleDriveOAuth2Api Output: Binary (default key: data) and original metadata. --- 3) Extract from File Purpose: Extract text from PDF (OCR as needed) for AI parsing. Key settings: Operation: pdf OCR: enable for scanned PDFs (in options) Output: JSON with OCR text at {{ $json.text }}. --- 4) Message a model (AI JSON Extractor) Purpose: Convert OCR text into strict normalized JSON array (invoice schema). Key settings: Node: @n8n/n8n-nodes-langchain.openAi Model: gpt-4.1 (or gpt-4.1-mini) Message role: system (the strict prompt; references {{ $json.text }}) jsonOutput: true Creds: openAiApi Output (per item): $.message.content → the parsed JSON (ensure it’s an array). --- 5) Create or update an account (Salesforce) Purpose: Upsert Buyer as Account using an external ID. Key settings: Resource: account Operation: upsert External Id Field: taxid_c External Id Value: ={{ $json.message.content.buyer.tax_id }} Name: ={{ $json.message.content.buyer.name }} Creds: salesforceOAuth2Api Output: Account record (captures Id) for downstream Opportunity. --- 6) Create an opportunity (Salesforce) Purpose: Create Opportunity linked to the Buyer (Account). Key settings: Resource: opportunity Name: ={{ $('Message a model').item.json.message.content.invoice.code }} Close Date: ={{ $('Message a model').item.json.message.content.invoice.issue_date }} Stage: Closed Won Amount: ={{ $('Message a model').item.json.message.content.summary.grand_total }} AccountId: ={{ $json.id }} (from Upsert Account output) Creds: salesforceOAuth2Api Output: Opportunity Id for OLI creation. --- 7) Build SOQL (Code / JS) Purpose: Collect unique product codes from AI JSON and build a SOQL query for PricebookEntry by Pricebook2Id. Key settings: pricebook2Id (hardcoded in script): e.g., 01sxxxxxxxxxxxxxxx Source lines: $('Message a model').first().json.message.content.products Output: { soql, codes } --- 8) Query PricebookEntries (Salesforce) Purpose: Fetch PricebookEntry.Id for each Product2.ProductCode. Key settings: Resource: search Query: ={{ $json.soql }} Creds: salesforceOAuth2Api Output: Items with Id, Product2.ProductCode (used for mapping). --- 9) Code in JavaScript (Build OLI payloads) Purpose: Join lines with PBE results and Opportunity Id ➜ build OpportunityLineItem payloads. Inputs: OpportunityId: ={{ $('Create an opportunity').first().json.id }} Lines: ={{ $('Message a model').first().json.message.content.products }} PBE rows: from previous node items Output: { body: { allOrNone:false, records:[{ OpportunityLineItem... }] } } Notes: Converts discount_total ➜ per-unit if needed (currently commented for standard pricing). Throws on missing PBE mapping or empty lines. --- 10) Create Opportunity Line Items (HTTP Request) Purpose: Bulk create OLIs via Salesforce Composite API. Key settings: Method: POST URL: https://<your-instance>.my.salesforce.com/services/data/v65.0/composite/sobjects Auth: salesforceOAuth2Api (predefined credential) Body (JSON): ={{ $json.body }} Output: Composite API results (per-record statuses). --- 11) Update File to One Drive Purpose: Archive the original PDF in OneDrive. Key settings: Operation: upload File Name: ={{ $json.name }} Parent Folder ID: onedrive folder id Binary Data: true (from the Download node) Creds: microsoftOneDriveOAuth2Api Output: Uploaded file metadata. --- Data flow (wiring) Google Drive Trigger → Download File From Google Download File From Google → Extract from File → Update File to One Drive Extract from File → Message a model Message a model → Create or update an account Create or update an account → Create an opportunity Create an opportunity → Build SOQL Build SOQL → Query PricebookEntries Query PricebookEntries → Code in JavaScript Code in JavaScript → Create Opportunity Line Items --- Quick setup checklist 🔐 Credentials: Connect Google Drive, OneDrive, Salesforce, OpenAI. 📂 IDs: Drive Folder ID (watch) OneDrive Parent Folder ID (archive) Salesforce Pricebook2Id (in the JS SOQL builder) 🧠 AI Prompt: Use the strict system prompt; jsonOutput = true. 🧾 Field mappings: Buyer tax id/name → Account upsert fields Invoice code/date/amount → Opportunity fields Product name must equal your Product2.ProductCode in SF. ✅ Test: Drop a sample PDF → verify: AI returns array JSON only Account/Opportunity created OLI records created PDF archived to OneDrive --- Notes & best practices If PDFs are scans, enable OCR in Extract from File. If AI returns non-JSON, keep “Return only a JSON array” as the last line of the prompt and keep jsonOutput enabled. Consider adding validation on parsing.warnings to gate Salesforce writes. For discounts/taxes in OLI: Standard OLI fields don’t support per-line discount amounts directly; model them in UnitPrice or custom fields. Replace the Composite API URL with your org’s domain or use the Salesforce node’s Bulk Upsert for simplicity.

Le NguyenBy Le Nguyen
942