💬 Build a private Llama chatbot with Ollama, Groq, Slack and Google Sheets

⚡ 32 views · 💬 Support Chatbots

Description

This workflow builds a fully private, self-hosted AI chatbot using Meta Llama models. Unlike cloud-based AI APIs, every conversation stays on your infrastructure — no data leaves your environment. The chatbot remembers conversation history per session, routes different query types to specialized Llama prompts, logs all interactions, and can escalate unresolved queries to a human agent via Slack.

Powered by Ollama (local) or Groq/Together AI (cloud Llama endpoints) — configurable in one node.

What’s the Goal?

To give businesses a production-grade private AI chatbot that:

• Runs on their own servers with zero data exposure
• Handles customer support, internal helpdesk, sales FAQs, and onboarding
• Remembers context across a full conversation session
• Routes intelligently: support vs sales vs general vs escalation
• Logs every turn for quality review, training, and compliance

Why Does It Matter?

Most businesses cannot send sensitive conversations to OpenAI or Anthropic due to:

• GDPR, HIPAA, SOC2, or internal data governance policies
• Confidential customer data in support queries
• Proprietary internal knowledge that must stay private

Llama models run fully on-premise. This workflow gives those businesses the same quality AI chatbot experience with complete data sovereignty.

Monetization: sell this as a private AI chatbot deployment package to enterprises. Setup fee plus monthly hosting — recurring revenue.

How It Works

Stage A — Message Intake Webhook receives incoming chat message with session ID and user message text. Set node stores Llama endpoint config and normalizes the payload.

Stage B — Session Memory Code node loads conversation history for the session from an in-memory store. Appends the new user message to build the full context window for Llama.

Stage C — Intent Router IF node checks the message for keywords to classify intent: support issue, sales inquiry, general question, or escalation request. Routes to the matching Llama system prompt branch.

Stage D — Llama Inference HTTP Request calls the Llama API (Ollama local, Groq, or Together AI). Sends full conversation history plus the matched system prompt. Returns the assistant reply.

Stage E — Response Handling Code node parses the Llama output, updates the session memory, checks if escalation is needed, and formats the final response.

Stage F — Logging and Delivery Google Sheets logs every turn. Slack fires only when escalation is flagged. Webhook responds with the chatbot reply and session metadata.

Configuration Requirements

• LLAMA_ENDPOINT — Your Ollama URL (http://localhost:11434) or Groq/Together AI base URL
• LLAMA_API_KEY — API key if using Groq or Together AI (leave blank for local Ollama)
• LLAMA_MODEL — Model name e.g. llama3, llama3.1:8b, llama3.1:70b, mixtral
• SLACK_WEBHOOK_URL — For human escalation alerts
• GOOGLE_SHEET_ID — Conversation audit log

Setup Guide

Option A (Local / Private):

1. Install Ollama: curl -fsSL https://ollama.ai/install.sh | sh
2. Pull model: ollama pull llama3.1
3. Set LLAMA_ENDPOINT to http://localhost:11434
4. Leave LLAMA_API_KEY blank

Option B (Cloud Llama via Groq — fastest):

1. Sign up at groq.com and copy your API key
2. Set LLAMA_ENDPOINT to https://api.groq.com/openai/v1
3. Set LLAMA_MODEL to llama-3.1-8b-instant or llama-3.1-70b-versatile
4. Paste your Groq API key in LLAMA_API_KEY

Option C (Together AI):

1. Sign up at together.ai
2. Set endpoint to https://api.together.xyz/v1
3. Set model to meta-llama/Llama-3.1-8B-Instruct-Turbo

Steps for all options: 5. Open Set Llama Config node — fill in all values 6. Set SLACK_WEBHOOK_URL and GOOGLE_SHEET_ID 7. Activate and POST to /webhook/llama-chat

Sample Payload

{ sessionId: user-abc-123, message: My order arrived damaged and I need a refund, userId: user_123, botPersona: support, userName: Sarah }

Explore More Automation: Contact us to design AI-powered lead nurturing, content engagement, and multi-platform reply workflows tailored to your growth strategy.

🔗 Nodes Used

Google Sheets, HTTP Request, Webhook

📥 Import

Download workflow.json and import into n8n: Workflow menu → Import from File

📖 Importing guide · 🔑 Credential setup