Hello, dev community! 👋 I’ve been working on a personal project lately: a WhatsApp AI Bot that actually keeps track of conversations. No more "forgetful" bots, and best of all: it runs entirely on my own hardware! 🧠💻
🛠️ The Tech Stack
Runtime: Node.js 🟢
AI Engine: Ollama (Running Llama 3 / Mistral locally) 🦙
WhatsApp Interface: WPPConnect 📱
Database: SQLite for persistent conversation memory 🗄️
OS: Linux 🐧
🚀 The Journey
The goal was to create an assistant that doesn't rely on external APIs like OpenAI. By combining WPPConnect with Ollama, I have full control over the data and the model.
Here is the project structure:
Bash
user@remote-server:~/whatsapp-bot$ ls
database.db # Long-term memory (SQLite)
node_modules # The heavy lifters
package.json # Project DNA
server.js # The brain connecting WPPConnect + Ollama
tokens/ # Session persistence (No need to re-scan QR)
🔍 Key Features
Local Intelligence: Using Ollama means zero latency from external servers and 100% privacy.
True Context: Instead of stateless replies, I use SQLite to feed the previous chat history back into Ollama. It remembers who you are! 🔄
Session Persistence: Thanks to the tokens folder, the bot stays logged in even after a server reboot.
💡 Quick Snippet (The Connection)
Here is how I bridge the WhatsApp message to the local Ollama instance:
JavaScript
const wppconnect = require('@wppconnect-team/wppconnect');
const axios = require('axios'); // To talk to Ollama's local API
async function askOllama(prompt) {
const response = await axios.post('http://localhost:11434/api/generate', {
model: 'llama3',
prompt: prompt,
stream: false
});
return response.data.response;
}
wppconnect.create({ session: 'ai-session' })
.then((client) => {
client.onMessage(async (message) => {
const aiReply = await askOllama(message.body);
await client.sendText(message.from, aiReply);
});
});
🚧 What's next?
I'm working on "System Prompts" to give the bot a specific personality and improving the SQLite query speed for massive chat histories.
Are you running LLMs locally? I’d love to hear how you optimize Ollama performance for real-time chat! 👇
Top comments (1)
Nice approach. I built something very similar recently using Node.js + Ollama, but went with whatsapp-web.js instead of WPPConnect.
The SQLite conversation memory is a good addition — that’s one area I haven’t added yet. How much previous conversation do you usually send back to Ollama before context size or response time starts becoming an issue?
Also curious whether you’ve had any stability issues with the WhatsApp session after long runtimes or Windows/Linux restarts.