IntentiAI

Intent-aware LLM routing that saves 60-90% on API costs while maintaining quality.

Installation

npm install -g intentiai

Quick Start

# Login (opens browser for authentication)
intentiai auth login

# Create a project
intentiai project create "My Chatbot"

# Add intents
intentiai intent add "refund_request" -d "Customer wants a refund"
intentiai intent add "order_status" -d "Customer checking order status"
intentiai intent add "product_question" -d "Questions about products"

# Set up routing (cheap models for easy queries, GPT-4 for hard ones)
intentiai route add -i "order_status" -d easy -p groq -m "llama-3.1-8b-instant"
intentiai route add -i "product_question" -d easy -p groq -m "llama-3.1-8b-instant"
intentiai route add -i "refund_request" -d hard -p openai -m "gpt-4o"

# Set system prompts
intentiai prompt set -i "refund_request" -s "You are a helpful customer service agent. Help with refunds politely."

# Train the intent classifier
intentiai train start --samples 20 --watch

# Test your chatbot
intentiai test "Where is my order?"
intentiai test "I want a refund for my purchase"

# Get embed code for your website
intentiai deploy

How It Works

1. Intent Classification - Queries are classified into intents (refund_request, order_status, etc.)
2. Difficulty Assessment - Each query is rated as easy, medium, or hard
3. Smart Routing - Easy queries → fast/cheap models (Groq Llama), Hard queries → powerful models (GPT-4)
4. Cost Savings - 60-90% reduction in LLM API costs

Commands

Authentication

intentiai auth login              # Browser-based login (recommended)
intentiai auth login --no-browser # Terminal login with email/password
intentiai auth whoami             # Show current user
intentiai auth logout             # Logout

Project Management

intentiai project create "Name"   # Create a new project
intentiai project list            # List all projects
intentiai project select <id>     # Select a project to work with

Intent Management

intentiai intent add "name" -d "description"   # Add an intent
intentiai intent list                          # List all intents

Routing Rules

Route queries to different models based on intent and difficulty:

• easy → Fast, cheap models (Groq Llama 3.1 8B)
• medium → Balanced models (Groq Llama 3.3 70B)
• hard → Powerful models (GPT-4o)

intentiai route add -i "intent_name" -d easy -p groq -m "llama-3.1-8b-instant"
intentiai route add -i "intent_name" -d medium -p groq -m "llama-3.3-70b-versatile"
intentiai route add -i "intent_name" -d hard -p openai -m "gpt-4o"
intentiai route list

System Prompts

intentiai prompt set -i "intent_name" -s "Your system prompt here"
intentiai prompt get -i "intent_name"

Training

intentiai train start --samples 20 --watch   # Train the classifier
intentiai train status                        # Check training status

Testing & Deployment

intentiai test "Your test message"   # Test a query
intentiai deploy                     # Get embed code and API info

API Integration

JavaScript/TypeScript

const response = await fetch('https://api.intentiai.com/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'X-Project-Key': 'your_project_key'
  },
  body: JSON.stringify({
    messages: [{ role: 'user', content: 'I want a refund' }]
  })
});

const data = await response.json();
console.log(data.choices[0].message.content);

Python (OpenAI SDK Compatible)

from openai import OpenAI

client = OpenAI(
    base_url='https://api.intentiai.com/v1',
    api_key='your_project_key'
)

response = client.chat.completions.create(
    model='auto',  # IntentiAI routes automatically
    messages=[{'role': 'user', 'content': 'I want a refund'}]
)

print(response.choices[0].message.content)

Links

• Website: https://www.intentiai.com
• Documentation: https://www.intentiai.com/docs
• Dashboard: https://www.intentiai.com/dashboard

License

MIT