AI Agent

Bot with Generative Artificial Intelligence

What is AI Agent?

AI Agent is Multibot's most advanced bot type, using generative AI models like GPT-4, Gemini, Claude and others to create natural, intelligent conversations. It supports file upload, RAG (Retrieval-Augmented Generation) and integration with CRM data.

Basic Configuration

  • Bot name and description
  • Choose AI provider
  • Provider API Key
  • Model (GPT-4, Gemini, etc)

Advanced Parameters

  • Temperature (0-2)
  • Max Tokens
  • System Prompt (up to 50KB)
  • Keep history

RAG & Documents

  • File upload
  • Training links
  • Assistants API
  • Vector Store

Supported Providers

OpenAI

GPT-4o, GPT-4-turbo, GPT-3.5-turbo

Google Gemini

Gemini Pro, Gemini Pro Vision

Anthropic Claude

Claude 3 Opus, Sonnet, Haiku

Groq

Llama 3, Mixtral (ultra-fast)

Up to 500 tokens/second

Perplexity

Model with web search

Updated in real time

Perplexity

With integrated web search

Custom API

Any compatible API

Setup Guide

AI Agent Concepts

RAG (Retrieval-Augmented Generation)

RAG is an advanced technique that combines information retrieval with text generation:

  • Retrieves relevant information from a knowledge base
  • Enriches the AI model's context
  • Generates more accurate, factual answers
  • Reduces hallucinations in responses

In Multibot, you can upload documents that will be used by RAG to improve your agent's answers.

Temperature and its Effects

Temperature controls the randomness of the model's responses:

Low Temperature (0-0.5)

  • More focused and deterministic answers
  • Ideal for tasks with correct answers
  • Less creative, but more consistent

High Temperature (0.6-1.0)

  • More creative and diverse answers
  • Greater variability in responses
  • Ideal for idea generation and creative content

Tokens and Limits

AI models process text in units called tokens:

  • What is a token? Approximately 4 characters in English
  • Context limits: Varies by model (e.g., 8K to 128K tokens)
  • Input vs. output tokens: Both count towards the total limit
  • Management: Conversation history is truncated when it reaches the limit

Available AI Providers

OpenAI

Models: GPT-4, GPT-3.5

Strength: Best overall performance

Google Gemini

Models: Gemini Pro, Gemini Ultra

Strength: Integration with the Google ecosystem

Anthropic Claude

Models: Claude 3 Opus, Sonnet

Strength: Safety and alignment

Groq

Models: Llama 3, Mixtral

Strength: Extreme speed

Perplexity

Models: Sonar, Online

Strength: Real-time web search

Cost Estimates

Prices per 1 million tokens (approximate):

Model Input Output
GPT-4 Turbo $10.00 $30.00
Claude 3 Opus $15.00 $75.00
Gemini Pro $7.00 $21.00
Groq (Llama 3 70B) $0.59 $0.79
Perplexity (Sonar) $3.00 $15.00

* Values may vary depending on the provider and usage volume.

Provider Documentation

OpenAI Docs Google AI Docs Claude Docs Groq Docs Perplexity Docs
1

Get API Key

Access the chosen provider's site and create an API Key:

2

Create Bot in Multibot

In the Multibot dashboard, click "Create Bot" and select "AI Agent"

3

Configure Provider

Select the provider, paste the API Key, and choose the desired model

4

Write the System Prompt

Define the bot's personality and behavior. Example:

You are a virtual assistant for company XYZ.
Be polite, professional, and helpful.
Answer in English.
If you don't know something, be honest.

Best Practices

Tips for System Prompts

  • Be specific about the bot's role
  • Define tone of voice (formal, casual, technical)
  • Include examples of desired responses
  • Keep it concise (max 50KB, ideally 30KB)
  • Use RAG for large contexts
  • Test and refine continuously

Limitations

  • System Prompt limited to 50KB
  • API costs are the user's responsibility
  • Response time depends on the provider
  • Some providers have rate limits

Security

Multibot was designed with security first, using multiple layers of protection to ensure your data integrity.

Learn More