AI Assistants

🏛️ Publisher: Sergi Rodrigues (IMASDEWEB)
🌐 User interface language: English / Catalan / Spanish / Italian / German / French / Portuguese
📖 User Manual and DEMO Access: https://imasdeweb.com

Install

1. Go into menu Home - Setup - Modules - Deploy an external module, and submit the zip file
    
2. Go into menu Home - Setup - Modules, and activate the module. If you're updating, deactivate and activate again.
    
3. Normally, you should also visit the module configuration section and review any new options that appear. This module will be adding new AI models every X weeks, which by default will not be activated when updating the module, but you will need to do it manually.

Note: once installed, the CHANGELOG tab in the module settings loads the remote version of the ChangeLog.md file on my server, allowing users to easily view new features before deciding to install the update.
 

OpenAI, Perplexity and other APIs (IMPORTANT to read BEFORE use it)

This module requires you open a "developer account" on the provider you want to use their API services. Take in account that if you're already for example paying a monthly subscription like the $20USD of the chaGPT PLUS, you need anyway to open that "developer account" and PUT MONEY in it, because it usually are PRE-PAID accounts. If you don't do this, maybe they let you create an API key, but you will only receive "null" responses from their API when trying to use it.

Indeed, if you're paying the monthly $20USD of Perplexity PRO, then they give you until $5USD to consume their API models each month. They offer opensource Llama LLM family of Meta, useful for chat_completion and for search online within the chat. The level and price is probably about 90% of the 4o-mini from OpenAI.

 
Feature Description

• Use AI at minimal cost: With this module, each of your Dolibarr users can use AI services from different providers without having to pay a fixed monthly fee to the corresponding company, but "barely" paying for the consumption of those services through the respective APIs of those providers. This represents an immense saving.
   
• Different AI Provider APIs: currently the following
  
   + OpenAI.com: GPT-4o (text), GPT-4o-mini (text), GPT-40-search, o1-mini and o3-mini (reasoning)
   + Anthropic.com: Claude 3.5 Sonnet (text + vision), Claude 3.5 Haiku (text)
   + Perplexity.ai: Sonar PRO & Sonar REASONING (text, web search, reasoning) y DeepResearch
   + Fal.ai: familia FLUX (images), StableDifussion (images)
   + DeepSeek.com: V3 (text) y R1 (reasoning)
   + Mistral.ai: Ministral & Mistral (text) y Pixtral (vision)
   
• Keep the history of AI queries under your control: The module builds a history of conversations and requests for each user with each of the AIs they interact with. So even if in a few years the current providers no longer exist or you no longer use them, you will keep all the conversations and generated images in your Dolibarr.
   
• Multi-AI interface: This module integrates API connection to AI services from different providers for different uses. Conversational chat, internet search, image generation, audio transcription (coming soon), etc. But what's more, the user can switch AI in the middle of a conversation, as many times as needed!
   
• Control for the company: From the module configuration section you can

 + configure the use of one, several or all providers allowed by the module, just by placing the API key of each provider
 + activate or deactivate the AI models of each provider that you want to allow your users to use
 + define a default provider & model to use
 + define a limit of interactions per user per hour, and per 24 hours
 + enable or disable the use of this module in the permissions tab of each user

• API response speed monitoring: from the module configuration section, you can check all calls made to AI providers' APIs to verify how many milliseconds they take to respond.
   

How much does AI consumption cost
 

• This module does not include or provide any AI service. You must be the one to open an account with the provider you are interested in (you will find the direct link in the module configuration section) and obtain a secret API key to place in the configuration of this module.
   
• The rates as of today (January 2025) are very affordable. For example, both chatGPT-40-mini and the smallest Llama 3.1 model (through Perplexity's API) costs about $0.15-$0.20 USD to process ONE MILLION TOKENS, which is equivalent to more than TWO THOUSAND PAGES of a book! Obviously, your consumption of these AIs will vary greatly depending on the volume of information you pass as INPUT and the volume of information you request to be generated as OUTPUT.
   
• Very important to keep in mind that: every time you add a new message in a conversation with the AI, THE ENTIRE CONVERSATION IS SENT AGAIN for it to use as context... so long conversations greatly increase AI consumption. My advice is that if you don't need to take advantage of what was said in a conversation, start a new one. You will easily save 100% of money.

Coming Soon
 

• Enable semantic search in the conversation history using embeddings (something that not even chatGPT offers today!)
• Integrate audio transcription by uploading MP3 files through OpenAI's Whisper
• Integrate some text-to-video generator
• Have the possibility to generate something similar to OpenAI's GPTs: pre-instructed assistants to do a very specific task very efficiently without having to spend time each time configuring them, but calling them when we need them.
• Be able to interact with information from other modules, initially only for consultation, summary and analysis.

DEMO & User Guide

Here: https://imasdeweb.com/index.php?pag=m_blog&gad=detalle_entrada&entry=106

List of my modules for sale