First AI Chatbot for Dolibarr (Ticket Plus & Helpdesk Module)

caos30 · August 22, 2024, 7:10am

Hello everyone!

I’m thrilled to announce that six months after its initial release, I’ve just updated the “Ticket Plus & Helpdesk” module with an exciting new feature: it now includes its very own AI chatbot!

This chatbot incorporates some cutting-edge features:

It uses RAG (Retrieval-Augmented Generation) techniques on articles from the native “Knowledge Base” module and leverages vector embeddings technology to find the most relevant information for visitor queries, whether they’re customers, users, or suppliers.
It utilizes the ChatGPT API (OpenAI) for MULTI-LANGUAGE conversational abilities. This means we can have our knowledge base articles in French, but the chatbot will UNDERSTAND AND RESPOND perfectly in OTHER LANGUAGES: Spanish, English, Catalan, Portuguese, etc.

To get started, you’ll need to:

Sign up (for free) as a developer on OpenAI (https://platform.openai.com). You may need to provide your credit card details.
Once your account is created, generate an “API key” (https://platform.openai.com/api-keys) and paste it into the TicketPlus module configuration.

Worried about the cost of using AI? Don’t be! The most expensive part is actually buying my module (€60, one-time payment). The API consumption is highly optimized, and the current rates for using the “gpt-4-turbo” model API are minimal (https://openai.com/api/pricing/):

$0.15 per MILLION INPUT TOKENS
$0.60 per MILLION OUTPUT TOKENS

What does this mean in practice? Well, a token is roughly equivalent to 1.5 syllables. In my tests, answering a visitor’s question using about 4 short articles from the “knowledge base” consumes approximately ONE THOUSAND TOKENS.

This means we can comfortably answer about 100 visitor questions for just ONE CENT USD. While this calculation isn’t exact, it gives you an idea of the AI cost for using a RAG-based chatbot focused on your business.

To me, this seems very reasonable: Can a visitor ask about 10 questions in a single chat conversation? So, the cost of serving 10 visitors 24/7 with good information and intelligence is just about one cent? Wow!

Disclaimer

This is the first version of my module with a chatbot. Until now, it only had semantic search (also using OpenAI embeddings). You can continue using it that way, or even choose not to use AI at all. It’s up to you, but I recommend giving it a try.

That said, PLEASE, if you have any questions or suggestions for improvement, don’t hesitate to contact me. I believe this is an opportunity for all of us to natively enjoy a chatbot for our Dolibarr without any software dependencies other than the API connection to an LLM provider.

Future Plans

I’ve read in this forum for several months that there’s a lot of interest in similar chatbot technology that would allow a customer to query information from OTHER Dolibarr databases (products, for example) and even enable “conversational” creation of an order (even if it’s just in draft mode).

If you’re an LLM enthusiast like me, you know there’s so much potential. Let’s go for it! I’m counting on your collaboration.

Links

My module on Dolistore
DEMO site: https://demo.bimex.tech/custom/ticketplus/

caos30 · October 2, 2025, 12:43am

One year later, I’ve just released a new version compatible with Dolibarr 22, now truly ready to leverage much more from AI chatbots using RAG.

The TicketPlus module now gives you a section called “Chatbot List” that:

Allows you to create as many chatbots as you want.
Lets you embed these chatbots on your website by copying and pasting a JavaScript code anywhere on the page, which will display the typical chat bubble in the bottom corner.
Each chatbot can be configured with your preferred OpenAI LLM model (including GPT-5, but Mistral coming soon) and its own system prompt. You can even customize the bubble color and UI elements of the chat.
For each chatbot, you specify which “knowledge base articles” it can consult. The module automatically (re)calculates embeddings for these articles to keep them well-indexed and always retrieve the 3 most relevant ones to answer user queries.
You can review conversation lists and filter them by date or by chatbot, allowing you to identify improvements needed for the chatbot’s system prompt or the articles it references, enhancing performance over time.
Contact information left by visitors through the chatbot is also recorded there.
Finally, you can set the email address where you want to receive notifications by email when a visitor starts a conversation with the chatbot

I’ve been using these chatbots in my two SaaS platforms:

for almost half a year now, and they’re proving really useful:

They work great for clarifying commercial questions from potential clients
Additionally, I explicitly state in the system prompt that their goal is twofold: to answer questions and offer human assistance to visitors, requesting their name, phone number, and email with the promise of being contacted within 24 hours

And it actually works: some visitors leave their contact details, with the added advantage that I can review their conversations and understand what each person was interested in.

Note about AI consumption

AI consumption is minimal:

For (re)calculating embeddings of knowledge base articles, it’s truly negligible… just cents per month.
For bot conversations with visitors (processing INPUT + RELEVANT ARTICLES + generating OUTPUT) it’s very cheap… It varies depending on the LLM model you choose, but GPT-4.1-mini costs about $0.20 - $0.80 USD to process ONE MILLION INPUT tokens… equivalent to a two-thousand-page book… which is really very little… A conversation with about 4 back-and-forth messages (4 from user + 4 from bot) uses about 3k tokens… meaning one million tokens would cover 333 such 4+4 conversations!! So the AI cost is ridiculously small

Very soon (I hope within a month), I’ll integrate the Mistral API, which I know is highly valued in this environment. Besides being very economical and excellent for these types of tasks, it’s particularly fast (I know because I often use it in my “AI Assistants” module, also available on Dolistore).