Artificial intelligence

A bespoke AI assistant, trained on your own data

It knows your company. Not just the internet.

Inside Web builds AI assistants that answer from your own content: catalogue, documentation, procedures, terms, history. On your site to advise your customers, or internally to save your team time. Every answer rests on your sources, cites them, and stays within the boundaries you set.

An assistant, not a chatbot from 2019.

The old chatbots followed decision trees: three buttons, two canned answers, and a visitor who left frustrated. A modern AI assistant understands a question asked in plain language, looks the information up in your documents and writes a clear answer, in the visitor's own language.

A useful assistant does not know everything. It knows everything you have taught it, and admits to the rest.

What your assistant can do.

  • Advise your customers

    Recommend the right product for the need described, compare two references, check availability.

  • Answer frequent questions

    Delivery, returns, opening hours, terms, warranties: accurate answers, day and night, with no queue.

  • Find internal information

    Procedures, technical sheets, model contracts, notes: your team asks the question instead of digging through ten folders.

  • Qualify enquiries

    Understand the need, gather the useful details and pass a complete enquiry to the right person.

  • Answer in several languages

    English, French, Dutch: the assistant answers in the language of the question, from the same sources.

  • Hand over to a person

    When a question falls outside its remit or turns sensitive, it says so and points to your team.

How it works.

The technique is called RAG, for Retrieval-Augmented Generation. Rather than answering from memory, the assistant first searches your content for the relevant passages, and then writes its answer from those alone.

  1. 01

    Gather your sources

    Pages from the site, the catalogue, PDFs, documentation, knowledge bases, and where needed data from Odoo or from your shop.

  2. 02

    Index the content

    Documents are split up and indexed so the assistant finds the right passage in a fraction of a second.

  3. 03

    Set the boundaries

    Tone, permitted subjects, information never to be given out, and the point at which a person takes over.

  4. 04

    Test with real questions

    We put the assistant up against the questions your customers or your team actually ask, and adjust until the answers are dependable.

  5. 05

    Go live and improve

    Conversations are reviewed to spot questions that went unanswered and to enrich the sources.

General-purpose AI or an assistant on your data?

General-purpose AIAn Inside Web assistant
Source of the answersThe model's general knowledgeYour documents, your catalogue, your procedures
Accuracy about your businessHit and miss, with a risk of inventionAnswers grounded in your sources, and cited
UpdatesDepend on the model's publisherAs soon as your content changes
Boundaries and toneGenericDefined by you
DataUnder the terms of a consumer serviceAn architecture chosen for confidentiality

Dependability and compliance.

  • Answers with sources

    The assistant works from identified passages and can display its sources, which makes every answer verifiable.

  • Clear limits

    Where the information does not exist in your content, it says so instead of inventing.

  • Your data protected

    Providers that do not use data sent through their API to train their models, with options for European hosting.

  • GDPR and the AI Act

    Users clearly informed that they are talking to an AI, personal data kept to a minimum, processing documented.

With an eye on what comes next.

An assistant that answers is a first step. The same foundation can then act: raise a quotation request in Odoo, track an order, book an appointment, or sit inside your mobile application. That is the step from assistant to AI agent.

Your questions.

(AI assistants on your data)

What is an AI assistant trained on its own data?

It is an assistant that answers from your company's content, such as your catalogue, your documentation or your procedures, rather than from general knowledge. Inside Web uses the RAG technique for this, which finds the relevant passages before writing the answer.

What is RAG?

RAG, for Retrieval-Augmented Generation, combines a search through your documents with writing by an AI model. The assistant first finds the relevant information in your sources, then frames its answer from that, which sharply limits invention.

Do we have to retrain an AI model with our data?

In the vast majority of cases, no. With RAG the model is not modified: your content is indexed and consulted at the moment of each question. An update to your documents is therefore taken into account without any retraining.

Can the assistant invent answers?

The risk exists with any AI, but it is greatly reduced when the assistant is limited to your sources, cites the passages it used and is configured to recognise when information is missing. Testing against real questions comes before any launch.

Which sources can the assistant use?

The pages of your site, your product catalogue, documents in PDF, Word or spreadsheet form, a knowledge base, and depending on the project data from Odoo, from your online shop or from other tools through an API.

Where can the assistant be integrated?

On your website or shop, in a mobile application, in an internal tool for your team, or in other conversation channels depending on the project. The same knowledge base can serve several channels.

Are our confidential documents safe?

Yes, if the architecture is right. We favour providers that do not use data sent through their API to train their models, options for European hosting and, where confidentiality demands it, self-hosted open source models.

Is an AI assistant compliant with the GDPR and the AI Act?

It can be. That means telling users clearly they are talking to an AI, limiting the personal data processed, documenting the processing and choosing compliant processors. These points are built in from the scoping stage.

Does the assistant speak French and Dutch?

Yes. Today's models understand and write in many languages, including English, French and Dutch. The assistant answers in the language of the question, even where your sources are written in another.

What happens when the assistant does not know the answer?

It says so clearly and offers an alternative: contacting your team, leaving their details or visiting a specific page. Those unanswered questions are then reviewed so the sources can be enriched.

How long does it take to set up an AI assistant?

A first working assistant on a well-defined scope usually takes a few weeks to build. The timing depends above all on the quantity and quality of the content to be gathered, and on the integrations required.

Can the assistant's effectiveness be measured?

Yes. The number of conversations, the most frequent questions, the share of questions answered, hand-overs to a person and the enquiries generated all let you follow its real value and improve it continuously.

Let's talk about your project.

A first conversation with no strings attached. One of the team gets back to you within 24 working hours.