+39 348 9323674 +39 011 9222277 info@webcreation.it
Italiano Tell us about your projectGet a quote

Documents·Procedures·Price lists·Archive

ChatGPT for business
on your documents

Tell us about your documents

Real case

Standards, certifications, technical data sheets.
Now the answer is found by asking.

At SAE Srl the people who sell fire extinguishers have to answer questions every day about standards, approvals and maintenance, with the answers scattered across hundreds of PDFs.

We built an AI assistant on company documents that answers in plain Italian, cites document and page and, if it cannot find the answer, says so. Everyone sees only the documents they are allowed to see.

  • 3,000+pages of technical documents searchable with a question
  • A few secondsfor an answer with the source cited
  • Every answerwith the reference document and page
Client SAE Srl I want an assistant on my documents

The answers are already in your company.
The problem is finding them.

A ChatGPT for business is an AI virtual assistant that answers the questions of the people working in a company starting from your documents: procedures, technical data sheets, price lists, past quotes, contracts, manuals, emails with customers. In every company the information exists, but it is scattered across shared folders, the ERP and mailboxes, and the people who really know it are two or three colleagues who get interrupted all the time.

With a custom-built ChatGPT for business, people ask a question in plain language and the assistant answers starting from your documents, citing the file and the passage the information comes from. It does not draw on the model's generic memory, but on an index of your content that we keep up to date.

We build it with permissions per department, a connection to the CRM or ERP and a log of the questions. In the cloud, with the models of the major providers under business contract terms, it costs €3,000 to €7,000 in 3-6 weeks; installation on your own servers falls under AI programming, from €5,000, like all our custom software. For professional practices there are dedicated pages for lawyers and accountants; all the services are under AI for business.

Company AI assistant searching internal documents

What it is for

What an AI virtual assistant
does with company documents

The use cases that pay back fastest are the ones where today an expert gets interrupted to answer the same questions over and over. If instead it has to answer customers, you need an AI chatbot; if it also has to enter data into your systems, AI agents.

Sales · Quotes

Faster quotes

The salesperson asks about prices, compatibility and terms applied to similar customers; the assistant prepares the draft quote from price lists and past quotes, with the references.

Let's talk

Technical · Support

Technical data sheets and manuals

Support people find the right procedure, spare part code or technical note in a few seconds, instead of opening ten PDFs or calling the colleague who knows.

Let's talk

New hires · Procedures

Onboarding new people

Internal procedures, rules, forms: a new employee asks the assistant instead of interrupting the others, and the answer is always the official one.

Let's talk

Archive · Email · CRM

A memory of customer relationships

Connected to the CRM or the archive, it reconstructs a customer's history: what they bought, what they asked, what problems they had, before the phone call.

Let's talk

How it works

Document search (RAG),
the foundation of every ChatGPT for business.

The technique is called RAG, retrieval-augmented generation. There is no need to train a model on your data: you build an index that the model consults for every question.

How an AI assistant with document search works

An index of your content

Documents, pages and records are split into parts with a method designed for the type of document and turned into vectors (embeddings) inside a vector database. The index is updated when files change.

Search and answers with sources

For every question the system retrieves the relevant passages, even if they are worded differently, and the model answers only from those, citing document and section. If the information is not there, it is instructed to say so.

Inherited permissions

Each person queries only the documents they are entitled to: administration sees contracts, sales sees price lists, support sees manuals. Permissions are applied before the search, not after.

Subscription or project

OpenAI's ChatGPT Business
or a custom assistant?

People looking for ChatGPT for business often find ChatGPT Business, OpenAI's plan for work groups. According to OpenAI's help center it costs $25 per user per month billed monthly or $20 billed annually, it requires at least two users and OpenAI does not train its models on workspace data. Since October 2025, with the company knowledge feature, it also searches Slack, SharePoint, Google Drive and GitHub, respecting existing permissions and citing sources. It is a good product and sometimes it is enough. The custom assistant we build is something else, software you own connected to your systems.

ChatGPT Business is enough if

  • The goal is to help people write, summarize and reason.
  • The documents are already in Google Drive, SharePoint or the other services ChatGPT connects to.
  • A per-user cost works for everyone who will use it.
  • There is no need to read internal ERP, CRM or databases.

You need a custom assistant if

  • Answers have to start from an ERP, CRM or databases that ChatGPT does not connect to.
  • The documents cannot leave the company and you need a model installed on your servers.
  • Many people use it, including sales agents and resellers, and paying for usage is cheaper than paying for seats.
  • You want to choose the model, interface and rules, with code you own.

Often the best combination is ChatGPT Business for personal work and the custom assistant for the company's knowledge and systems. OpenAI's prices and features change often: we check them with you before deciding.

The technical choices

How we build your
ChatGPT for business

The quality of an assistant on documents depends almost entirely on how the data is prepared and where the model runs.

  • Document preparation

    Cleanup of scanned PDFs with OCR, tables converted into a readable form, obsolete documents excluded, duplicate versions resolved. A clean index is worth more than a bigger model.

  • Hybrid search

    Search by meaning with embeddings plus search by exact words, useful for item codes, names and standard numbers. That way the assistant also finds a code written in a single document.

  • AI in the cloud or on-premises

    Models from the major providers with business contracts and data processing in Europe where available, or an open source model installed on your own server when documents must not leave. We evaluate quality and hardware cost on your case.

  • Fine-tuning only when needed

    To teach company content, RAG is cheaper, easier to update and more verifiable than fine-tuning. We use fine-tuning only for a very specific style or format, and we say so up front.

  • Connections to CRM and ERP

    Up to two integrations included in the base range: the assistant reads customer records, orders or items in real time, with read-only permissions.

  • Logging and unanswered questions

    Every question is kept in a log. The questions the assistant cannot answer show which documents are missing: it is the fastest way to improve the company's knowledge.

On-premises AI

Local AI
and open source models.

When documents must not leave the company, the model is installed on your own server or in a data center of your choice. It is the on-premises AI solution, much more practical today than a few years ago.

Company server running local AI with open source models

Open source models you can install

There are openly licensed models that you download and run without any connection to the providers. One example is the gpt-oss models released by OpenAI in August 2025 under the Apache 2.0 license: the smallest runs with 16 GB of memory, the largest on a single 80 GB GPU.

Hardware and quality, tested on your case

A local model is paid for in servers, not in usage. On complex tasks it performs worse than the largest cloud models, which is why we compare it on your documents before choosing, with the questions your colleagues actually ask.

Local and cloud together

Often the right choice is a mix: confidential documents processed locally, generic texts handed to a cloud model under a business contract. The system decides where to send each request according to the rules we set together.

Pricing

How much does
ChatGPT for business cost

The price depends on the quantity and type of documents, the number of integrations and where the model has to run. We set it at a fixed price after seeing a sample of your files.

Most requested

Virtual assistant

€3,000 to €7,000

ChatGPT for business · RAG

Timeline 3-6 weeks

  • Assistant for your employees
  • Search across company documents
  • Two integrations (CRM or ERP)
  • Dedicated web interface
  • User permissions and log
  • Training for the people who use it
Custom

AI programming

€5,000 to €15,000

API · MCP server · On-premises

Timeline 6-8 weeks

  • Integration into your software
  • MCP server for Claude and ChatGPT
  • Model installed at your company
  • Data update pipelines
  • Multiple models where needed
  • Your code, documented

Chatbot

€1,000 to €2,000

Website · WhatsApp

Timeline 1-2 weeks

  • The same knowledge for customers
  • Connection to catalog and FAQ
  • Integration on one channel
  • Your brand's tone of voice
  • Conversation log
  • GDPR privacy notice and consent

Small jobs

from €350

Hour packages

5 hours €350 · 10 hours €600 · 20 hours €1,100

  • New sources to index
  • New permissions or departments
  • Answer review
  • Model change
  • Support for the people who use it
  • Night work at a higher rate

Where to start? Make your company secure.

An analysis of how your employees already use AI, clear rules against shadow AI and training on secure tools. It is the first step we propose, before any development. Priced per project, with a fixed price after a first conversation.

Let's talk in the chat →

Cloud or on-premises changes the monthly costs.

In the cloud you pay for model usage based on the questions; on-premises there is no usage cost but you need a suitable server. We compare the two routes on your volumes before the quote.

Tell us about the project →

On request, payment can be spread over installments with no interest. 10% discount for payment in advance on amounts over €3,000.

Learn more

The next step,
connecting AI to your systems.

MCP · Claude · ChatGPT

Do you already use Claude or ChatGPT in your company?

With an MCP server, the clients your employees already use can read the ERP, databases and company documents, with the permissions you decide. Sometimes it is the fastest route.

MCP server →

Law firms

Do you run a law firm?

For law firms, the assistant on documents has to be designed around professional secrecy and the obligations of Italian Law 132/2025.

AI for lawyers →

Let's talk

Which documents would you like
to be able to query?

Tell us what kind of documents you have, how many there are, where they are kept and who should use the assistant. We will tell you what we would build and how much it costs. To talk about it directly, leave your number with "I want to be called back".

AI

AI chat · ChatGPT for business

online · instant replies
Hi, I'm Webcreation's AI assistant. Tell me which documents you would like to make searchable and for whom, and I'll tell you how we would build it.

Frequently asked questions

Frequently asked questions
about ChatGPT for business

What is a ChatGPT for business?

It is an AI assistant that answers the questions of the people working in a company using the company's documents and data, with sources cited and permissions respected. It uses a language model like the ones behind ChatGPT, but connected to your knowledge instead of only to general knowledge.

How much does ChatGPT for business cost?

It depends on what you mean. OpenAI's ChatGPT Business, according to its help center, costs $25 per user per month billed monthly or $20 billed annually, with at least two users. A custom assistant on the company's documents, built by us, ranges from €3,000 to €7,000 with a timeline of 3-6 weeks, plus model usage; on-premises installations or more complex integrations fall under AI programming, from €5,000 to €15,000.

What is the difference between ChatGPT Business and a custom assistant?

ChatGPT Business is a per-user subscription from OpenAI: excellent for writing and reasoning, and with company knowledge it also searches services such as Google Drive, SharePoint and Slack. A custom assistant is software you own, which also reads your ERP, CRM and databases, can run on a company server and is paid per project plus usage. It often makes sense to use them together.

Do the documents stay in the company?

With a model installed on your own server, yes, they do not leave. In the cloud, OpenAI states that it does not train its models on business data by default, and it indicates data residency in Europe for eligible Enterprise and API customers. We choose the architecture based on how confidential the documents are.

Can it be installed on-premises?

Yes. We install an open source model on your own server or in a data center of your choice. Local models need suitable hardware and on complex tasks they perform worse than the largest cloud models: we check this on your documents before deciding.

What is the best AI for business?

There is not just one. For personal work, the business versions of ChatGPT, Claude or Gemini are fine; for answering from the company's documents and systems, how the data is prepared matters more than the name of the model. In our projects we compare several models on your colleagues' real questions and choose the cheapest one that answers well.

How long does it take?

An assistant on your documents in the cloud takes 3-6 weeks, from preparing the files to training the people who use it. On-premises installations and more complex integrations take 6-8 weeks. The first step is to look at a sample of your documents.

Do you have to train the AI on our data?

Usually not. With document search (RAG) the model consults an index of your content for every question: it is cheaper, it updates at the next indexing when a file changes and every answer can be verified. Fine-tuning is needed only in particular cases.