The whole company
A reliable answer to every internal question
- Procedures, internal memos, HR or quality rules found simply by asking the question
- Every answer comes with its source document, so you can check it in one click

Your teams already use ChatGPT or Copilot to write an email or summarize a document. But when it comes to finding the latest version of a procedure, the rate negotiated with a client or the history of a case, the assistant knows nothing about your company. And everyone ends up pasting internal documents into a consumer tool, without really knowing where they go. At Kreante, we build custom assistants for SMEs and mid-sized companies. Your assistant draws on your own documents and software, cites its sources, respects each person’s access rights, and can act in your tools when you authorize it to.
265+ projects delivered in 35 countries · 4.9/5 average rating on Clutch · 110+ client companies
































These situations may sound familiar:
A new salesperson who takes three months to learn the products, prices and terms of sale
The same questions asked every week to HR or accounting
Procedures stored in a SharePoint where nobody can find the right version
Client contracts copied into the free version of ChatGPT to get a summary
An AI assistant tested six months ago, which made up answers and that nobody opens anymore
A customer service team answering by hand requests whose answer is already in your documents
According to Bpifrance Le Lab, 32% of SMEs and mid-sized companies use AI, and 54% of them rely on a free solution [1]. It is a good way to get familiar with the subject. But a free, general-purpose tool knows neither your clients, nor your products, nor your rules, and it offers no guarantee on your data.
A reliable answer to every internal question
Arrive at meetings already knowing the client
Faster answers that are always accurate
Less time spent on repetitive tasks
The right information to make decisions
The assistant is connected to the documents and software you choose: SharePoint, Google Drive, procedure database, CRM, ERP.
For each question, it finds the most useful passages in your sources, respecting the access rights of the person asking.
The AI model writes a clear answer from these passages and cites the documents used. If it finds nothing, it says so.
When planned, the assistant prepares an action in your tools: creating a task, updating a client record, sending an email. This way, we can connect AI to your existing software without changing your habits.
Unanswered questions and user feedback are tracked, to complete the sources and improve the assistant every month.
Several approaches exist, from the simplest to the most complete. The choice depends mainly on what the assistant needs to know about your company, and on what it must be able to do in your tools.
| Approach | What it does well | Its limits | Ideal for |
|---|---|---|---|
| Free consumer assistant | Discovering AI without spending | Knows nothing about the company, few guarantees on data | A personal trial, with no confidential documents |
| Professional plan (ChatGPT Business, Claude, Copilot, Le Chat) | High-quality writing, summarizing and analysis, protected data | Company knowledge limited to the documents added, few business rules | Writing and analysis needs across the whole team |
| AI built into your software | Works directly in the CRM, email or accounting software | Each tool has its own assistant, which only sees its own data | Needs limited to a single piece of software |
| Turnkey subscription assistant | Quick start on everyday tasks: letter templates, reminders, social media | Hard to adapt to your procedures and business tools | Very small businesses with standard needs |
| Kreante custom assistant | Connected to your sources and software, access rights, sourced answers, approved actions | Initial scoping and a few weeks of development | SMEs that want an assistant that truly knows their company |
If your teams mainly need to write, summarize and translate, a well-configured professional plan will be enough. Custom development makes sense when the assistant must rely on your internal documents, query your software or respect access rights that differ from person to person. We will tell you from the very first conversation.
Your assistant will consult contracts, price lists and sometimes personnel files every day. Before writing a single line of code, we decide with you what it can read, what it can do and what it must forget.
A salesperson querying the assistant will never get information taken from an HR file: everyone only finds the documents they can already open in your tools
Your conversations with the assistant are not used to train AI models, and its data can remain hosted in Europe
Only the personal data needed for the answer is processed, as required by the GDPR [2]
The assistant can prepare an email to a client or an order change, but an employee approves the sending
Every question, answer and action remains logged, so you know at all times who asked for what
If you ever change provider, you keep everything: the code and documentation are in your own GitHub repository from day one
When a technician asks for the tightening torque of a part, the assistant gives the value and opens the manual at the right page. And if the answer does not exist anywhere in your documents, it simply says so.
Nobody will have one more piece of software to open. The assistant answers in Teams, Slack or your intranet, and fetches information from SharePoint, Google Drive, your CRM or your ERP.
AI models evolve every few months. We select the one that offers the best balance of quality, cost and hosting for your uses, and the assistant is designed to switch the day a better one appears.
Every unanswered question points us to a missing or poorly written document. We review them with you every month, and the assistant improves at your teams’ pace.
Our first projects were built with Bubble, WeWeb and Xano. Today, we develop with Cursor, Codex and Claude Code, on PostgreSQL, Supabase and Python. Our AI agents write the code, and our senior developers review and approve it before each release.
Free
30 minutes with an expert
From €3,900 excl. VAT
Up and running in about six weeks
Custom quote
At the pace of your needs
We spend half a day with the people who will use the assistant. They show us what they look for every day, and where they look for it. You leave with the list of sources to connect first.
Three versions of the same procedure, a 2023 price list still online: together we identify what must be discarded, updated or restricted to certain profiles.
We connect it to your sources and to your teams’ work tool. From the third week, you ask it your own questions and judge its answers.
Around ten employees use it every day. Every answer judged wrong or incomplete is analyzed and corrected before it opens to everyone.
The assistant opens to the teams concerned, with a short onboarding for each role. You receive the code and documentation, and we monitor usage during the first few weeks.
Come with the questions that keep coming back, the documents where the answer is hiding, or an assistant you have already tried without success. In 30 minutes, we look together at what an AI assistant could change for you. It is free, with no commitment.
It is an artificial intelligence tool, built on a large language model, that helps employees in their daily work: answering a question, writing, summarizing, analyzing a document or preparing an action. What sets it apart from a consumer assistant is its knowledge of the company. It relies on your documents, your software and your rules, and it respects each person’s access rights.
| Tool | What it does | Example |
|---|---|---|
| Generative AI | Produces content, whether text, image or code, from a request | Writing a follow-up email from three keywords |
| Chatbot | Answers questions according to scenarios planned in advance | A website help module offering three answer choices |
| AI assistant | Understands a free-form request, relies on your information and helps an employee complete their task | Finding the product return procedure and writing the reply to the customer |
| AI agent | Chains several actions on its own to reach a goal, under human supervision | Reading return requests, checking the warranty, creating the credit note and notifying the customer |
For an assistant to know your company, it is connected to your documents: for each question, it fetches the useful passages and uses them to answer. This method, called RAG, costs far less than training, updates as soon as a document changes, and makes it possible to cite sources. Training a model is only justified for very specific needs, such as technical vocabulary that off-the-shelf models do not master.
Yes, for simple needs. ChatGPT lets you create custom GPTs with your instructions and a few documents, and Microsoft offers Copilot Studio to build agents within its environment. It is a good way to test an idea. The limits appear when you need to manage different access rights, query business software, keep hundreds of documents up to date or monitor answer quality over time.
Start with the task, before the tool. List the repetitive tasks in each department, estimate the time they take, and choose the one that combines a clear gain with low risk. Then check whether your current software already offers an AI feature, and compare market solutions with custom development. If nobody in-house has time to lead this thinking, an AI consultant present a few days a month can carry it out with you.
Yes, and this is often how AI enters an SME. Connected to the executive’s calendar, email and documents, a personal assistant prepares meetings, summarizes the week’s exchanges and finds a file in seconds. It requires the same precautions as an enterprise assistant: a professional plan, well-defined access and clear rules on what it can do on the executive’s behalf. To configure it and learn to use it on one’s own files, AI coaching for executives is often the fastest way.
Yes, like any tool based on a language model. It can misunderstand a question or, without safeguards, make up a plausible answer. Three precautions greatly reduce this risk: requiring the assistant to answer only from your sources, having it cite the documents used, and allowing it to answer that it does not know. For sensitive actions, human approval remains essential.
By placing it where they already work, in Teams, Slack or the CRM, and by starting with the questions that waste the most of their time. A pilot group sharing its first results is more convincing than an internal memo. Also plan a short AI training adapted to each role so that everyone knows what the assistant does well and what must always be checked.
Yes. The European AI Act requires informing people that they are interacting with an AI system, unless this is obvious from the context [3]. An assistant that answers your customers by chat or email must therefore state this clearly. For an assistant reserved for employees, information at launch and a usage policy are generally enough.
The market brings together general-purpose assistants, AI features built into software, turnkey assistants and custom development.
| Solution | For whom | Strengths | Limits |
|---|---|---|---|
| Kreante | SMEs and mid-sized companies with 15 to 200 employees | Assistant connected to your documents and software, sourced answers, access rights, code handed over | Requires scoping and a few weeks of development |
| Professional plan of a general-purpose assistant | All companies | Quality writing and analysis, immediate onboarding | Knows little about the company, few business rules |
| AI built into software | Companies already equipped | No extra tool to install | Limited to each software’s data |
| Turnkey subscription assistant | Very small businesses and small SMEs | Quick start on everyday tasks | Hard to adapt to your procedures and tools |
| No-code creation tools | Technically comfortable teams | Simple assistants built in-house | Maintenance, access rights and quality to manage yourself |
This list is not exhaustive.

Written by Jorge del Carpio, CEO and co-founder of Kreante. Since it was founded in 2020, the agency has delivered 265+ projects in 35 countries.
Published Updated
From the blog
Guides, comparisons and field notes from the projects we deliver.
DevelopmentPilot AI coaching for one to three teams over 6 to 8 weeks, measure workflow KPIs, and demand embedded deliverables to keep skills in house.
DevelopmentDiscover how to choose effectively between RAG and fine-tuning for your enterprise AI needs, ensuring rapid results and compliance.
ToolsWebflow AI Site Builder generates complete website layouts from text prompts, while the AI Assistant speeds up design tasks inside the editor. Learn how to use both tools to launch a professional site in a fraction of the time.
ToolsPick Lovable to get a hosted, Supabase backed live app fast. Pick Claude Design when engineering teams need code first prototypes and tighter code control.
DevelopmentMost AI automation projects show no measurable result. A Kreante project manager explains the sequence that works, with pricing and a real case study.
Insights & TipsOutcome first AI playbook for SMBs: use a 5-criteria scoring rubric, run a 4-week before and 4-week after task audit, and pilot in 6–12 weeks to prove ROI.