AI for SMEs, connected to your businessNot another demo: one use case in production, inside your tools, within six weeks.

You tried a general-purpose AI assistant and it knows neither your customers, nor your procedures, nor your documents. Value appears when the model is connected to your real data, framed by your rules and integrated where your teams already work. That is what Kapsloc does, with SMEs and public bodies in Nouvelle-Aquitaine, since 2025.

Book a free diagnostic

Your situations

The situations we meet

The same questions arrive every day by phone or e-mail, and the answer depends on who picks up.

Dozens of pages to read for every bid, quote or tender, with scoring that varies from one reviewer to the next.

A valuable internal tool that only runs on one person's computer, with no versions and no tests.

A rich document base that nobody can search: binders, PDFs, browser tabs, the memory of three colleagues.

Invoices and delivery notes retyped by hand into the management software.

A monthly AI subscription used by two people, with no measurable effect.

What we deliver

Five ways to put AI to work

Assistants connected to your systems

ERP, CRM, document base, e-mail: the assistant reads, suggests and prepares; your teams validate. Nothing leaves without a human.

Agents that act inside your tools

An agent reads an order e-mail, checks stock in the ERP, prepares the quote and submits it to your sales rep. It acts within a written scope, and every action is logged.

Searchable knowledge bases

A graph or an index of your documents, with plain-language search that cites its source and date, and says so when it does not know.

Automatic document reading

Invoices, delivery notes, tender bids: extracted or scored against your grid, then reviewed and corrected by the person in charge.

Turning prototypes into products

A notebook, a macro or a fragile internal tool becomes a maintainable, tested, deployed application whose code you keep.

How it works

Six weeks, four steps

01
We pick the use case
A frequent request, a document that eats time, a poorly informed decision. We size it together, in time lost today and result expected.
02
An engineer joins your team
Two to three days a week, with the people who do the job. They connect the model to your data, inside your tools.
03
First version by week 2
Your team uses it, we measure the errors, we adjust. No development tunnel, no hundred-page report.
04
In production by week 6
The use case runs, the documentation is written, your team is trained. You keep the code and the access.

Our principles, in six lines

AI assists, it does not decide. Every suggestion is reviewed by a person: this is what keeping a human in the loop means.
Every answer cites its source and date. When the base does not know, it says so instead of inventing.
No personal data leaves without need, and never without your knowledge, in line with the CNIL's guidance on AI.
The model is a choice, not a religion: hosted in France or in Europe when required, and replaceable.
Connected through open standards. We link models to your tools with the Model Context Protocol (MCP), the open standard adopted by the major vendors, not with a proprietary connector that would lock you in.
You stay the owner of the code, the access and the data. Short notice, no lock-in.

What an SME can concretely expect

Two examples, taken from recent projects:

A front-desk team gets the right service in under thirty seconds, with phone number, address and access conditions, from a request phrased as on the phone.
A buyer applies the same scoring grid to every bid, with a justification per criterion to review before deciding.

What it costs

The engagement: between twelve and eighteen engineer days depending on scope, at a daily rate announced before we start.
Running costs: hosting and model calls included, from a few tens to a few hundred euros per month depending on volume. A document assistant for ten people costs less than an office subscription; an agent reading thousands of documents a month is priced before we start.
Funding: for French companies with 10 to 2,000 employees and over one million euros of revenue, Bpifrance's Data AI diagnostic, part of the national Osez l'IA plan, covers 40% of an assessment and action plan. We build the application with you.

And afterwards

An AI solution lives on: sources change, models evolve, usage grows. Kapsloc operates, monitors and evolves it as part of its managed-services offer, with the same contact as on day one.

That is what sets us apart from an agency that delivers a prototype and moves on to the next client.

Useful references

The EU AI Act: what a company needs to know about transparency duties and high-risk uses.
The Model Context Protocol: the open standard that connects an AI model to your tools and data.
The CNIL's guidance: personal data and AI, what is allowed and how to do it properly.
Osez l'IA, the Bpifrance plan: funded diagnostics and support for SMEs.

Proof

What we have already delivered

An orientation platform for a professional health community in Gironde: a graph of 723 services, organisations and venues, queried in plain language by front-desk teams, with an assistant that cites its sources.
A tender-analysis tool for an engineering office: the mandatory scoring grid applied to every PDF bid, a justified score out of 100 reviewed by the buyer.

Frequently asked questions

Questions we are asked

Do we need large amounts of data to start?

No. Most use cases that help an SME rely on documents you already have: procedures, catalogue, e-mail history, customer base. We do not retrain a model, we connect an existing model to your information, which works from a few dozen documents.

Does our data go to OpenAI or elsewhere?

Only if you accept it, and only what is needed. We can use models hosted in France or in Europe, or a model installed on your own server for sensitive data. Documents read by your teams stay with you; the model only receives what it needs to answer.

What happens when the AI is wrong?

It will be wrong sometimes, which is why it never decides alone. Every suggestion is reviewed by a person, cites its source, and the tool says when it does not know rather than inventing. We measure the error rate with you during the engagement and adjust before production.

How long before a visible result?

A first usable version in the second week, the use case in production by the sixth. If a project needs more, we say so before starting, not after.

Do we have to train our teams?

Half a day is usually enough, because the tool is designed with the people who will use it and integrated into their usual software. Training is part of the engagement, with a short guide written for your team.

What if we want to stop or change provider?

The code, documentation and access are yours from day one. The operations contract, if there is one, ends with a short notice period. You can take the tool in-house or hand it to someone else without asking us.

Are we affected by the EU AI Act?

Since August 2026, any company deploying an AI system has transparency duties: telling people they are interacting with an AI, keeping a human accountable for decisions, documenting the use. High-risk uses, such as recruitment, credit or health, carry additional requirements. The use cases we deliver to SMEs almost always fall under minimal or limited risk; we check it with you at the start and write the accompanying documentation.

Let's talk about your use case

Thirty minutes with an engineer, no commitment: together we identify the quickest use case to put into production.