Frequently asked questions about our prices, timelines and your data.

A landing page starts at €900 excl. VAT, an AI-ready site at €2,900 excl. VAT, an AI audit at €1,500 excl. VAT and an AI Sprint at €4,500 excl. VAT. The final price depends on scope, which is agreed in writing before work starts. Every price is on the pricing page.

By comparing two numbers: the time the task takes today and the price of the workflow. If someone spends five hours a week keying in invoices, the automation is measured against those hours. The AI audit does that calculation process by process. Our prices are set in writing before we start, so the cost doesn't move along the way.

Not without your agreement. The price covers the scope written into the quote. If you add something, we send you an extra quote before doing it.

The AI Sprint is sold on the outcome: one workflow, with a scope we fix before work starts. We reuse building blocks we've already made, and the two weeks go into that single deliverable. With an FDE you pay for time inside your team on an open backlog that changes from week to week.

You pay the vendors directly: the automation tool, such as n8n or Make, and the AI model's API. For a small volume these subscriptions usually stay modest, but they depend on how many documents you process. We give you a costed estimate for your case during scoping, before we start, so you know what to expect.

Not this one. Bpifrance's Diagnostic Data IA goes through experts it accredits, and Klyx is not one of them. If that funding matters to you, talk to an accredited expert. If you want an answer within two weeks, at a fixed price and with no application to file, our private audit is built for that.

No. The audit is paid separately and the roadmap is yours. Part of the recommendations can often be done with the tools you already have. If you hire us for the next step, the quote takes the audit you already paid into account, and the terms are written down before you sign.

The AI automation price is credited 100% against an AI Sprint signed within 60 days. So you can start small, watch the workflow run on your real documents, then move to a fuller agent without paying twice for the discovery of your processes. After 60 days, the two stay separate fixed-price packages.

Allow 1 to 2 weeks for a landing page and 3 to 4 weeks for a 5 to 8 page AI-ready site. The schedule is written into the quote.

One to two weeks between the first interview and the review. Interviews take about an hour per person, and we ask your team to show us their tools the way they really use them. The rest happens on our side: mapping, scoring the possible automations and writing the roadmap. You have nothing to prepare beforehand.

1 to 2 weeks for the audit, then about 2 weeks to put a first automation in place.

The fixes are live within 2 weeks. Engines that search the web live, like Perplexity, can pick them up within days. Others rely on older data and take longer. That is why we run the test again at day 30, then every month if you take the GEO monitoring package.

Your accounts and tools stay in your name. For the AI to handle a request, data goes through a model provider's API, such as OpenAI or Anthropic, which by default doesn't use it to train its models. If you need EU hosting or a local model, we plan for it in the quote.

Yes, if it is designed to. Before writing code, we list the personal data the workflow handles, keep only what it needs, and set a retention period. Accounts stay in your name. If your data is sensitive, we choose EU hosting or mask some fields before they are sent to the model.

Rarely in a heavy way for an SME. Sorting email, extracting invoices or answering product questions are not among the high-risk uses. CV screening or assessing the creditworthiness of individuals are. An assistant that talks to your customers must say it is an AI, and the team using an AI tool must understand its limits. We plan for both. For legal advice, ask your lawyer.

Yes, if the data is processed within a framework you control. Accounts stay in your name, we can choose an EU-hosted model provider, mask names and amounts before sending, or keep some steps free of AI. We set that framework with you during the audit. Your profession's rules remain the reference.

It answers only from the data you give it, and says it doesn't know when information is missing. Before launch, we test it on real buyer questions. Every exchange is recorded in a log, so you can fix a property sheet or an instruction as soon as an answer isn't right.

Think of a task that comes back every week: reading email, re-keying documents, chasing clients. If it follows roughly the same steps each time, it can probably be partly automated. Book a 20-minute call: we pick one with you and tell you honestly whether it's worth it.

An agent or automation wired into your real tools and data, used by your team every day, with a log of what the agent does. A demo running on test data doesn't count.

An automation follows fixed rules: when an email arrives, it adds a row to the CRM. An AI agent reads, decides and acts within limits you set, for example sorting a customer request and drafting a reply. Many projects use both.

AI automation is one linear workflow: a trigger, an AI step that reads or drafts, an action. The AI makes no decision on its own and the workflow does not connect five tools. As soon as the process needs to choose between several actions, or act across several tools with a test set, it is an AI Sprint, delivered in two weeks.

That is planned for. We add a human approval step wherever a mistake is costly: the person sees the extracted fields, fixes them if needed and approves. The workflow also emails you when a run fails. Before launch, we test the AI step on 10 of your real documents to see where it gets things wrong.

Your team, using the runbook and onboarding included in the sprint, or us through the monthly run.

No, that is the case for most SMEs. We handle the technical side, your team keeps its usual tools and approves what the AI prepares. Every delivery includes onboarding and written documentation, and what we build stays with you: the code in your repo, the n8n or Make workflows in your accounts. Nobody needs to know how to write a prompt.

Yes, that is often where it pays back fastest. In a company of 10 to 50 people, a few repetitive tasks (quotes, invoices, follow-ups, email triage) take one or two people several hours a week. The audit finds them, puts a number on that time and checks that an automation costs less than what it saves.

A written document, not a slide deck: the map of your processes, the possible automations scored by hours saved, cost and risk, the GDPR points to settle and a 90-day roadmap. We go through it together in the review meeting. It belongs to you, and you can hand it to another provider if you prefer.

No. Most of the small companies we work with have no in-house developer. We write the roadmap for the person who decides, without jargon, and each recommendation says who can carry it out: you with a tool like Make or n8n, us, or another provider. When a point needs real technical skills, we say so.

You keep the process map and the roadmap, even if you go no further.

No. It answers the questions whose answer is on the property sheet, qualifies the request and offers a slot. As soon as a question needs an opinion, a negotiation or information missing from the sheet, it hands over to a negotiator, with the conversation history. Your team spends less time on the phone and more on viewings.

Usually, yes. Most portals send their requests by email, and the agent can read those emails. For the CRM, we use its API when it has one, otherwise a file import or a tool like Make. We check your tools during the call, before we put a price on anything, so you know what is possible.

Often, but not always directly. When the software has an API or a file import, the workflow drops the entries there for approval. Otherwise it prepares a file ready to import by hand. We check your software during the audit, and tell you what is possible before we price the workflow.

A little. They keep their usual tools and approve what the workflow prepares for them. We plan an onboarding session at delivery and short documentation written for them. If a staff member needs to change a step themselves, we show them how to do it in your automation tool.

Partly. GEO stands for Generative Engine Optimization: the work that makes a site readable and citable by AI engines such as ChatGPT, Perplexity, Gemini or Google's AI Overviews. SEO aims for a rank in a list of links. GEO aims for a place among the few sources an AI engine cites in its answer. The two overlap a lot.

No, and nobody can honestly promise that. AI engines change their answers from one week to the next and do not publish their rules. What we commit to is the method: a baseline, documented fixes on your pages, and the same measurement at day 30 on the same questions. You see what moved, with the numbers to back it.

No. AI engines rely largely on the same signals as Google: fast, well-structured pages with clear answers and verifiable facts. Good SEO is still the base. GEO adds direct answers at the top of the page, structured data, an llms.txt file and tracking of what AI engines say about you, question by question.

Usually, yes. The offer is built for existing sites. We need editor access to change the 5 key pages and add the structured data. If your tool does not allow it, we hand over the copy and the markup ready to paste, and you publish them yourself. We check your platform during the call, before you sign.

Maxime Mansiet, our lead engineer, co-founded CiteMe, a tool that measures how often a brand appears in AI engines' answers. We tell you so the link is clear before you sign. You don't have to subscribe: the report, the tracked questions and the fixed pages belong to you, and you can run the tests yourself.

A freelancer usually delivers a project defined up front. As an FDE, our lead engineer works inside your team on whatever matters most right now, joins your stand-ups and ships every week.

It takes one to two weeks to really learn your codebase and your team. Below four weeks, you'd mostly be paying for that ramp-up.

No. Three days a week is our maximum per client, so every engagement gets the same quality of work.

Remote by default. If your team is in Bordeaux, we can come to your office.

The AI Sprint delivers one defined workflow at a fixed price in two weeks. The FDE fits when your backlog keeps changing and you need someone inside the team for several weeks.

Next.js, TypeScript and Supabase for products, WordPress when your team wants to edit on its own. On the AI side: OpenAI, Anthropic and Gemini, MCP servers, and agents connected to your tools.

WordPress if your team wants to edit pages without going through us. Next.js if you want the fastest possible site or custom features. We decide together during the brief.

It depends on the process and on who will maintain it. n8n or Make work well when your team wants to change the steps itself. Custom code makes sense when you need fine-grained control, tests or high volume.

In your own n8n or Make account, in your name, so you stay in control. If you don't have a tool yet, we help you pick one and open the account. We also hand over the workflow export, which you can give to another provider if you ever change partners.

We send you a content brief, then structure and rework your copy with you. You approve everything before launch.

Yes. The code, the domain and the hosting accounts are in your name.

You edit your content yourself. For a new page or feature, we send you a quote.

Your project keeps running. The code is in your repo from day one, the n8n or Make workflows sit in your accounts with their export, the tools are in your name, and every project ends with a written handover. Any developer can take over from that documentation, without depending on us.

If you want, we can join meetings. How we are introduced is decided with you and written down before the project starts. Otherwise we stay behind the scenes and you carry the relationship with your client alone. These rules sit in the same document as the scope and the price.

Your team, with the written handover and the code in the repo. Follow-up by us, which you bill on, is discussed project by project. If we stop working together, the documentation lets another Next.js developer take the project over without depending on us.

Not many, and we say so before you sign. As an FDE, we take two clients at most at the same time. For an AI Sprint or a site, the quote gives a start date. If we can't meet your timeline, we tell you straight away rather than take the project and deliver it late.

Yes. Klyx is based in Bordeaux and we work remotely with teams across France and Europe, in French or English.

Maxime Mansiet, our lead engineer, from the first call to the handover. He hands well-bounded tasks to AI coding agents, and the code they write is reviewed before it goes to production. If the project needs design or a specialty, we team up with a specialist and tell you first. The code is in your repo from day one, and every project ends with a written handover so another developer can pick it up. For Forward Deployed Engineer work, we take on two clients at most at a time, and we don't offer 24/7 on-call.

No. We handle the technical side, and your team learns to use the tool. You don't have to install anything yourself.