We design and integrate custom AI agents and workflow automation for SMBs who want to save time without losing their voice. An independent design studio in Paris, where the brand comes before the agent.
Saved / week on average
First agent in production
GDPR, EU or France hosting
Raw Make workflows, white-on-white chatbots, dashboards that look like Excel sheets. Technically it works. Visually, it kills your brand.
Design studio first, AI second. Branding before the agent, interface before the prompt, voice before the LLM.
An agent that speaks in your brand's tone, lives in a designed interface, strengthens your image instead of diluting it.
Identity, site, AI agent.
One studio, one system.
A chatbot and an AI agent are two different things. The first follows a script. The second reasons, acts, and learns from your context.
A chatbot is a dressed-up decision tree. An AI agent is a digital collaborator who knows your business and acts on its behalf.
Six cases that come up in most of our projects.
An agent that reads each incoming request, asks 2-3 targeted questions, and routes the lead to the right bucket.
An agent plugged into your knowledge base that answers recurring questions in the right tone.
An agent that follows up politely at the right moment, with a message tailored to each customer.
An agent trained on your existing articles that writes first drafts consistent with your line.
An agent that reads your meetings, KPIs, Slack channels and produces a weekly brief for the leadership team.
An agent that knows your procedures, your Notion, your HR processes. End of "can you resend me the template contract?"
Outdoor e-commerce scenario
Composite case · anonymised data
Tell us where the project stands. We come back with one recommendation and the order we would do things in. The call is free and commits you to nothing.
(04)
We map your processes and rank three quick wins. You walk away with a 2-page PDF: diagnosis, ROI estimate, concrete recommendations. Whether you sign with us or not, the deliverable is yours.
We define tone, rules, interface. We write prompts the way we write a brand book: with method, with taste, with an eye on your identity. Your agent should sound like your brand, not like a generic SaaS.
We code. We connect to your CRM, Slack, Gmail, Airtable, Notion, whatever runs in your shop. We test with your real data, not a demo dataset. You follow progress week after week, not by discovering a demo at the end.
Go-live, reinforced monitoring for the first 10 days, team training, written documentation. You're autonomous. If you want us to stay, an AI Care subscription takes over. Otherwise, you walk with everything.
Most AI projects stall after three months for lack of maintenance. AI Care is the contract that keeps your agents useful over time.
Real-time alerts if an agent breaks, slows down or drifts. We know before you do.
We check where the agent hallucinates, where it underperforms, where it wastes time. And we fix.
Prompts are never "done". We rewrite them based on new cases, new data, team feedback.
Claude 4, GPT-5, Mistral Large: when a better model ships, we switch. No surcharge, no downtime.
Monitoring detects, an alert reaches us. Essential: fixed within 48 business hours. Studio: fixed within 12 business hours, with automatic compensation if we miss.
Yes, at no extra cost. We handle the switch as part of monthly optimisation. API costs always go to your account, not ours, so zero conflict of interest.
Creating an entirely new agent, architecture overhauls, or adding a channel (WhatsApp, Teams, voice) are billed separately as missions. Studio includes 4h/month of evolutions absorbing small requests.
After the first 3 months of commitment, yes, with 30 days notice. No complicated retention clause. If we don't do the job, you leave.
What every contract guarantees on data and sovereignty.
DPA signed for each mission, declared sub-processing, registry kept up to date.
European servers by default, French sovereignty available (Mistral, OVH, Scaleway).
Providers with "zero training on your data" clauses (Claude API, private Mistral).
Complete data flow review once a year, written report, proposed fixes.
Emmanuel Lamare
Co-founder, Pyrene Performance®
We needed a tool that spoke to our athletes in Pyrene's language, not a generic chatbot. Richard designed the agent the way he designed the brand: coherent from the first word to the last. We plugged it in in 2 weeks, weekends included.
Florian Billard
Co-founder, Pyrene Performance®
What we gained is hours. Ingredient questions, pre-race nutrition tips, post-purchase follow-ups: the agent handles them. We can finally spend our evenings improving products instead of answering the same question forty times.
Laurent B.
Owner, Domaine Le Courti®
Running a guesthouse is 90% repeat questions and 10% real conversations. The agent filters, replies instantly on rooms, dates, nearby restaurants. I got my evenings back, and guests never wait.
Full case study
Pyrene Metabolic Scan
Four sectors where a generic agent falls short.
On an €80 basket, a generic agent smells like a ChatGPT template. The buyer feels treated like volume, not like a premium customer.
An agent that qualifies, advises, reassures, follows up. In the brand's voice, not the LLM's.
80% of inbound (bookings, room questions, nearby restaurants) is repetitive and eats into the owner's evening.
80% handled as if it were you, 20% escalated where your presence actually makes the difference.
One coach can't answer 200 athletes, 24/7, on training plans, nutrition and follow-ups.
An extension of the coach's methodology, with the same precision and tone. Never a substitute.
3-month cycles, technical quotes, multi-step follow-ups, too many cold leads to warm up by hand.
The agent filters, qualifies, nurtures. The team focuses on leads worth closing.
An AI agent is a software assistant built on top of an existing model (Claude, GPT, Mistral) that performs concrete tasks for your business: qualifying leads, answering customers, sorting documents, chasing quotes. Unlike a generic chatbot, it knows your business context, speaks in your brand tone, and connects to your tools (CRM, Slack, Gmail, Airtable).
A chatbot follows fixed scripts and returns pre-written responses. An AI agent reasons, reads your data, makes decisions and triggers actions in your tools. The first is a dressed-up decision tree. The second is a digital collaborator.
We don't publish a price list: an agent that answers FAQs and one that qualifies leads across three tools are not the same project. We scope your case in a free 15-minute call and send a precise quote within 48 hours, as a one-off mission or an AI Care monthly plan. APIs (Claude, GPT, Mistral) are billed directly by the provider, on your own account.
We use existing market models: Claude (Anthropic), GPT (OpenAI), Mistral, Gemini depending on the use case. We don't create proprietary models. We're integrators, not R&D lab. Model selection happens case by case based on task, cost and confidentiality constraints.
Yes. We sign a GDPR data processing agreement for every mission. Sensitive data can be isolated in dedicated databases. We prioritise providers with no-training-on-your-data commitments (Claude via API, Mistral hosted in France if required). European or French server hosting available depending on your sensitivity level.
API costs are billed directly to you by providers, on your own account, without hidden markup. Typical range: €30 to €150/month depending on volume. You keep control of your usage and can switch models at any time.
A simple agent (lead qualification, enhanced FAQ, message sorting) can be delivered in 2 to 3 weeks. Multi-tool automation integrated with your CRM and internal channels takes 4 to 8 weeks depending on complexity. The free audit provides a precise estimate.
Yes, after a preliminary audit of your existing setup, deducted from your first month if you subscribe. The audit identifies necessary fixes and ensures we can confidently maintain agents we didn't build. We scope it on the 15-minute call.
Yes. The studio is based in Paris but most projects run remotely. We have clients across Europe, with regular video check-ins and occasional on-site visits for kickoffs or team training.
Pyrene®
French nutrition for ultra-endurance.
Hoooz®
Multilingual SaaS marketing site.
Menao®
B2B app for artisans and craftsmen.
Pace®
Triathlon and individual sports coaching.
Le Courti®
Seaside guest house.
Ultra Bike France®
French ultra-cycling event series.
AI & Innovation Anthropic built the most capable creative AI ever released, then a US government directive erased it from every account in 72 hours. The lesson isn't about the model. It's about what happens when your creative stack can be switched off by someone else.
AI & Innovation On June 17, xAI shipped Grok Imagine Video 1.5 at 4.20 dollars a minute. Production quality just stopped being the moat. The thing AI video still can't do is hold a brand steady across a campaign, and that gap is the new craft.
AI & Innovation On May 20, Figma rolled out an AI agent that lives inside your design file. Canva pushed its brand kit into Gemini the same week. The new moat is no longer the tool you license. It's the brand system the agent reads from when it works for you.