We connect AI to your business.
And we keep it running.
A senior engineer embeds and builds an AI tool that produces in six weeks, and that grows with you, because we keep it in production.
- 01Diagnosis — 5 days, €6,000 excl. VATLivré
- 02Build — 6 weeks, fixed priceEn cours
- 03Operation — monthly subscriptionActif
Why so many AI projects stall before production
- 01The consultancy delivered a plan and a nice demo, then moved on to the next client.
- 02The IT firm billed days, with no commitment to a result or an end date.
- 03The freelancer ran the tool on their own laptop, then left with it.
- 04The prototype worked in a meeting room, never on real data or with real people.
- 05Nobody stayed to connect the tool, watch it, and fix it when it breaks.
- 06Six months later, nobody in-house knows how to run what was delivered.
What an embedded engineer changes
AI can transform your business: answering requests, reading your documents, preparing your decisions. It still has to reach production, and stay there.
A use case in production, not a report
The deliverable is a tool that runs, used by your team, with its documentation and access held by you. Not a demo, production.
A senior engineer, on your premises, with your business
Embedded in your rituals and your code, alongside the people who do the work. They understand the job before writing code, and are reviewed by colleagues.
A fixed price, not hourly time and materials
The diagnosis and the build are fixed-price, agreed up front. You are measured on the business result, not on hours. No running meter, no surprise.
We keep it running afterwards
The edge: once in production, we keep the tool alive and keep improving it, monitoring, updates, security. That is what makes it still producing a year later.
Pick an input. See what AI does with it.
Cases in production — and the scenarios we know how to build, on real businesses.
Odoo ERP — In production — U'Wine, Wine trade. Assistant that guides orders
Your systems
- Odoo ERP
- Vintage catalogue
- Customer history
MCP · Memgraph graph
- queries the Odoo ERP
- reads the graph (vintages, tastes)
- recommends, in production
Assistant that guides orders
Most AI projects never reach production. Ours run, and keep producing, because we operate them.
The method, in three stages
Diagnosis
€6,000 excl. VAT5 days- Your data and processes mapped
- The projects worth doing
- The ones to stop
Build
The core of the offerFixed price6 weeks- A written scope
- An end date
- A price agreed up front
Operation
Subscriptionmonthly- Monitoring, updates, security
- Evolutions
- Or skills transfer and documentation
A tool that lasts, and keeps improving
Most AI projects stall before production. Ours run, and keep producing, because we operate them: managed services, monitoring, security. You leave with an asset that grows, not a demo.
A model connected to your data is alive: sources evolve, uses expand, the model improves. By keeping it in production we do more than keep it alive, we improve it over time. A consultancy advises, an IT firm places days, a studio delivers then leaves; we build the tool and grow it under the same roof. Nobody else on this market connects the AI and operates it together.
What it costs
Fixed prices, announced before we start:
- Diagnosis: fixed €6,000 excl. VAT. Five days to map your data and processes, separate the real from the decor, and leave with the projects worth doing and the ones to stop.
- Build: fixed price per project, agreed up front. No hourly time and materials, no meter, no overrun. A written scope and an end date.
- Operation: monthly subscription depending on scope: monitoring, backups, updates, security, evolutions.
To start quickly, an engineer on time and materials at €600 per day stays available as a bridge. But the offer is fixed-price, and that is precisely what separates an embedded engineer from plain time and materials. For eligible French SMEs, a scheme like Bpifrance's Data AI diagnostic funds around 40% of a diagnosis of this kind: we help you check your eligibility, without yet being an approved provider. The first conversation is always free.
What it is not
It is not raw staff augmentation or an isolated contractor: the engineer comes with Kapsloc's methods, tooling and lessons learned, and is reviewed by colleagues. It is not AI strategy consulting either: we do not deliver a plan, we deliver a tool that runs, and the plan comes afterwards, once we have seen what works.
At night too
By day, the engineer works with your teams; at night, our AI agents can carry the work forward, each proposal checked by another agent before a human reads it in the morning. That is the night team pilot: four weeks, €1,500 excl. VAT.
Our commitment
We separate the real from the decor, even when it costs us a sale. We deliver a tool that runs, not a slide. We keep our word on scope and date. And we stay: operation is what makes the tool still run a year later.
References
- The market shifts to the forward deployed engineer model, SFEIR analysis, 2026.
- The Model Context Protocol, the open standard we use to connect models to your tools.
- The EU AI Act, to know what applies to your use case.
Engagements delivered in this format
- For U'Wine, a wine-investment company: an AI assistant connected to their Odoo ERP through the MCP protocol, with a Memgraph knowledge graph as its memory. In production, on Claude, continuously enriched, and being opened to their end customers.
- For the CPTS Haute Gironde: a health-orientation platform, knowledge graph and an assistant that cites its sources, running on Claude, tested by six panels of front-desk staff, with a regional funding application drafted right after.
- For IMING: SmartCharging, charging optimisation for electric bus fleets. A notebook tool turned into a web platform (Python engine, simulated annealing and genetic algorithm, API and web interface), delivered in four phases, production rollout under way.
Questions we are asked about this format
What is a Forward Deployed AI Engineer?
It is a senior engineer who comes to work inside your team to connect AI to your real tools, real data and real people, deliver a use case in production, then keep it alive. The format was invented by vendors of complex software for their largest clients; Kapsloc brings it to the scale of an SME, fixed-price and with operation on top.
Why do most AI POCs fail?
Because nobody stays to make them last. The consultancy advises and leaves, the IT firm bills days, the freelancer moves on. The tool is delivered, then data changes, a dependency ages, a model drifts, and with nobody to watch and repair, it dies in silence. Our operation offer exists precisely to prevent that.
Why fixed-price rather than hourly time and materials?
Because a fixed price commits us to a result and a date, not to an hour meter. The diagnosis is €6,000 excl. VAT, the build is a fixed price agreed up front, operation is a monthly subscription. Time and materials at €600 per day stays possible as a bridge to start quickly, but the headline offer is fixed-price: that is what separates an embedded engineer from plain staffing.
What if the chosen use case does not work?
We know by the second or third week, thanks to the first version and the error measurements. Two options: we adjust the scope, or we switch use case. And the diagnosis is there for that upstream: we separate the real from the decor and tell you which projects to stop before you spend.
How is it different from a freelancer on time and materials?
A freelancer on time and materials arrives alone, with only their own tools, and leaves alone. Our engineer arrives with Kapsloc's methods, tooling and lessons learned, is reviewed by colleagues, and leaves you a tool that runs and a team that can carry on.
A use case in mind?
Thirty minutes, free, to identify the first use case together and check that the format suits you.