We build the AI systems that survive production
Agents that do real work, LLM applications you can evaluate, retrieval that cites its sources, and the data and cloud foundations underneath. We build this for clients, and we run four products of our own on it — which is where most of what we know comes from.
Software we build and run
AI systems, and the foundations they need
Agents, language models and retrieval on top — data platforms, cloud and security underneath, because that is where AI projects actually fail.
What working with us
is actually like
No claims here you cannot hold us to in the first month.
We build our own products
one3.ai, SalesBrace, adrox and a2lp are ours. We operate them, carry the pager for them, and learn from them. That is a different kind of experience from having delivered a project and left.
Engineers, in the room
You talk to the people writing the code. There is no account layer translating your requirements into a brief and translating the result back.
We start with how you will measure it
Before the build, we agree what success looks like and how it gets measured. Without that, an AI project is impossible to evaluate — which is why so many quietly continue forever.
Security is part of the build
Permissions, audit trails and the specific risks AI adds — prompt injection, over-scoped agents, data leaking through a model — are designed in, not reviewed afterwards.
We will tell you not to build it
Sometimes the honest answer is that a query, a rule or fixing your data solves the problem, and a model would be an expensive way to be less reliable. We would rather say that early.
Handover is part of the work
Documentation your team can operate from, runbooks written before the incident, and a period where we run it alongside you rather than throwing keys over the fence.
Where we think this is going
The interesting shift is not that models can write. It is that software can now take actions on your behalf — and that makes permissions, audit trails and human oversight the difference between an agent that helps and one that quietly does damage. We build for the second case, because it is the one that arrives eventually.
Where we have done this work
The problems change by sector, but the failure modes rarely do — bad data, no evaluation, and nobody owning the system after launch.
All industries
Software we build and run ourselves
Each one started as a problem we kept meeting in client work. You can go and use them — which is a more useful kind of proof than a case study.
All productsone3.ai
The whole back office, with AI agents you can rent by the month
One connected suite covering ERP, CRM, accounts, HRM and video — with a library of AI agents that work inside those modules and can be rented rather than built.
SalesBrace
A complete ERP for businesses that outgrew their spreadsheets
Full enterprise resource planning — inventory, purchasing, sales, production and finance — built to be implemented in phases rather than in one traumatic cutover.
adrox
Digital marketing automation, end to end
Software that runs the mechanical half of digital marketing — campaigns, audiences, sequences, publishing and attribution — so the team spends its time on the half that needs judgement.
a2lp
Landing pages and websites assembled by agents, per visitor
Landing pages and sites where AI agents compose what the visitor sees, following the A2UI approach — the interface is generated for the context rather than authored once and served to everyone.
Platforms and models we build on
Tell us what is actually in your way
Bring us the process that is costing you the most time, or the AI project that has stalled. We will tell you how we would approach it — including when the honest answer is that you do not need us.