AI agents
Agents that keep working after they are delivered
We design, build and host AI agents, chatbots, knowledge bases and automated workflows, and the same people keep them running afterwards. An agent is never finished: models change, data shifts and the output has to stay monitored. We also build our own agents; our own work runs on them.
What we build
AI agents
Agents that carry out a task on their own, within limits set in advance: which data they may use, which actions they may take, and who reviews the outcome.
Knowledge bases and chatbots
An assistant that answers from your own documents. Your knowledge base is kept separate from other clients and is deleted or handed over to you at the end.
Automated workflows
The work that is now retyped between systems or prepared by hand. The steps that need judgement stay with a person.
Locally hosted models
Where personal or production data is involved, the models run locally or on infrastructure we host ourselves, not on a third-party AI service.
How the work is controlled
What makes an AI project different is in the contract
- AI output is a proposal. Output may be inaccurate, incomplete or outdated, and may vary from one run to the next. We say so up front. Where an outcome affects people, a person decides, not the agent.
- Models change, the agent keeps working. Providers change, replace or withdraw models. We then deploy an equivalent alternative. If that materially changes the result or the cost, we consult you first.
- Your data trains no models. Without your prior written consent we do not use your data to train or improve models, and knowledge bases holding your data are kept separate per client.
- The AI Act is part of the design. If an application could qualify as prohibited or high-risk under the AI Act, that belongs on the table before work starts. AI literacy of your staff who work with the agent is your responsibility, as the Act requires. Applications we judge unlawful or unethical, we decline.
How we work
Plan the work, then work the plan
Contract review
Before we quote, we determine whether we can actually meet what is being asked, including the statutory and regulatory requirements you may not have stated, and which side of the hosting boundary the engagement will sit on. If the answer is that we are the wrong supplier, that is a legitimate outcome and you hear it in week one.
Design and development
Planned stages with defined inputs and outputs. Review, verification and validation are separate recorded activities. Security and accessibility are requirements in the same list as the features, not a phase at the end.
Release acceptance
A defined acceptance against the recorded requirements, with nonconforming outputs handled explicitly rather than shipped with a note. Nothing reaches production on the strength of a verbal yes.
Go-live and service transition
The workload enters the hosted estate inventory with its backup schedule, monitoring, restore test date and named owner. This is the step most suppliers skip, and it is the one that decides what the next three years feel like.
FAQ
Output an agent generates specifically for you belongs to you, to the extent rights can exist in it, and you are free to use it in your own business. The agent itself, with its prompts, models and workflow configuration, remains ours; you receive the right to use it. Without our consent the service and its output may not be used to train other models, and the agent is not to be copied, rebuilt or integrated into other AI systems. If you want to own an agent, or run it yourself or at your own customers, we agree a separate agreement and fee for that. We say so up front because it differs from client work, where the code becomes yours.
Yes. Personal and production data is then processed with models we host locally or ourselves, not on a third-party AI service, unless we agree otherwise in writing. A knowledge base holding your data is kept separate from other clients and is deleted or handed over to you at the end.
Besides building and running it, there are consumption costs such as tokens, calls and compute time. We pass those on unless agreed otherwise, and we can set usage limits so the costs do not build up unnoticed.
Tell us which work takes too much time
That is usually the most useful first conversation. Whether the answer is a new system, an AI agent, an integration between what you already have, or nothing at all, you get an honest assessment.