AI agents & automation
AI that does the work, not AI that demos well. We build agents into the systems you already run, with a person in the loop where it matters.
Click to send a request
Who it's for
You are probably here because
-
01
Two staff spend most of the day answering the same twelve questions.
-
02
Invoices and receipts are being retyped from one system into another.
-
03
Nobody can find anything in five years of contracts and tenders.
-
04
You have run a pilot that impressed everyone and shipped nothing.
What's included
What you actually get.
The hard part is never the model. It is the connection to your CRM, the guardrails, the handover to a human, and the log of everything it could not answer.
Customer-facing agents
WhatsApp, web chat and phone agents that answer from your own documents and hand off to staff cleanly.
Back-office automation
Invoice and receipt extraction, procurement matching, report generation, data entry between systems.
Document intelligence
Search and question-answering across contracts, policies, tenders and records. Retrieval over your own corpus.
Operational copilots
Internal assistants for sales, HR and support, wired to your CRM, ERP or ticketing.
Private model hosting
Open-weight models on your own hardware when data cannot leave the building. Pairs with GPU clusters.
Evaluation and guardrails
Test sets, accuracy tracking, fallbacks, audit logs and cost ceilings before anything goes live.
How it runs
Each stage, and what it produces.
-
01
Use-case workshop
We score candidates on value and feasibility together.
Output
A shortlist with a written recommendation, sometimes 'not yet'.
-
02
Pilot
One use case, real data, measured against a human baseline.
Output
A working pilot and a go/no-go number.
-
03
Production
Integration, guardrails, monitoring and staff training.
Output
A live system and a runbook.
-
04
Operate
Accuracy review, prompt and model updates, cost reporting.
Output
A monthly report on accuracy and spend.
What you are handed
- An evaluation set you own, so accuracy is measurable later
- Audit logs of every answer given and every handover
- A documented cost ceiling and what happens when it is hit
- Staff training on when to trust it and when not to
Note panel
Deployment options, plainly
Hosted API. Fastest to stand up, lowest fixed cost, your data leaves the building under the provider's terms.
Private cloud. Your own tenancy and keys, data stays in a region you choose, moderate fixed cost.
On-premise. Open-weight models on your own GPUs. Nothing leaves the building. Highest fixed cost, lowest marginal cost, and the only option when a contract or a regulator forbids the other two.
Worth saying
When we say don't
If the task is rule-shaped and stable, a script is cheaper, faster and easier to audit than an agent.
If nobody can say what a right answer looks like, there is nothing to measure and the pilot cannot pass or fail.
If the underlying data is wrong, an agent will simply be wrong faster. Fix the data first.
The questions people actually ask first
“Will it leak our data?”
Three deployment options, and you pick: a hosted API, a private cloud instance, or fully on-premise on hardware you own. We will tell you what each costs to run before you choose.
“Will it make things up?”
Answers are grounded in your own documents and cite what they came from. Anything that triggers an action goes through a human approval step. Everything it could not answer is logged for you to read.
“What does it cost to run?”
We estimate token or GPU cost during the pilot, set a monthly ceiling before go-live, and report actual spend against it every month.
“Will it replace our staff?”
It removes repetitive work. People stay in the loop on anything that commits the business. Every deployment we have run has changed what staff spend the day on, not how many there are.
Technology
Proof
One project, in full.
On-premise AI cluster for a research institution
Research data that could not leave the country, on cloud GPU hours nobody could afford.
Overnight
Job turnaround
Research data could not be sent to overseas cloud providers, and cloud GPU hours were unaffordable at the volume required. Jobs queued for days, and researchers had begun to shrink their experiments to fit …
Overnight
Job turnaround
Eight
GPUs on site
On premise
Data residency
Commercials
How buying this works.
Engagement
Paid use-case workshop, then a fixed-scope pilot, then production.
Price
On request, after scope
What sets the price
Number of systems it must connect to, how much of your content needs preparing, whether it runs on a hosted API or your own hardware, and the accuracy bar you set.
Token or GPU cost is estimated before the build and monitored after it, so the running cost is never a surprise.
Every project is quoted after we understand the scope. You will have a written, fixed quote before any build work begins. No open-ended billing.
Questions
Answered plainly.
The pilot is deliberately narrow: one use case, real data, measured. You see it running on your own material rather than on a demo.
Yes for understanding and for most replies. We test this during the pilot with your own message history rather than assuming it.
It says so and hands over to a named person, and the question is logged. That log is the most useful thing the first month produces.
Only if your data cannot leave the building or your volume makes per-token billing more expensive than hardware. We will do that arithmetic with you before you buy anything.
You do. They transfer with the rest of the project.
Often bought alongside
Build
Talk to an engineer about one use case
Bring the task that eats the most staff hours and we will tell you honestly whether an agent should touch it.
We reply to every enquiry within one working day.