Custom AI Agents
Agents for the internal work that eats your week — quoting, intake, reconciliation, reporting.
We price it on the audit and put it in writing before we start. Usage charges from third-party platforms, such as per-minute call rates, are separate and vary with volume. Book the audit to get a figure.
The problem, plainly
The expensive work is rarely customer-facing. It is the quote built by hand, the invoice keyed twice, the report rebuilt every Monday.
What we build
We identify the two or three processes with the highest hours-per-week cost, then build agents that execute them end to end against your real systems. Each agent has a written scope, a permission boundary, an audit log and a human approval gate wherever money, contracts or patient data are involved.
What this service moves
12–30h
Weekly hours returned
100%
Actions written to an audit log
2 weeks
Shadow period before cutover
These figures are design targets and illustrative examples, not guarantees. Results depend on your business and on the third-party platforms involved. Your baseline is measured during the audit, before anything is built.
What changes
What it is designed to deliver
- Hours returned to the people you would otherwise need to hire
- Consistent output on work that currently depends on who is on shift
- An audit trail for every action an agent takes
- Approval gates on anything financial, legal or clinical
What is included
Everything in the deployment
- Process audit with an hours-and-error baseline
- Least-privilege service credentials, scoped per system
- Structured audit log of every action and input
- Human-in-the-loop approval gates where they matter
- Regression test suite run before each change ships
- Runbook and handover so your team can operate it
How it gets built
Four stages, scoped in 1 week, built in 3–6
Nothing is a black box. You sign off the scope before we build, and you review the test log before anything goes live.
- 01
Time the process
We measure the current process in hours and error rate, so the business case is arithmetic, not a pitch.
- 02
Draw the boundary
Exactly which systems the agent may read, which it may write, and what always requires a person to approve.
- 03
Build against staging
Nothing touches production data until the agent passes a scripted test suite you sign off.
- 04
Shadow, then switch
The agent runs alongside the human process until outputs match, then takes over with monitoring.
Often deployed together
The systems that sit either side of this one
AIOS dashboard
All your operations data captured and visualised on a single dashboard.
Scoped in 1 week, built in 2–4Smart chat agent
Site chat that answers from your documents and books, instead of collecting emails.
Live in 5–7 daysVoice receptionist
Answers on the first ring, at 2am, on public holidays, during the rush.
Live in 10–14 days
No obligation · no sales sequence
Start with custom AI agents?
The audit tells you whether this is the right first system for you. If it is not, we will say which one is. Scoped in 1 week, built in 3–6 from a signed scope.