AI agents, with a person in charge
Teams of agents, built around how your business runs. Each one reads what arrives, checks it against your rules and records, and takes only the next step it is allowed to.
IBI's built-to-order agents are coming soon. This page is the standard they are designed to.
Already at work.
Agents IBI built into its own product, DraftRight, and for The Coaching Corner Academy, running today.
Voice, on a live call
Joins the call, follows the conversation and advises what to say and which document to use.
Contract review
A five-pass review that scores risk against UK standards, flags red lines and maps each finding to legislation.
Agreements from a conversation
Takes a recorded negotiation and drafts a numbered UK agreement.
Transcripts
Audio and video to speaker-labelled text, with timestamps.
A client to practise with
Coaches in training talk with an AI client by voice, then get a review against the ICF competencies.
Built to order, for any business.
The brain is designed to pick the agents your work needs, so a housing association, a clinic and a law firm each get a different team. Each agent is built after the audit, for one measure it found.
- 01Sorting
Messages labelled as they arrive: an enquiry, a booking change, a complaint, a repair, a payment.
- 02Reply drafts
Replies written into your mailbox for a person to read and send.
- 03Complaints
A message that reads as a complaint, flagged for a person to log.
- 04Website enquiries
Visitors answered only from your own pages and the facts you approve; times offered, and one booked when the visitor confirms.
- 05Your CRM
Each enquiry written up as a brief, in any CRM with an API.
- 06Payment reminders
Reminders drafted to chase what is owed.
- 07Booking changes
The reply to a request to move or cancel, drafted for a person to send.
- 08A weekly report
The week's operations, reported for you to read.
- 09For housing providers
The Awaab's Law written summary, drafted from the investigation record for an officer to send.
- 10Inside your own AI tools
Some of these, built in Microsoft 365 Copilot, Google Gemini, ChatGPT or Claude, under your own administrators.
Built from the brain down.
- The brain
- Your rules
- The agents
- Where it works
- A person in charge
Reads what arrives.
Checks it against your rules and records: crisis, risk, facts, privacy, links and hidden instructions.
Each takes only the next step it is allowed to.
It reads and writes only in the mailboxes and systems you name. Your IT team can revoke it at any time.
A person you name is in charge of every agent and can stop it. Every action is logged for you to read.
Then it gets to work: the same blocks, in the order a message moves through them.
Three ways to run it.
One codebase; where it runs is configuration. The same tests run against all three before every release, and a release that fails on one does not ship.
For small businesses that want it run for them. IBI's cloud, in Ireland by default or the UK where your terms require it. Each business is kept apart by its own id, row-level rules and its own data key.
For clients who want their own database without running it. IBI's cloud, by the same region rule, with one database and one worker for you alone.
For regulated buyers, and any IT team that says "inside our tenancy". Your own Azure subscription, Google Cloud project or a server you control, from IBI's signed images. IBI never holds a token.
Every output, checked.
Every output passes these before a person sees it. If a check errors, times out or loses its model, the item is held for a person and an alert is raised. No failure lets output through.
- 01Crisis rule
A message read as a crisis gets no AI output. A named person is alerted.
- 02Distress screen
Built to catch too much rather than miss one, with every exception pinned by a test.
- 03Two-model risk check
Two independent models must both clear a message.
- 04Nothing invented
Every price, date, credential, policy or promise must match your approved facts.
- 05No one else's details
No email, phone, name or reference of any other customer of yours can appear in a reply.
- 06Your links only
Links come only from your own list.
- 07Hidden instructions ignored
Text read from mail, forms or pages is treated as material, never as an instruction.
- 08Always says it is an AI
Every conversation opens by saying so, and every draft carries a note to the person who will send it.
Risk levels are set when you join and re-checked by the audit: high for practices, housing, and anyone whose customers may be vulnerable; a business cannot lower its own level without IBI's review.
Who decides what.
- Sending an email
- Changing your records
- Any decision about a named person
- Every draft, sent from their own mail client
- Every new mailbox or system it reads
- Every release in your own cloud
- Reading the mailboxes in scope
- Drafting, screening and sorting
- Health checks
How an agent goes live.
- 01The audit's finding
Each agent is built for one measure the audit found, such as how long replies wait.
- 02A pilot
Tested on a labelled sample of your own work, kept inside your boundary.
- 03Live, predicted
What it should change is written down before it ships.
- 04Day 30
The same measure, run again.
- 05Day 90
Again. If it did not move the measure, we say so.
What our agents will never do.
- 01Email your customers on their own
They write drafts; a person sends.
- 02Answer a crisis with AI
They show the support details you chose and alert a named person.
- 03Quote what you have not approved
Every price, date or promise is checked against your approved facts.
- 04Pretend to be a person
Every conversation says it is an AI and how to reach someone.
Your next system, with IBI.
Tell us what slows you down, on a 30-minute call with Malik.
IBI Technologies Ltd · Founder Malik Makinde · Registered in England and Wales, No. 17055807 · 293 Archer Road, Stevenage SG1 5HF