On servers we operate
Our own AI runs on our servers, under the same rules as the rest of the system. A model built for a client belongs to that client alone.
WHAT WE BUILD · AI
AI assistance and integration, under strict protocols
We build AI into a client’s system on rules set before the first line. The AI is our own, on our servers, or a model reached through an API. The rules stay the same either way.
ONE REQUEST, UNDER THE PROTOCOLS
- 01A requestA question or a task for the AI
- 02Zero trustNo standing access, including ours
- 03 Our own AIOn our servers A modelThrough an API
- 04 VerifiedAn answer Not verifiedA clarifying question
- 05A person confirms the actionPeople confirm each action
The client decides where AI works, and can turn it off.
TWO WAYS TO RUN IT
Our own AI, or a model reached through an API
Either way, the protocols below stay the same.
Or a model reached through an API
Where an outside model fits the job, it is connected through an API, on the same protocols as our own.
THE PROTOCOLS
Six rules, set before the first line
The same rules for our clients and for ourselves. We practise them every day on our own systems.
No standing access
Nobody gets standing access, including us. Access ends when the work ends.
People confirm each action
A person stays in the loop. People confirm each action.
An answer, or a question
It answers only when the answer is verified. Otherwise it asks a clarifying question. It never guesses.
Built from several agents
AI projects are built multi-agent, to the same Gigagrams protocols.
Where AI works, and off
The client decides where AI works in their project, and can turn it off.
No training on it
We do not train AI on client data. A model is never trained on another client’s data.
WHERE IT STANDS TODAY
What is taken, in training, in build and planned
Each line with its status today, as we publish it everywhere else.
| AI projects for clients | Taken. Built to Gigagrams protocols: zero trust, human in the loop, multi-agent. | TAKEN |
|---|---|---|
| TEZZ AI | Our own AI. Runs on Gigagrams’ own systems. API access only. TEZZ AI | IN TRAINING |
| Real Estate CRM copilot | An AI copilot, TEZZ, inside our Real Estate CRM. Real Estate CRM | IN BUILD |
| AI Console | Ours, built for our own work. Access for clients may open later. | PLANNED |
OUR OWN AI CONSOLE
We built it after we studied how such systems work
We built our own AI Console for our own work, only after we studied in depth how such systems work. Access for clients may open later.
Planned HOW AN AI PROJECT RUNS
The same six steps as every project we build
The process of every Gigagrams project, from the brief to care after launch.
01
Your answers
People write the brief
Built from your own answers, one question at a time. Not written by AI.
02
Signed by both
The project passport
One card per question. Both sides sign off on the same text.
03
3 rounds per
milestone
Staging, with your comments
A private copy of the system. Comment on any part of it.
milestone
04
Score with date
and method
Stress tests
Speed, enterprise cybersecurity, function, QA and UX/UI.
and method
05
You own it
from day one
Launch
The work goes live after your review and the tests.
from day one
06
90 days
Care
Defects in our work fixed at our cost, then a care plan.
Our approach
Each step, and the reason for it.
STANDARDS
The controls AI work is built against
Each with how it is assessed. No certificate is claimed that a client has not earned itself.
| ISO/IEC 42001 | AI work is built against these controls, self-assessed. |
|---|---|
| SOC 2 · ISO/IEC 27001 | We build to these controls; each client certifies its own system. |
| NIST CSF | Tier 3 (Repeatable), self-assessed. Target: Tier 4 (Adaptive). |
We never claim that a system is safe from all attacks. No such system exists.How we secure systems
STRAIGHT ANSWERS
Questions about AI in your system
Either, under the same protocols. Our own AI runs on servers we operate. Where an outside model fits the job, it is reached through an API. The rules do not change with the model: zero trust, a person in the loop, and an answer only when it is verified. TEZZ AI, our own system, is in training and is reached by API only.
No, Gigagrams does not train AI on client data. A model built for a client belongs to that client alone. It is never trained on another client's data. The client also decides where AI works in their project and can turn it off. We run our own systems by the same rule.
It asks a clarifying question instead of answering. Our AI gives an answer only when the answer is verified, and it never guesses. People confirm each action, so a person stays in the loop. The same rule holds whether the model is our own or reached through an API.
AI work is built against ISO/IEC 42001 controls, self-assessed. Systems are built to SOC 2 and ISO/IEC 27001 controls, and each client certifies its own system; Gigagrams does not certify. Our security practice is self-assessed at NIST CSF Tier 3 (Repeatable), with Tier 4 (Adaptive) as the target. We never claim that a system is safe from all attacks.
Tell us where AI should work, and where it should not.
We ask one question at a time, and your answers become the project passport.