Trust
Read-only, reconciled, and yours to leave with
The three questions every security review asks, answered directly, plus the two nobody asks until it is too late.
Last reviewed 3 August 2026
The direct answers
No qualifications on any of these.
Does it write to our systems
No. Every connection is read-only by design. SQOR never modifies your data and never writes to a source system.
Will our data train a language model
No. Nothing about your business trains a language model, and no language model retains your data. Prompts, payloads and answers are processed in the moment.
Does it replace our books
No. SQOR sits beside your audited books and never writes to them. It is not an accounting system and it does not replace an audit.
Can we see how a number was produced
Yes. Every figure traces to its data source, its calculation and its logic, with the date it was computed, reconciled to the ledger to the cent rather than to a tolerance band.
What happens if we stop using it
Your data stays where it lives, because we read it rather than take custody of it. If we provisioned a warehouse for you, its contents are yours.
The mechanisms
Four properties that are structural rather than promised.
01
The model never holds the number
Machine learning computes each figure and reconciles it to your source. The language model explains and cannot alter it, so a figure cannot be invented. That is architecture, not a setting.
02
The same question returns the same answer
A governed answer is computed, reconciled and stored, so asking twice returns the identical figure and two people asking in different words get the same number.
03
It withholds rather than guesses
When a figure cannot be reconciled, the platform declines to answer and says so, then builds the corrected metric. Nothing plausible-but-wrong is served to fill a gap.
04
Google Cloud underneath
SQOR is Google Cloud native and Marketplace listed, so enterprise security controls and existing Google Cloud commitments both apply.
On data retention
Two postures, and the difference matters.
The standard posture ships with every deployment: no language model retains your data and nothing about your business trains one. That covers the question most security reviews are actually asking.
For regulated buyers, medical, government, finance and other compliance-bound organizations, a walled-off arrangement is available in which the model runs only for that customer. It is an add-on rather than the default, the analytics subscription is still required, and the walled-off model is licensed separately. If your requirement is genuinely zero retention at the model layer with contractual backing, say so on the first call and we will tell you plainly what is and is not in place.
Questions
Trust and security, asked plainly.
Is it safe to connect an AI tool to our financial systems?
The two questions that decide it are whether it writes and whether it trains. SQOR is read-only on every connection and never modifies your data, and nothing about your business trains a language model. The third question, which fewer people ask and which matters more, is what the system does when it cannot reconcile a figure. SQOR withholds and says so rather than producing something plausible.
Does the analytics tool write to our systems?
No. Read-only, always, on every source. That is a design property rather than a configuration option, which means it cannot be switched on by mistake.
Will our data be used to train an AI model?
No. Nothing about your business trains a language model and no model retains your data. Prompts, payloads and answers are processed in the moment. For regulated buyers there is a walled-off option in which the model runs only for that customer, licensed separately.
Can auditors rely on numbers from an AI analytics platform?
They can rely on numbers they can trace, which is why traceability is the requirement rather than accuracy claims. Every SQOR figure traces to its source record, its calculation and its logic with the date it was computed, and reconciles to the ledger to the cent. It sits beside your audited books rather than replacing them, and it is not represented as an audit.
What happens to our data if we stop using the platform?
It stays where it lives, because SQOR reads your systems rather than taking custody of your data. If we provisioned a warehouse as part of the subscription, its contents are yours. There is no proprietary store holding your history hostage, which is worth confirming with any vendor before you sign rather than afterwards.
Is there a zero data retention option for analytics?
The standard posture already covers the model layer: no language model retains your data and nothing about your business trains one. For regulated organizations that need a contractual, walled-off arrangement with a model that runs only for them, that exists as a separately licensed add-on. Ask on the first call and we will be specific about what is in place today rather than what is planned.
Bring your security review to the first call.
We would rather answer the read, retention and reconciliation questions with your team in the room than send a document.