Honest AI
AI that cites its sources
Four capabilities, one principle: grounded. Every AI output starts from your tenant's data and ends at a human decision.
Questionnaire assist, cited, scored, reviewable
Q: How is access to production systems reviewed?
Access to production systems is reviewed quarterly by the platform owner, with results recorded and exceptions remediated within 30 days.
The audit trail records what was proposed, what was accepted, and by whom.
Illustrative, computed live per tenant in the platform.
The dividing line
Autonomous upkeep. Accountable decisions.
The category is racing to hand the programme to an agent that acts on its own. We built the opposite kind of autonomy. The tedious upkeep runs continuously, and no change reaches your programme without a named human on the record.
Autonomy that acts
An agent changes the programme on its own
Fast, until the examiner asks who approved it. What you are left holding is a log of what a machine decided, with no person accountable for the decision.
Autonomy that maintains
The machine keeps state current. A human owns every change.
Monitoring, drafting and crosswalking run continuously. Every proposed change carries a source and a confidence score, and waits for a named approver. The record shows who decided, on what evidence.
- 01
Propose
The AI drafts a mapping, an answer or a crosswalk from your tenant's own documents. Nothing is applied to the programme yet.
- 02
Cite
Every suggestion carries its source and a confidence score, so a reviewer sees exactly where it came from and how sure the model is.
- 03
Approve
A named human accepts, edits or rejects. Only then does it enter the programme, and the decision stays on the record.
The question in the room is never what the AI did. It is who approved this, on what evidence, and where it is written down. Compli-Once answers that by construction.
Four capabilities
One principle: grounded
Every capability starts from your tenant's data and ends at a human decision.
Auto-mapping
Upload a document and the AI proposes which controls it evidences across all adopted frameworks. Every link carries a confidence score and stays reviewable.
Crosswalk suggestions
Control equivalences proposed across frameworks, so one implementation satisfies many obligations. Equivalent or related, each mapping reviewable and auditable.
Questionnaire assist
Upload any questionnaire in PDF, DOCX, XLSX or CSV. Questions are extracted, answers drafted from your vault with a source and confidence per answer. What takes a team a week takes a reviewer an afternoon.
Grounded chat
A conversational assistant over your policies, controls and evidence, scoped to your tenant. It never answers from thin air, and never from someone else's data.
Questionnaire assist, cited, scored, reviewable
Q: How is access to production systems reviewed?
Access to production systems is reviewed quarterly by the platform owner, with results recorded and exceptions remediated within 30 days.
The audit trail records what was proposed, what was accepted, and by whom.
Illustrative, computed live per tenant in the platform.
The guarantee
Every AI output starts from your tenant's data and ends at a human decision. Nothing enters the programme silently. The audit trail records what was proposed, what was accepted, and by whom.
Connected to
Frequently asked questions
Is the AI trained on our data?
No. It retrieves from your tenant at answer time. It never answers from another tenant's data, and your documents are not used to train shared models.
Can we reject AI suggestions?
Yes. Rejection is a first-class recorded action. The audit trail shows what was proposed, what was accepted, what was rejected, and by whom.