How it works

AI understands. Policy authorises.

Mantle lets the task proceed while keeping deterministic organisational policy in control of data, model access, approvals and evidence.

01
Understand the task
Determine the user’s intended outcome and the information genuinely needed to complete it.
02
Inspect data and context
Evaluate prompts and documents for relevance, sensitivity and declared restrictions.
03
Apply organisational policy
Use deterministic organisation rules as the final authority over data and model access.
04
Transform and minimise
Keep, mask, remove or escalate information before it crosses the trusted boundary.
05
Show the exact payload
Let the user approve the precise outbound content—not a summary of it.
06
Route to an approved model
Send only approved context to a model eligible for the task and classification.
07
Restore approved context
Reintroduce approved masked values only where policy and the workflow permit.
08
Retain evidence
Record decisions, approvals, routing and permitted actions without keeping raw removed values.

Three governed planes

Control what AI can see, receive and do.

Data

Classification · relevance · minimisation · permissions · retention

Model eligibility · routing · private or local options · policy

Action

Propose · approve · execute · expire · audit

Pilot and design partner conversations

Start with one workflow, one team and one real policy boundary. We’ll explore where AI can create useful productivity without sacrificing control.