Use cases
Who buys. Where the agent acts. One Register → Gate → Prove loop.
Who buys
DPO
Close the CNIL file
Export agent register + AIPD + hash from the same inventory — without rebuilding a full RoPA suite.
- Outcome: Art. 30 / AIPD pack ready to attach
- Art. 12 provenance on every event
- Assist to frame wording — not to govern for you
CISO / RSSI
Stop the dangerous action
The action stays on hold until you decide — deny / HITL / suspend before apply.
- Outcome: nothing risky applies without you
- Approvals inbox as the home screen
- Deny / HITL webhooks for your SOC
Builder
Ship with policies that bite
A few SDK lines, per-skill policies, sandbox → prod — without flying blind.
- Outcome: agent instrumented in minutes
- SoR onboarding wizard
- Deny / require_approval demo on a real skill
Agents in production
Same Gate — risk changes by domain. Here’s where deny / HITL bite first.
Call center / support
Tickets, customer PII, CRM writes
The agent reads tickets, proposes a refund, or updates the CRM — nothing applies until the Gate decides.
- Classifier: email, phone, IBAN in the payload
- HITL on CRM write / credit / close
- Art. 12 trail: who approved what, hashed
HR
Candidates, scoring, HRIS access
Hiring and HR are high-risk under the AI Act — the Gate blocks scoring / decisions / exports until policy and HITL are in place.
- `rh` category → elevated risk signal
- Deny on CV / salary export out of scope
- AIPD prefilled from the agent inventory
Procurement
Vendors, POs, payments
The agent proposes a purchase order or a transfer — require_approval before execution, deny on amounts / IBANs outside the rules.
- Finance + IBAN classifier
- Amount thresholds → HITL or deny
- Proof for internal audit / DPO
Healthcare
Patient data, records, prescriptions
Health data is a sensitive category. The agent reads / writes / exports only with an explicit policy and human oversight.
- `sante` category → red feu / HITL
- Deny by default outside the allowed skill
- Compliance pack ready for inspection