AI Mode · private beta
Sovereign AI operations.
Without giving up control.
AI Mode turns operational intent into a signed, diff-able infrastructure plan — designed so nothing applies without explicit human approval. Constrained to Rectangle's verified blueprint catalog.
The verified plan loop
From prompt to signed deploy — with a human in the loop.
AI Mode is constrained by design. It is configured to resolve only to verified blueprints from Rectangle's curated catalog — it does not author free-form code. Every proposed change is diffed, signed and gated on human approval.
- Configured to use verified blueprints only — designed to avoid arbitrary execution
- Signed plan hashes, attached to approver identity
- Designed so apply requires explicit human approval
- Prompts, plans and approvals captured as evidence in LTA
- Available for selected sovereign deployments in private beta
- plan hash
- 0x9c41…ab2e
- residency
- fr-par · SecNumCloud
- blueprint
- peppol-ap@1.4 (verified)
- retention
- 10y · NF Z42-013
- Residency check passed
- Verified blueprint resolved
- Plan diff reviewed
- Approver identity attached
Sovereign AI stack
Designed for European models. Operated on sovereign infrastructure.
LeCloud is designed to support sovereign AI stacks — including Mistral AI and open-weight European models — running on Outscale SecNumCloud infrastructure, with the same verified plan loop wrapping every change.
Mistral AI & open-weight EU models
Verified blueprints designed to host European models inside your sovereign perimeter.
Configured for EU residency
Inference and training data designed to stay within FR/EU SecNumCloud zones.
Operator gates
Model deployments are gated on the same human approval workflow as the rest of the platform.
Peppol AP
verifiedPDP/PA-ready access point with SMP.
Keycloak Realm
verifiedSovereign identity with OIDC + SAML.
LTA Bucket
verifiedProbative archive aligned to NF Z42-013.
Mistral Inference
verifiedSovereign-hosted EU model endpoint.
Tyk Gateway
verifiedmTLS API gateway with quota guards.
ArgoCD App
verifiedGitOps app wired into the verified plan loop.
Bounded scope
AI Mode is configured to act only on verified blueprints.
Diff-able plans
Every change is rendered as a signed, reviewable diff.
Human gate
Designed so apply requires an explicit human approval with identity attached.
Evidence by default
Prompts, plans and approvals captured in LTA — exportable on request.
- Prompt receivedp:0x71…ceops-lead@rectangle · 14:02:11
- Plan signed0x9c41…ab2eai-mode / verified blueprint · 14:02:18
- Human approvalsig:0x4b…9dcompliance.peppol · 14:05:44
- Deployedrev:1.4.0argocd · fr-par-secnumcloud · 14:09:56
- Archivedlta:0xaa…12LTA · NF Z42-013 · 10y · 14:10:02
Private beta fit
Who AI Mode is built for — today.
AI Mode is in private beta for selected sovereign deployments. We're prioritising teams that are already operating regulated infrastructure and want a human-approved AI layer on top.
Internal platforms standardising on verified blueprints and signed plans.
Compliance, fraud, KYC and document AI pipelines that need approval gates by design.
Teams running PDP/PA, e-invoicing or probative archive operations under audit.
Mistral and open-weight European models running inside sovereign perimeters.