Skip to content
LumeSecLumeSec

Platform · Architecture

Eight layers.
One contract.

For IT, architecture and audit: how the platform is built, where your data lives and at which point each rule applies.

01Overview

From your systems to your processes.

This is how we explain the platform to business teams: five levels, the platform in the middle.

01Your processes

  • Statement analysis
  • Loan application
  • Annual accounts
  • Quote
  • Claim
  • Building permit

02Specialised agents

03LumeSec AI Platform

  • Identity
  • Permissions
  • Models
  • Rules
  • Approvals
  • Audit trail
  • Operations

04Models and data

  • Language models
  • Classic machine learning
  • Document recognition
  • Your data

05Your systems

  • Core banking
  • RZL · BMD
  • SAP
  • Microsoft 365
  • SharePoint
  • Line-of-business apps
Fig. · The platform handles identity, permissions, models and audit trail once. Every agent inherits it.

02Reference architecture

The eight layers in detail.

Each layer has one clear job and a documented interface to the next.

  1. L01ClientsPortal, command line, desktop and mobile apps. All speak the same API.
  2. L02Enterprise IdentitySingle sign-on via SAML, OIDC and SCIM. MFA with passkeys. Tenants as a hard boundary.
  3. L03Enterprise DataRelational data, search and vector search, with classification and retention rules.
  4. L04Platform CoreAssistant, tasks, integrations, knowledge, agents, roles, budget and approvals.
  5. L05Agentic AutomationDeterministic processes for agents. Your own agents can connect through the Open Orchestrator.
  6. L06APIOpenAPI 3.1. Every response carries a trace hash that locates it in the audit trail.
  7. L07Micro ServicesDocument processing, recognition, calculation, export. Scaled one by one.
  8. L08SDK & Development.NET SDK for your own services, integrations and modules, with checks before every release.

03Principles

Three things every layer does.

  • 01Correlate

    Correlate

    Every event, document and decision is linked to a person, an object and a process.

  • 02Contain

    Contain

    Agents act only within their permissions, rules and budgets. Anything beyond needs an approval.

  • 03Attest

    Attest

    Every step is signed and logged so it cannot be altered. The audit trail produces evidence for audit and supervisors.

04 · Traceability

The same inputs. The same result.

Calculations run on fixed, versioned formulas. Extractions, rules and model versions are stored with each step. So a result can be traced and reproduced months later.

  • Formulas versioned
  • Rules versioned, with source
  • Model version per step
  • Signed audit trail that cannot be altered

Marcus M. Ziegler

Founder & CEOBirgitz, Tyrol--:--

A demo directly with Marcus.

30 minutes on your example. Marcus shows you what the LumeSec AI Platform can do. Afterwards you know what an agent takes over for you.

30 min · Microsoft Teams · German or English

What should we look at?

We show: the analysis of one set of annual accounts up to a draft vote

Pick a time
Calendar opens herePrefer e-mail