A-Team

An open framework as the starting point

The A-Team connects existing tools into an open, controlled, and replaceable framework. At its centre are reusable building blocks for sign-on, collaboration, knowledge, and AI. Existing systems are not replaced pre-emptively; they are connected, retired, or extended deliberately.

Every organisation needs similar foundations: access, files, knowledge, automation, operations, and clear boundaries for AI. The difference lies in processes, data, roles, and risks. The method therefore starts not with a catalogue of tools but with a survey of the existing system landscape.

Anti-patterns: what the method avoids

Migration in advance: Systems are replaced before the value is clear.
Tool fixation: A single product dictates the architecture even though the function should remain replaceable.
AI as a black box: Prompts, data outflow, model choice, and access remain invisible.
Knowledge silos: Documentation lives in slides, tickets, or drives, but not as a reliable source for people and agents.
All-or-nothing programmes: A large rebuild blocks small, verifiable steps.

Process

Make the current state visible: Sign-on, collaboration, files, knowledge sources, data flows, integrations, and permissions are described as the current architecture.
Set sovereignty boundaries: For data, models, and external services, it is defined what stays under direct control and what may deliberately leave that boundary.
Select building blocks: Open components such as identity providers, knowledge bases, RAG layers, automation, and monitoring are selected by function, not by product lock-in.
Connect what already exists: Microsoft 365, specialist applications, websites, ERP, CRM, or databases stay where they make sense and are connected through clear interfaces.
Integrate controlled AI: A model gateway, access control, source binding, and logging limit which data goes to which model.
Deliver in small steps: Every step must stay operable, verifiable, and replaceable. Documentation, operations, and decommissioning are part of the delivery.

Operating model

The operating model separates function, tool, and infrastructure. Sign-on is a function; Authentik, Entra ID, or another identity provider are possible implementations. Knowledge management is a function; Wiki.js, Nextcloud, GitLab, or existing systems can take that role. AI is not a single application but a layer of access, context, model choice, and evaluation.

architecture-beta
    group org(server)["Organisation"]
    group data(database)["Knowledge and processes"]
    group ai(server)["AI layer"]
    group ops(server)["Operations"]
    service roles(server)["Roles"] in org
    service access(server)["Access"] in org
    service knowledge(database)["Knowledge and sources"] in data
    service processes(server)["Processes and applications"] in data
    service agents(server)["Agents RAG gateway"] in ai
    service models(server)["Internal or external models"] in ai
    service monitoring(server)["Monitoring backup updates"] in ops
    service infra(server)["Controlled infrastructure"] in ops
    roles:R -- L:access
    access:R -- L:knowledge
    access:B -- T:processes
    knowledge:R -- L:agents
    processes:R -- L:agents
    agents:R -- L:models
    agents:B -- T:monitoring
    monitoring:R -- L:infra

The architecture stays readable because every component has a clear responsibility. When a tool is replaced, the function remains. When an external AI service is used, access runs through a deliberate approval instead of arbitrary direct calls.

Practical example

A non-profit association already works with Microsoft 365. Sign-on and collaboration stay there. A structured knowledge base, a RAG layer, and a controlled model gateway are added. This creates a foundation for internal answers, campaign work, and agentic software development without rebuilding the existing collaboration.

The value does not come from replacing every tool. The decisive point is the controlled connection: knowledge becomes citable, AI access is bounded, and new applications are built on a foundation that stays openly documented and replaceable step by step.

Ask AI

These links open external AI services, the conversation and its content are sent to their providers.