Integration Landscape

End-to-end view of the integration architecture in meinGPT

This page explains the architecture model. If you simply want to connect a service, start here: Integrations Home.

The Full Picture

In meinGPT, every integration type is reduced to one shared runtime model:

LLM + tool calls as the central integration fabric.

This allows very different systems to be operated, secured, and scaled in a consistent way.

How requests flow
Assistants & workflowsmeinGPT runtimeConnector / data sourceExternal systems

Your options

Five ways to connect external systems to meinGPT — from no-code connectors to your own MCP server.

Connectors

Pre-built integrations for standard tools, ready to use in the UI.

Use cases

Send messagesCreate ticketsFetch CRM data

Databases

Direct queries on structured data (e.g. SQL), including live access.

Use cases

Generate reportsAnalyze dataUpdate records

Data Pools

Makes documents and files searchable so assistants can answer with context.

Use cases

Search documentsRetrieve knowledgeAnswer FAQs

No-Code

Connects existing automations like Make/Zapier with meinGPT.

Use cases

Trigger workflowsSync dataAutomate processes

Custom MCP

Your own interface for internal APIs, custom logic, or legacy systems.

Use cases

Connect legacy systemsCustom APIsSpecial protocols

Side-by-side

Which path covers which capability? For orientation — exact scope varies by connector and target system.

Read

  • Connectors
    Full
  • Databases
    Full
  • Data Pools
    Full
  • No-Code
    Full
  • Custom MCP
    Full

Actions / write

  • Connectors
    Full
  • Databases
    Full
  • Data Pools
    Not suited
  • No-Code
    Full
  • Custom MCP
    Full

Live access

  • Connectors
    Partial
  • Databases
    Full
  • Data Pools
    Not suited
  • No-Code
    Partial
  • Custom MCP
    Full

User permissions

  • Connectors
    Full
  • Databases
    Partial
  • Data Pools
    Full
  • No-Code
    Partial
  • Custom MCP
    Full

Private networks

  • Connectors
    Not suited
  • Databases
    Full
  • Data Pools
    Full
  • No-Code
    Partial
  • Custom MCP
    Full

Easy setup

  • Connectors
    Full
  • Databases
    Partial
  • Data Pools
    Full
  • No-Code
    Full
  • Custom MCP
    Not suited
Full support · Partial · Usually not the right fit

Permissions, identity, and network rules apply consistently to every integration type:

  • Permissions (RBAC, team sharing)
  • Identity (JWT, audit trail)
  • Connections (cloud & on-premise)

In 30 Seconds: What This Means for You

  1. You connect systems through one integration model, not separately per assistant.
  2. Your assistant selects the right tool or data source per task.
  3. Security, permissions, and network access apply consistently across all integration types.

Security and Access Model (Cross-cutting)

These layers apply across all integration types:

  • Permissions and assistant sharing (users, teams, groups)
  • Identity forwarding (JWT) for secure context propagation
  • Cloud/on-prem connections (IP allowlisting, Enterprise Connection Network, VPN)

Typical User Paths

I want to...Start here
Connect Microsoft 365, Google, Jira, or ConfluenceConnectors Overview
Make documents/files searchable as knowledgeData Pools (RAG)
Query databasesDatabases Overview
Connect internal APIs/servicesCustom MCP
Reach private systems in company networksConnections (Cloud/On-Prem)

When to Use Which Building Block

Building BlockUse Case
ConnectorsExecute actions in external tools (Slack, Jira, etc.)
DatabasesQuery structured data precisely (SQL, NoSQL)
Data Pools (RAG)Retrieve and ground on document knowledge
No-CodeIntegrate existing process automations (Make, Zapier)
Custom MCPConnect customer-specific protocols/backends
Custom AI AppsProvide guided custom UIs for specific workflows

Cloud Default vs. On-Prem Advanced

  • Default: For most teams, setup in the meinGPT UI is sufficient.
  • Advanced: On-prem/Data Vault is for teams needing own runtime and network control.

Related

Was this page helpful?