Architecture overview
How the neuland.ai HUB is built. From data source to answer.
From bottom to top: ingestion takes in your data, the ontology organises it, the Orchestrator turns it into answers, and the harness sets the framework for every step.
Hover over or tap a component: it explains its role and highlights its connections.
External engines
Cohere North · n8n · any engine
Connected via SDK · MCP
Teams already working with Cohere North or n8n use the knowledge, permissions and orchestration of the neuland.ai HUB through them.
Platform to platform
Customer and partner platforms, e.g. from testing and certification organizations, banking groups or insurers
These platforms use the neuland.ai HUB as their orchestration core, white-label or embedded
Connected via SDK · APIs · MCP · federation
Customer and partner platforms use the neuland.ai HUB as their orchestration core, white-label or embedded.
Enterprise data systems
Wherever the system lives: secure, open MCP
Public cloud / API
Microsoft 365 / SharePoint · Salesforce · etc.
On-prem / private cloud
SAP S/4HANA · Navision · Abas · Orbit · Documentum (DMS) · File servers · custom applications · etc.
Your data systems, wherever they live: ingestion takes in their data, the neuland.ai HUB accesses them directly via MCP, and each system has its own connector ontology.
neuland.ai HUB
Sovereign AI management and orchestration platform
Native interfaces: built on the HUB
Chat
Direct dialogue with the neuland.ai HUB, on the same core as every other interface.
Agents
Agents with fixed specifications: created once, shared with the team, with the same permissions and the same knowledge.
Projects
Workspaces with their own knowledge, shared with the team, on the same core.
Custom apps
Interfaces built on the neuland.ai HUB for specific use cases.
SDK and APIs
Programmatic access to the core · REST · MCP · platform federation
Programmatic access to the core via REST, MCP and platform federation. This is where external engines and your own platforms connect.
Core
Orchestrator
Routes every task to the right ontology and the right model per department and application
deterministic or agentic
It breaks every task into steps and selects knowledge, tool and model for each step. Fixed workflows run deterministically, open-ended tasks agentically, and every step is logged.
Harness
Context · model · check · execution: one loop per step
Information · tools · budget · human sign-off
The harness is the framework every agent works within: which information it may use, which tools it has, how much it may spend and where a person has to decide. If the Orchestrator is the conductor, the harness is the score.
Agentic retrieval and reasoning
Agentic RAG · recursive language models (RLM)
Agentic threads · deep agents · multi-hop reasoning
KAG (knowledge-augmented generation) · query planning · semantic re-ranking
Plans the search, links knowledge across multiple steps and ranks results by meaning before a model answers.
Ontology layer
Multiple ontologies, orchestrated per department and application · generated automatically from your data (Ontology Builder) and kept in sync
Enterprise ontology
Business logic · entities · relations · domain knowledge
Connector ontologies
One per system · gives every access the context of its system
Materialized as
Knowledge graph · skills · rules · vector index. The basis for transparent, verifiable answers
The ontology layer describes your company and every connected system. It is generated automatically from your data and stays continuously up to date.
Governance and compliance
Roles and permissions · guardrails · audit trail
incl. EU AI Act · DORA · BaFin · GDPR · BRAO
Corporate values · stance · ethics
Explainability and traceability
Roles, permissions and guardrails apply to every interface and every agent. Every step remains traceable in the audit trail.
Observability and logging
Across all connectors, platforms and agentic processes
Tracing · metrics · cost and token usage
Prompt and response logs · evaluations
Tracing, metrics, costs and logs across all connectors, platforms and agentic processes.
Ingestion pipelines
Petabyte range · scalable · Kubernetes GPU cluster
The ingestion pipelines take in data from your systems, scalable to the petabyte range, on a Kubernetes GPU cluster.
Models: agnostic
Sovereignty means: where the model runs
External · cloud API
incl. GPT · Claude · Gemini
not sovereign · via cloud API
On-prem / sovereign · open source
incl. Qwen · gpt-oss · DeepSeek · Mistral · Llama · Cohere · Nemotron
self-hosted SLMs · fine-tuned
The Orchestrator selects the right model per department and application: external via a cloud API, or operated sovereignly, self-hosted and fine-tuned.
Sovereign deployment
The neuland.ai HUB runs in Schwarz Digits Cloud (STACKIT), on Microsoft Azure, Google Cloud, in an NVIDIA AI Factory, on-premises or hybrid. Your data stays under your control.
Options
The interchangeable foundation: the environment follows your protection needs, while the neuland.ai HUB on top stays the same.
Structured knowledge
The ontology maps your organisation.
A shared vocabulary that writes itself.
In the CRM the customer is an account, in accounting a debtor, in sales a partner. The ontology is your organisation’s shared vocabulary: it defines what exists and how it connects, for example that customers have contracts, contracts cover products and certain decisions need sign-off from a specific role. The Ontology Builder derives it automatically from your systems: from data, structures, metadata, permissions and processes. It becomes more precise with every system that docks on, and the harness works with it.
One term across all systems
The field in the ERP, the record in the CRM and the column in the spreadsheet all map to the same business term, even when customer, contract and sales team sit in three systems.
Permissions built in
The permission model is part of the ontology, not a filter in front of it. AI never reveals anything that would otherwise stay hidden.
Knowledge graph, not keyword search
Which contracts held by customers with an open complaint expire by year end? No single document holds the answer. The neuland.ai HUB follows the relationships and backs every answer with its source.
Grows as you work
New sources, objects and relationships are added while the system runs, and every completed case feeds back in. A hand-maintained ontology always lags behind. The Ontology Builder keeps it current.
Model-agnostic
When a better model comes along, you swap it in without rebuilding: agents, knowledge, permissions and workflows stay as they are. Admins decide which models a group may use.
Including these models, closed and open
OpenAI
Claude
Gemini
Llama
Mistral
DeepSeek
Qwen
Kimi
GLM
Cohere
Aleph Alpha
Nemotron
Where your models run: in the EU cloud, in Schwarz Digits Cloud, in your own data center or hybrid. And how you define per group which model does the work.
See deployment and model choice
How strong, fast and expensive common models are, measured independently by Artificial Analysis, and which ones can run in your own data center.
Go to the model benchmark
Tools and data
Public cloud or your own data center: the ingestion pipelines take in the data from your systems, and the neuland.ai HUB accesses them directly via MCP. Complex system landscapes and legacy systems can be connected too.
Petabyte-scale ingestion
Even large data sets such as complete DMS archives and file servers are ingested, with ten times the cost efficiency in querying.
Open MCP
Through the open MCP standard, the neuland.ai HUB also reaches systems that have no ready-made connector, such as custom applications.
Platform to platform
If you run your own AI platform, for example as a testing and certification organization, banking group or insurer, the neuland.ai HUB provides orchestration, knowledge and model selection in the background. Your customers keep working in your interface, under your brand.
Your brand stays up front
Your customers only see your interface; the neuland.ai HUB orchestrates in the background.
APIs, MCP and federation
Your customer portal queries the core via API or MCP, for example for framework agreement RV-2027-014, and gets the term and conditions back with their source.
Permissions and audit trail included
Roles, permissions and the audit trail of the neuland.ai HUB also apply through your platform.
Your systems, your models.
We bring the core.
Together we define which systems connect, which models you approve and where the neuland.ai HUB runs.
Talk to us
After your request
01
You name your systems and your deployment requirements
02
We show where models, data and permissions come together in the neuland.ai HUB
03
You receive a proposed architecture for your system landscape