Technology & Architecture

Architecture Overview

The FINO Suite follows a modular cloud architecture with clear layer separation. Each component is independently scalable and replaceable.

Frontend - Web Components (framework-independent, integrable into any website)

↕

API Layer - REST API with authentication, rate limiting and tenant separation

↕

AI Orchestration - Prompt management, context preparation and response validation

↕

RAG Pipeline - Retrieval Augmented Generation with vector and hybrid search

↕

Knowledge Layer - MCP servers, knowledge bases, document indices

Cloud infrastructure: Operated by default on EU data centres. Optionally available on STACKIT (Schwarz Group) as a sovereign cloud alternative - on customer request.

Language Models (LLMs)

FINO is model-agnostic and supports various Large Language Models. The choice of model can be configured per tenant and use case.

Provider	Models	Use Case	EU Hosting
Anthropic	Claude Sonnet family	Dialogue management, form assistance, complex follow-up questions	✅ via EU infrastructure
Amazon	Nova Pro, Nova Lite, Titan	Document analysis, image processing, embeddings	✅ EU (Frankfurt)
Amazon	Nova Sonic	Speech processing (FINO Voice)	✅ EU (Stockholm)
Others	Configurable on request	Customer-specific requirements	Depends on provider

Important: No model is trained with customer data. All requests are processed statelessly - no conversation data is permanently stored with the model providers.

RAG - Retrieval Augmented Generation

FINO uses RAG to base AI responses on verified facts rather than general model knowledge. The result: professionally accurate, up-to-date and traceable answers.

Retrieval (Knowledge Retrieval)

For each user query, relevant information is retrieved from the connected knowledge bases.

Vector search: Semantic matching via embeddings
Hybrid search: Combination of semantic and keyword search
Ranking: Relevance scoring and filtering of results
Source references: Every piece of information is traceable to its source

Generation (Response Generation)

The language model generates a response based on the retrieved information and conversation context.

Context window: Relevant documents are provided to the model
Prompt engineering: Domain-specific instructions control tone and accuracy
Validation: Responses are checked for consistency
Multilingual: Response in the user's language, form in German

Why RAG instead of pure LLM?

Without RAG (pure LLM):

Responses based on training data (outdated)
Hallucinations possible
No source references
Not tenant-specific

With RAG (FINO):

Responses based on current, verified sources
Fact-based and traceable
Source references with every response
Individual knowledge base per tenant

MCP - Model Context Protocol

FINO uses the Model Context Protocol (MCP) as an open standard for communication between AI systems and knowledge sources. This enables a flexible, extensible architecture.

Modularity

Knowledge sources as independent MCP servers
Easy addition and removal of data sources
Independent scaling per source
Standardised interfaces

Distributed Architecture

Multiple knowledge sources queryable in parallel
Multi-tenant configuration
Real-time knowledge base updates
Interoperability with various AI models

Direct Use in AI Clients

Thanks to MCP, FINO knowledge bases can be integrated directly into AI assistants like Claude Desktop, JetBrains AI or other MCP-compatible clients. Knowledge sources are available to the AI model as tools - no custom development, no middleware required.

This means: Your specialist content is instantly available in any MCP-compatible client - whether for internal research, customer consulting or automated workflows.

Integration & Interfaces

Frontend Integration

FINO is provided as a Web Component - a single HTML tag is all you need for integration.

                            <chat default-language="en"></chat>
                        

No framework required
Works in any website
Responsive and accessible
Customisable design

Integration guide →

Backend Interfaces

Standardised APIs are available for deeper integrations.

REST API: Standard HTTP interface for all products
MCP Protocol: For knowledge base connectivity
Webhooks: Event-based notifications
Form mapping: Automatic mapping of AI responses to form fields

CMS Plugins

WordPress (available)

Ready-to-use plugin with graphical configuration interface. Installation via WordPress admin, branding customisation without code changes.

All branding options (colours, texts, logos)
Multilingual configuration
Page-level visibility control

Other CMS (on request)

Integrations for Drupal, Joomla and Shopware are planned. Contact us for your specific use case.