Sovereign AI Based on Free Software

PrAIvate is a privacy-focused open-source RAG system for the controlled use of internal and confidential knowledge – in an EU cloud or entirely within your own infrastructure

Prototype under development Repository in preparation Live demo in preparation

Technological sovereignty as an architectural principle

PrAIvate combines controllable data processing, traceable retrieval and freely selectable operating models. Data protection and sovereignty are considered from the outset rather than added later.

Control and data protection

Data, models and infrastructure remain within a controllable operating model.

Traceable answers

Sources and metadata show which documents an answer is based on.

Open and interchangeable components

Free software and configurable models reduce unnecessary vendor dependencies.

From documents to robust answers

  1. Provide documents: Approved content is imported by upload or through an interface.
  2. Prepare content: Documents are extracted, segmented and enriched with metadata.
  3. Search for relevant content: Sparse and dense retrieval combine term-based and semantic search.
  4. Rank results: A reranker prioritises the passages most relevant to a query.
  5. Generate an answer: The language model receives only the selected context and generates a source-based answer.

Put knowledge to targeted use

PrAIvate is suitable for internal knowledge assistants, document-based research, technical documentation and limited pilot projects that assess data quality, retrieval and organisational integration.

Public administration

Making policies, procedural descriptions and internal knowledge accessible.

Regulated industries

Controlled use of confidential documents under heightened compliance requirements.

Small and medium-sized enterprises

Internal knowledge assistants and document-based research without reliance on proprietary AI platforms.

Controlled access to approved knowledge

More extensive tenant isolation is planned for the first public release. Data-protection-compliant operation also depends on the configuration, types of data, processes and responsibilities of the operating organisation.

Single sign-on

Central authentication is fully implemented in the current prototype.

Multi-factor authentication

Additional protection for user accounts is fully implemented.

Users and roles

User management and role-based access control are fully implemented.

Operation without mandatory external AI services

The architecture is designed for varying requirements concerning data protection, hardware, language support and organisational control.

EU cloud

The current prototype operates in a cloud environment within the European Union.

Fully local

PrAIvate can operate entirely within your own infrastructure.

Configurable models

Models and model-based components can be replaced with compatible alternatives.

Component and service architecture

Used components

Application

FastAPI, Uvicorn, Jinja2, HTMX, Vanilla JavaScript, custom CSS and Server-Sent Events.

Retrieval and ranking

Qdrant/bm25 for sparse retrieval and ms-marco-MultiBERT-L-12 for reranking.

Models and inference

Ollama as the inference runtime; granite4.1:3b as the current default configuration for the LLM and other model-based tasks; bge-m3 for dense embeddings.

Configurability

All models are configurable and can be replaced with compatible alternatives.

Further components for vector storage, parsing, persistence, deployment, identity providers, logging and monitoring will be fully documented before the public release.

Upcoming open-source release

PrAIvate is currently at the prototype stage. Tenant isolation, complete technology documentation and a consolidated installation process are among the next steps.

Public repository

Source code, the architecture overview and installation instructions are being prepared for the first public release.

Public live demo

A demonstration environment with dedicated sample data is in preparation.

Pilot projects

Individual pilot projects can systematically assess adoption, data quality, retrieval and the operating model.

Open-source software with strategic and technical support

PrAIvate will be released as free software. Consulting and individual pilot projects are also available to organisations with specific requirements for data protection, sovereignty and integration.

Potential services include use-case analysis, AI and RAG strategy, data protection and operating models, system and enterprise architecture, integration of free components, pilot implementation and knowledge transfer.

Contact and further information

Information about PrAIvate, its architecture, the planned open-source release and potential pilot projects can be requested using this form.

human test