HomeSovra RAG
The sovereign Swiss RAG platform

Sovra RAG

SovraRAG — also written Sovra RAG — turns a company's document archive into an AI analyst that answers with citations. Hybrid GraphRAG retrieval, agentic reasoning and per-tenant encryption, engineered, hosted and operated entirely in Switzerland.

Sovra RAG is an enterprise document-intelligence platform: you upload your documents, ask questions in natural language, and get answers with citations back to the exact passages they came from. What distinguishes it from generic AI assistants is where it runs and what it can prove — the entire data path stays in Switzerland, every document is encrypted with a key unique to your organisation before storage, and every answer is traceable to its sources.

What Sovra RAG does

Most organisations already have the answers to their hardest questions — spread across thousands of contracts, reports, policies, dossiers and scans that nobody has time to read. Search finds documents; it does not answer questions. Sovra RAG closes that gap: it reads the corpus, builds a map of the entities and relationships inside it, and answers questions that span many documents at once — which of our supplier contracts contain this clause?, what changed between these policy generations?, what do we know about this counterparty? — with a citation for every claim.

How it works

  • Ingest. PDFs, Word, Excel, PowerPoint, plain text and scanned images go in; structure, tables and headings are preserved, and scans are read with OCR. Encryption is applied before anything is stored.
  • Understand. Beyond indexing the text, the platform extracts the entities and relationships in your documents into a knowledge graph, so the system knows that two documents are talking about the same company, product or obligation.
  • Retrieve. A question triggers semantic and keyword retrieval combined with the graph, across languages, restricted to the documents the asking user is allowed to see.
  • Answer. Complex questions are researched step by step by a bounded agentic process before a single, cited answer is written. Simple questions take the direct path.

The sovereignty guarantee

Sovra RAG is built so that sovereignty is a property of the architecture, not a clause in a contract. AI inference and embeddings run on a Swiss AI cloud; compute and object storage run in a Geneva datacenter; backups stay in the same Swiss bucket. No US-incorporated service appears anywhere in the data path, so there is no CLOUD Act exposure to argue about and no cross-border disclosure to justify under the revised Swiss Data Protection Act.

On top of that, every document is encrypted with AES-256-GCM using a key unique to your organisation before it reaches storage. The platform operator cannot read customer content — a property auditors can verify architecturally rather than take on trust. Background: what a Swiss RAG platform is.

What you can switch on

ModuleWhat it gives you
Hybrid searchSemantic and keyword retrieval across German, French, Italian and English, with cited passages
Knowledge graphEntities and relations extracted from your corpus, for questions that span documents
Agentic reasoningBounded multi-step research for comparative and aggregating questions
Reports & presentationsGenerated from your documents, with the sources attached
Access controlPer-document permissions enforced at retrieval time, with departments and sharing
Audit & analyticsQuery log, audit log and consumption tracking per user and per tenant

Not affiliated with Sovra (sovra.io)

Two unrelated products share this name. Sovra RAG on this site is the sovereign Swiss platform described above, operated from Geneva, Switzerland. It has no connection to Sovra (sovra.io), a digital-identity provider for Latin America whose platform includes an assistant module also called SovraRAG. Different companies, different countries, different product categories — there is no affiliation, partnership or shared technology between them. If you are looking for the Latin American digital-identity platform, you want sovra.io, not this site.

Sovra RAG (sovrarag.ch)SovraRAG module by Sovra (sovra.io)
CompanySovraRAG — Geneva, SwitzerlandSovra — digital identity for Latin America
What it isEnterprise document intelligence over your private corpusAssistant module inside a digital-identity and government-services platform
Data pathEntirely in Switzerland; no US-incorporated service anywhereNot stated publicly
EncryptionPer-tenant AES-256-GCM applied before storage; operator-blindNot stated publicly
RetrievalHybrid search + knowledge graph + bounded agentic reasoningSemantic search, re-ranking, context-aware synthesis
LanguagesGerman, French, Italian, EnglishEnglish, Spanish, Portuguese
Compliance postureSwiss nFADP/nDSG, FINMA-oriented, per-document access control, audit logNot stated publicly
Who it is built forSwiss regulated enterprises and public bodiesLatin American governments and institutions

Comparison based on the information published on sovra.io as of 28 July 2026. Where that site states nothing on a point, the table says so rather than assuming.

The practical difference for a Swiss buyer is the data path. Sovereignty here is a property of the architecture — inference, embeddings, storage and backups all inside Switzerland, content encrypted under a key unique to your organisation before it is stored — and it is the reason this platform can be used for client dossiers, case files and patient records that must not leave Swiss jurisdiction. See the twelve questions we recommend asking any vendor, including us.

Who uses it

Sovra RAG is used where the corpus itself is the sensitive asset: banks and wealth managers, insurers, pharma and life sciences, legal and fiduciary firms, hospitals, manufacturers, watchmakers, commodity traders, public administrations and universities. The common thread is a regulatory or contractual reason that the archive cannot be sent to a foreign cloud — and a real need to ask questions of it. The industries overview sets out typical workflows per sector.

Deployment and pricing

Sovra RAG is delivered as a managed Swiss service. Team and Business plans run on shared Swiss infrastructure with per-tenant encryption and isolation; the Sovereign Enterprise tier runs on dedicated single-tenant Swiss infrastructure with an SLA and a compliance pack, for institutions whose supervisors or clients require infrastructure separation. Pricing is in CHF and published on the pricing section of the homepage — from CHF 49 per month for a single user to dedicated deployments from CHF 3’900 per month.

Getting started

The usual sequence is a guided demo, an architecture review with your security or compliance function, then a 30-day pilot on your own documents — running on the production Swiss data path from the first day, not in a sandbox somewhere else. Before that conversation, the twelve-question evaluation checklist is worth sending to every vendor you are considering, including us. Talk to the team via the contact page.

Sovra RAG — questions buyers ask

Is this related to the SovraRAG module from Sovra (sovra.io)?

No. They are unrelated products from unrelated companies that happen to share a name. Sovra RAG here is Swiss enterprise document intelligence, operated from Geneva for organisations whose documents must stay under Swiss jurisdiction. The sovra.io module belongs to Sovra, a Latin American digital-identity provider, and is part of its identity and government-services platform. There is no affiliation, partnership or shared technology, and claims made about one do not apply to the other.

Where do inference, embeddings and storage actually run?

All four locations are in Switzerland: language-model inference and embeddings are served from a Swiss AI cloud, compute and object storage run in a Geneva datacenter, and backups stay in the same Swiss bucket. No US-incorporated service is part of the data path, which is why there is no cross-border transfer to assess.

Can SovraRAG read our documents?

No. Each organisation's documents are encrypted with AES-256-GCM under a key unique to that organisation before they reach storage, so the operator holds ciphertext it cannot decrypt. This is an architectural property rather than a policy commitment — it is what “operator-blind” means.

How is our data isolated from other customers?

Each tenant has its own encryption key and its own logical data boundary across the document archive, the vector index and the knowledge graph. Retrieval is scoped to the tenant before any access-control check is applied. Organisations that require physical separation as well use the dedicated single-tenant Sovereign Enterprise deployment.

Who inside our organisation can see which documents?

Only the users you permit. Access control is per document and is enforced in the retrieval layer, not just in the interface, with department grouping and user-level sharing. An answer can only ever be built from documents the asking user is allowed to open, and citations make that visible.

How do we know an answer is correct?

Every answer carries numbered citations to the passages it was built from, so a reviewer can open the source and check the claim. The query log records what was asked, which documents were used and when, which is what makes the system auditable in a regulated review.

What happens to our data if we stop using the platform?

Your documents are exported in a usable format and then deleted, with deletion cascading to chunks, vector entries and knowledge-graph entries; backups expire on their rolling schedule. Export format, deadlines and a deletion confirmation are part of the contract rather than a support request.

Which languages and file formats are supported?

German, French, Italian and English, including asking in one language and retrieving from documents written in another. Formats: PDF, DOCX, XLSX, PPTX, plain text and scanned images via OCR, with document structure, tables and headings preserved. See multilingual RAG for Swiss documents.

Is a dedicated single-tenant deployment available?

Yes — the Sovereign Enterprise tier runs on dedicated Swiss infrastructure with a 99.9% SLA and a compliance pack, for institutions whose supervisors or client contracts require infrastructure separation. It remains a managed service; you do not operate it yourself.

How long does onboarding take?

A guided demo takes an hour; a proof-of-concept on your own corpus typically runs 30 days, of which the first two weeks are scoping and ingestion and the rest is evaluation with real users. The gating factor is almost always document quality and the permission model, not the technology.

How does this differ from a general AI assistant?

A general assistant answers from its training data on foreign infrastructure with no access control. Sovra RAG answers from your archive, cites the passages, enforces per-document permissions at retrieval time, logs every query for audit, and never lets the content leave Switzerland.

See it on your own documents

Get a guided demo and a proof-of-concept with your corpus — on Swiss infrastructure from day one.