Patrivox
European SaaS for heritage institutions: archives, museums, libraries and learned societies upload PDFs in bulk, then get Mistral OCR, entity extraction, full-text and semantic search, and a source-citing AI chat, with unlimited readers and EU hosting.
What is Patrivox?
Patrivox is a European SaaS platform for digitising, indexing and searching archival collections. It is published from Tournai, Belgium, by an independent practitioner who sits on the board of a historical and archaeological society founded in 1846: the product answers that kind of institution's need rather than being adapted to it afterwards. The pipeline is deliberately short. An institution drops PDFs into the platform, each page is read by Mistral OCR 3, named entities - people, places, organisations, dates - are extracted automatically, the text is indexed twice (Typesense for full text, Qdrant for semantics), and an AI chat answers questions in natural language while citing the exact document and page. The publisher presents three pillars: Automation (upload and it is ready), Search (find in one second) and Constellation, a knowledge graph that links people, places and organisations across the collection and lets a researcher move from one entity to the next. A built-in viewer adds zoom, page navigation and copyable OCR text. Access is organised around three roles: administrator, contributor and reader. Administrator seats are capped by plan, but readers are unlimited on every plan, the free one included, and they consult and search without creating an account, which matches the public-access mission of archives, libraries and learned societies. Five plans are published - Decouverte (free), Starter, Essentiel, Association and Pro - alongside a quote-based custom offer; an Institution plan exists in the interface strings but is not exposed publicly. Higher plans widen page, storage and AI query quotas, extend audit log retention from 7 days to 5 years, and add long-term S3 archiving, CSV export, per-project API access and a public AI chat announced for March 2026 on Pro. The figures displayed are modest and consistent with the product's age: 10,000+ pages digitised, 206 entities identified by the AI, 100% hosted in Europe, and claimed adoption among heritage institutions in Belgium and France. No blog post and no changelog entry had been published as of 13 August 2026.
What it does
- Upload PDFs one at a time or by the hundreds through batch import
- Run Mistral OCR on every page, including old and handwritten documents
- Automatically identify people, places, organisations and dates in the text
- Search the whole collection in full text with typo tolerance, or by meaning
- Question the collection in natural language and get answers sourced page by page
- Navigate the Constellation knowledge graph from one entity to the next
- Share documents through per-document links with optional expiry, view cap and password, open consultation to unlimited readers without an account, and export documents, extracted text and metadata
When to use Patrivox / When not to
A quick filter to help you decide if Patrivox is the right fit.
When to use Patrivox
- Municipal and communal archives holding deliberations, registers and correspondence that are already digitised but not searchable
- Historical and learned societies with bulletins and documentary collections that currently depend on volunteers for manual indexing
- Heritage libraries and documentation centres wanting to open special collections to researchers, students and the general public
- Dioceses and church fabrics keeping parish registers and ecclesiastical archives
- Institutions bound by EU data sovereignty requirements that also want unlimited readers to consult their holdings without creating an account
When not to use Patrivox
- Teams that need a mobile workflow: no iOS or Android application is published
- Organisations requiring a standard self-hosted or on-premise deployment, which exists only within the quote-based custom offer
- Developers who want to evaluate the API before committing: API access is reserved for the Pro plan and no public API documentation is available
- Institutions with irregular monthly volumes: page quotas do not roll over from one month to the next, and deleting a document does not give back the OCR pages already consumed
- Buyers whose procurement demands audited certification or a corporate vendor, and teams needing many administrator seats: plans include 1 to 5 administrators (extra seats at 15 EUR/month) and the publisher is a Belgian sole trader claiming no SOC 2 or ISO certification
How to use Patrivox
A typical end-to-end flow, from setup to results.
- Open a free account online; no credit card is required
- Drag and drop your PDFs, one at a time or by the hundreds
- Use single upload for files up to 250 MB, or bulk import through an S3 client for files up to 1 GB (Association plan and above)
- Let the AI analyse: OCR runs page by page and entities are extracted, with results announced in about two minutes, no manual keying and no configuration
- Explore the collection with full-text search (typo tolerant, filters by date, author, type and tag) or semantic search
- Ask the AI chat your questions and check the document and page cited with each answer
- Adjust the tone of the answers (accessible, formal, academic) and the language they are written in
- Browse the Constellation graph to move between entities and see which documents they share
- Share your work: per-document links with expiry from 1 hour to 30 days, optional view cap, password protection and direct PDF download, or open consultation to unlimited readers
- For larger needs, book a 30-minute or 60-minute demonstration slot, or request a quote covering assisted migration from CSV, EAD/XML, Access or FileMaker databases and loose scans, metadata mapping included
Pros & Cons
Pros
- A complete chain from the raw PDF to a sourced answer, with no configuration and no manual data entry, designed by a heritage practitioner rather than by a generalist software vendor
- Explicit European sovereignty: object storage with Scaleway in France, OCR by Mistral AI, four named subprocessors all located in the EU, and no transfer outside the EEA
- Written commitment that customer content is not used for model training, both by the publisher and through contractual guarantees obtained from model providers
- Unlimited readers on every plan, the free one included, which fits the public-access mission of heritage institutions
- A permanent free plan with 100 lifetime pages and no credit card, plus add-on packs that avoid an early plan change (the publisher publishes a worked cost example)
- 15% discount for associations and non-profits on presentation of their statutes, and 15% off on annual commitment
- Verifiable deletion - files, extracted text, search index and AI vectors are purged - alongside a DPA on request, published subprocessors and an audit log
Cons
- A very young product: domain registered on 31 January 2026, first Wayback capture on 16 March 2026, and both the blog and the changelog still empty on 13 August 2026
- The publisher is a sole trader, not a capital company: no team page, no about page, and no social presence on LinkedIn, Instagram, YouTube or X
- Help documentation is published but empty, and no public API documentation exists to assess the API before subscribing to Pro
- No security certification is claimed, neither SOC 2 nor ISO 27001; the guarantees are contractual rather than audited
- Pricing inconsistencies between language versions: the English add-on example quotes outdated prices (Essentiel 99 EUR, Association 199 EUR, +2,500-page pack) where the plan cards show 129 EUR, 299 EUR and a +1,000-page pack, and the in-app strings announce 20% off annual where the public page announces 15%
- A gap between the declared EU hosting and the observed infrastructure: the front end resolves to Hetzner in Falkenstein, Germany, while the FAQ names only Scaleway in France
- Modest traction figures displayed (10,000+ pages, 206 entities), no mobile application, and page quotas that never roll over - deleting a document frees storage but not consumed OCR pages
Pricing & Plans
A permanent free plan is offered: Decouverte provides a one-off lifetime credit of 100 pages, 200 MB of storage, 50 AI queries per month, one administrator and unlimited readers, with no credit card required. The lowest paid entry point is the Starter plan at 29 EUR per month excluding VAT; all prices are quoted excluding VAT, which applies according to the customer's country. An annual commitment reduces the price by 15% against the monthly rate and is invoiced once a year, while quotas continue to reset every month. Associations and non-profits obtain a 15% discount on all plans on presentation of their statutes. Monthly add-on packs may be purchased instead of changing plan: +1,000 pages for 15 EUR, +500 AI queries for 12 EUR, +25 GB of storage for 8 EUR, and an additional administrator for 15 EUR. A tailored offer is quoted for institutions. Payment is processed by a secure provider, and the service may be suspended after a 15-day formal notice in the event of non-payment.
- free - 100 lifetime pages (one-off credit)
- 200 MB of storage
- 250 MB per file
- 50 AI queries per month
- 1 administrator
- unlimited readers
- 7-day log retention
- 29 EUR/month excl. VAT - 1
- 000 pages per month
- 2 GB
- 150 AI queries per month
- 2 administrators
- email support
- 30-day log retention
- 129 EUR/month excl. VAT - 5
- 000 pages per month
- 10 GB
- 1
- 000 AI queries per month
- 3 administrators
- priority email support
- 6-month log retention (flagged 'Popular')
- 299 EUR/month excl. VAT - 15
- 000 pages per month
- 25 GB
- 1 GB per file on bulk import
- 3
- 000 AI queries per month
- 3 administrators (+15 EUR per extra administrator)
- unlimited bulk import
- 899 EUR/month excl. VAT - 45
- 000 pages per month
- 75 GB
- 9
- 000 AI queries per month
- 5 administrators
- public AI chat (announced for March 2026)
- per-project API access
- on quote - dedicated sovereign infrastructure
- SSO and directory integration
- white label
- assisted migration
- 99.5%+ SLA and extended API quotas
- Pages pack x3 (Starter) up to x10 (Pro)
- AI pack x2 to x10
- Storage pack x1 to x10
- extra administrator +1 to +5
Data, GDPR & hosting
A consolidated view of how Patrivox handles your data.
GDPR overview
GDPR implementation is documented in detail. The privacy policy and the terms of service are both version 1.0 dated 11 February 2026. The controller is the publisher, a sole trader based in Tournai, Belgium, registered under BCE 0640.997.081; a data protection officer is reachable by email and by post, and the terms (6.4) add a second address for exercising rights. The stated framework is EU Regulation 2016/679 and the Belgian act of 30 July 2018. Seven rights are detailed - access, rectification, erasure, restriction, portability, objection and withdrawal of consent - answered within one month, extendable by two. Breaches are notified to the Belgian Data Protection Authority within 72 hours and to data subjects where the risk is high. A separate, detailed DPA is signed on request. No transfer outside the EEA, no non-European subprocessor. Belgian law, Tournai courts.
Who owns the data?
Users and their institutions keep all intellectual property rights over the content they upload, and the terms state that the publisher acquires no ownership of it (ToS 5.2). The licence granted to the publisher is non-exclusive, worldwide, limited to the subscription period and restricted to operating the service - storage, OCR, indexing and search - and it ends when the content is deleted or the subscription terminated. The publisher claims no intellectual property over AI-generated outputs (ToS 5.3): OCR and AI results are derivatives of the customer's content and remain freely usable. Where an institution uploads records naming individuals, it acts as controller and the publisher as an Article 28 processor.
Reuse rights
Customers may reuse everything the platform produces without asking permission: original documents, extracted text and metadata can be exported in full at any time, and per-document sharing links can be issued with configurable expiry, an optional view cap and a password. On the publisher's side, the stated purposes are service delivery, commercial administration, security, communication, service improvement through anonymised analysis, and legal obligations, on the legal bases of contract performance, legal obligation, legitimate interest (security, improvement) and consent for functional and analytics cookies. OCR runs through Mistral OCR; for AI queries, the user's question and the relevant passages are sent to language models. The privacy policy (6.3) commits to no training of third-party models on customer data, no sharing of results between institutions, strict separation between organisations, and contractual non-reuse guarantees obtained from model providers. Recipients are limited to authorised staff, technical subprocessors, a PCI-DSS payment provider and competent authorities. Data is never sold, rented or transferred to third parties for commercial or advertising purposes. Authentication is handled by Better Auth.
Data retention & training
Hosting summary
The publisher commits to hosting and processing data exclusively on infrastructure located in the European Union. The privacy policy (7.2) names four subprocessors: Scaleway for S3-compatible object storage in France, Neon for the PostgreSQL database, Typesense for full-text search and Qdrant for the vector database, all in the EU. The FAQ states that storage is on Scaleway in Paris and OCR runs through Mistral AI in Paris, with no data transiting outside Europe. No transfer outside the EEA takes place and no non-European subprocessor is involved; should a transfer become necessary, the publisher undertakes to rely on an adequacy decision or standard contractual clauses. Organisations are strictly isolated from one another, including at infrastructure level. One nuance is worth noting: an independent network check resolves the domain to 188.245.225.89, AS24940 Hetzner Online GmbH, in Falkenstein, Germany. That is still EU territory, but Hetzner is not named in the published subprocessor list. Quote-based options extend the arrangement with an isolated S3 bucket, bring your own storage, a separate instance, or on-premise deployment at a European host of the customer's choosing.
Things to keep in mind
Risks and trade-offs to weigh before adopting Patrivox.
- The '100% European hosting' claim is supported by four named EU subprocessors, but the web server resolves to Hetzner in Falkenstein, Germany, which is in the EU yet absent from the published list: ask the publisher to reconcile the two
- Outdated example prices in the English version of the add-on block (99 EUR, 199 EUR, +2,500 pages) against 129 EUR, 299 EUR and +1,000 pages on the plan cards: have the applicable price confirmed in writing
- In-app strings announce a 20% annual discount and a '99.9% SLA' where the public pages announce 15% and '99.5%+', and one checkout string mentions a 14-day free trial that appears nowhere on the pricing page
- Age rules diverge: the privacy policy sets a minimum age of 16 while the terms restrict registration to adults, which matters for institutions working with school or student groups
- No SOC 2 or ISO certification is claimed; the guarantees rest on contractual commitments rather than on a third-party audit, so compliance teams will need to review the contract itself
- Only the French version of the terms and of the privacy policy is legally binding, whatever language the interface is used in
- API access is listed in the plan comparison but no public documentation exists to evaluate it, and the publisher is a natural person rather than a company: continuity of service is for the buyer to assess before entrusting a collection to the platform
Setup & Integrations
Technical difficulty
Low. There is nothing to install: the platform is a web application, so an account can be opened and used straight away. No manual data entry and no configuration are announced, with results reported about two minutes after upload, and readers consult without creating an account. Technical effort appears in only two places: bulk import through an S3 client, available from the Association plan, and API access by key on the Pro plan. Migrating an existing collection (CSV, EAD/XML, Access or FileMaker databases, loose scans) is offered as a quote-based assisted service rather than left to the institution.
Deployment
Integrations
Supported languages
Behind Patrivox
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
How does the OCR work?
Where is the data stored?
What is the difference between administrators and readers?
Can the AI chat invent answers?
Is mass import possible?
Is there a discount for associations?
What happens if I exceed my AI query allowance?
Should I pay monthly or annually?
Can I change plan?
What happens to my data if I cancel?
Should you pick Patrivox?
Patrivox is a niche tool, closely aligned with a real heritage need and built by a practitioner rather than by a generalist vendor. The technical chain is credible - Mistral OCR, entity extraction, Typesense for full text, Qdrant for semantics, a chat that cites its sources - and the subprocessor list is fully European and named, with no transfer outside the EEA. Unlimited readers on every plan, the free one included, genuinely fits institutions whose mission is public access. The reservations concern maturity more than design. The domain was registered in January 2026, the blog and the changelog were still empty in August 2026, and the traction figures displayed are modest. The publisher is a Belgian sole trader with no team page and no social presence, so continuity of service is a question a buyer should raise directly. No third-party certification is claimed: the guarantees are contractual, backed by a DPA signed on request. Internationalisation inconsistencies also deserve a check before signing, since the English add-on example quotes outdated prices and the in-app strings announce a 20% annual discount where the public page announces 15%. The pricing grid is otherwise readable, with a permanent free plan, a 29 EUR entry point, add-on packs that avoid an early plan change, and a 15% association discount. Recommended for municipal archives, historical societies, heritage libraries and church archives that already hold digitised but unsearchable material, that require EU data residency, and that can accept a young product from a very small publisher. Institutions needing a mobile app, standard self-hosting, a documented public API or audited certifications should either wait or negotiate through the custom quote offer. A 30-minute or 60-minute demonstration is the natural next step.
- Choosing a selection results in a full page refresh.
- Opens in a new window.