What's new

Marvin Systems can see!

Beyond text and metadata, Marvin Systems goes even further: the visual elements in your documents (logos, signatures, stamps, faces, etc.) are now detected automatically.

The platform now allows you to group all data types within a category under a single label, thereby expanding detection and standardizing the sequencing of pseudonyms.

• Example without grouping: `90-64 Melrose Ave, London NW2 4JT` → `ADR1, LOC1 CP1`
• Example with grouping: `90-64 Melrose Ave, London NW2 4JT` → `POS1`

Grouping can be done directly from the selection panel for the data types to be detected.

In addition, file names are now processed by default and anonymized or pseudonymized depending on the selected processing method.

It is now possible to create your own data types when the available types do not fully cover your use cases.

Once the new data type to be detected has been created, new sessions will include it in addition to the already selected data types.

In each session, you can now:

• Correct the boundaries of an incorrectly detected entity when a text fragment has been mistakenly aggregated to a pseudonym.
• Merge or split pseudonyms.
• Set a new reference value.

The reference value is the value used by the technology when restoring pseudonymized content.

Generate and download your organization’s processing history.

Each session now keeps a record of every processing operation: timestamp, metadata (file name, size, creation date, etc.), types of entities selected for detection (person’s full name, company name, email address, folder name, etc.), number of words analyzed, and number of occurrences detected.

The processing log does not contain any personally identifiable information.

This is available directly from your user account.

A dedicated page allows you to view the status of Marvin Systems platform services and, in particular, to receive real-time notifications of issues. This page is available under “Help and first steps”.

Grouping of pseudonyms is now configurable.

Why two modes? Depending on the use case, instances of the same entity may need to be matched more or less broadly. Two settings are now available to match values against one another.

Three modes to choose from:

• Strict: Matches values of the same type, including those nested within one another. Example: “Mathilde,” “Dupont,” and “Mathilde Dupont” are recognized as a single entity.

• Dynamic: Matches values across different types (such as individuals and companies), as well as syntactic variations, including spelling errors. Example: “Isabelle,” “Nguyen,” “Nguven,” and “EI Isabelle Nguyen” are recognized as a single entity.

• No grouping

Enabled by default → dynamic grouping is enabled for all logged-in users. The mode can be adjusted directly in the settings.

Marvin Systems now supports .msg files.

Your entire email conversation thread, along with all included attachments, can now be anonymized or pseudonymized with a single click.

Specifically, the credit calculation algorithm now only counts unique character sequences in a document and ignores repetitions and separator characters. Credit usage is adjusted accordingly, without any action required on your part.

Take advantage of comprehensive technical documentation, right at your fingertips.

This comprehensive documentation, in the form of a “user guide,” is now available directly from your user account in a dedicated section.

It helps you understand how the platform works, configure your setup, or explore the technology on your own.

Pseudonymisation does not stop at the body of the text. To cover the entirety of a document, every field that may contain identifying data must be processed.

• Images contained within Office documents can now be anonymised. There are three options to choose from beforehand to configure this feature: skip processing, remove the images from the final document (if they are not relevant to your analysis), or process the images. As usual, the document’s original format is preserved!
• Identifying data found in document comments, their authors and in version tracking blocks is now included in the detection scope.
• Marvin Systems now supports your PowerPoint presentations, adding another format to the list of supported documents.

Do you need to create automations? Access to the Marvin Systems API is available from the Pro plan onwards. Good news: this new feature comes with a new data management policy!

• You can now integrate Marvin Systems with your own tools using the API and create the automations you need.
• Configure data retention in just a few clicks: you can now choose how long your data is retained (ZDR policy, no retention, or the default setting for files).

The retention period is now configurable – why is this?

In practice, a user works on a case over time. From the very first documents processed, pseudonyms are created and associated with that case. Whether using Claude Cowork or the web application, they can upload new files linked to the same case at any time and find their pseudonyms intact, allowing them to continue working on the case with complete consistency.

However, not all contexts are the same. If pseudonymisation is not central to your use case, or if your internal policy requires it, you can opt for the immediate deletion of all data.

Greater granularity, without compromising on confidentiality!

The editor plays a key role in how you use Marvin Systems. The interface has been designed and is constantly being improved to meet users’ needs with ease.

• The editor/preview toggle has been redesigned to streamline the interface.
• Automatic zooming in the PDF reader has been restored when layout changes are made.
• Loading times have been optimised during saves.

Several improvements have been made to the session system and the management of scanned documents.

Sessions can be renamed and can now hold up to 50 files each. Large scans are now processed more efficiently, resulting in smaller output files.

When a file is not supported by the detection engine, the error message is now clear, including when using the technology directly within Claude Cowork, and unexpected binary files are detected at an early stage.

Secure guest access allows you to create sessions and process documents without an account, then retrieve them once you have registered.

• Use without logging in, to explore our technology in just a few clicks.
• Most guest restrictions have been lifted: you can explore all features (only exporting remains restricted to logged-in users).

You can now also download all files processed during a session in a single ZIP archive.

Support for Word documents has been improved.

• The document editor now allows you to view Word documents with detected entities, following the same format as PDF documents and text files.
• Line breaks are now handled correctly, resulting in a display that is more faithful to the original document.

A new page lets you paste plain text and anonymize it instantly, without going through a file.

The app has been completely redesigned around sessions, which are designed to mirror your working folders.

• List of sessions and a dedicated page for each session.
• Real-time updates to your files.
• ‘Editor’ and ‘Pseudonyms’ tabs to review your results.
• ‘Restore’ tab to access documents shared by your AI assistant with a single click.
• New home page.

Choose to export your documents, once processed, in their original format or as Markdown. The Markdown format converts the document to plain text and adapts the layout. In particular, it optimises interaction with AI assistants.

Until now, restoration (de-pseudonymization) only worked on plain text. This was useful for retrieving a text response from an AI assistant, but it was limited when it came to formatted files.

Now we’re taking it to the next level: Restoration now supports Word, Excel, and other files. With just one click, you can restore the file generated by your AI while preserving its formatting. No more copying and pasting the content into a new document and reformatting it afterward.

In practical terms, what does this change? You ask your AI assistant to rework documents that have been pseudonymized beforehand using Marvin Systems and to generate a new one. You run the new file through Marvin Systems, and you get it back in its final, ready-to-use version, with the original values restored in place of the pseudonyms and the formatting intact.

A dedicated integration enables Claude Cowork to pseudonymise your documents before processing them. Claude never has access to your documents in plain text once you use the Marvin Systems skill.

• Access to the web platform’s features directly from the Claude Cowork environment.
• File preview and manual approval before processing by Claude.
• Trigger the Marvin Systems skill at your own discretion: you remain in control of when your files are sent via our API.
• Automatic detection of new versions, with a prompt to update the skill to benefit from the latest fixes.
• Monitor your usage via the API.

The editor provides a side-by-side preview of the original document and its anonymised version, with synchronised scrolling.

• Side-by-side preview of the original and anonymised versions, with synchronised scrolling to allow you to easily view how the detected entities appear.
• Support for landscape-oriented pages.

Did you anonymize data by mistake? Don’t panic.

You can undo anonymization with a single click. The original text reappears, as if nothing had happened.

Do you know in advance that a specific entity should never be anonymized? With a single click, add it to the exclusion list.

Result:

• The entity is restored in the current document.
• It is automatically ignored in all other documents in the session.
• No need to go back and check document by document (your preferences are saved for the entire session).

Of course, an exclusion list also means an inclusion list...

The Pro plan is now collaborative: create an organisation and invite members to share your subscription.

• Centralised member management.
• Invitation system to grow your team.
• Preferential pricing for each additional licence.

Freely draw areas to redact on the document (logo, letterhead, signature, initials).

• Free redaction: draw your own areas to hide and duplicate these areas on every page of the document.
• Open the results directly in Word or Excel using the new buttons.

In practice, the same person may appear in various forms in your documents: full name, first name only, abbreviation, alternative spelling (Anne Pascolini, A. Pascolini, Mrs Pascolini, Mrs Pascolini). These should not be treated as separate entities.

Our technology recognises them as a single entity, using the context of the document to make an accurate determination. Without intelligent grouping, the analysis becomes fragmented.

Greater clarity, less ambiguity, and faster proofreading. And behind the scenes, the system is also more robust: reduced processing time and better performance with large or multi-entity documents.

Whether you’re using an integration such as Claude Cowork or the web app, the session system has been designed to allow you to find the items you’ve been working on and return to them later.

The editor offers new tools for greater versatility and complete control over data redaction.

• Toolbar (text selection, free-form redaction, redaction removal) with zoom and page selection.
• Format options: original or Markdown.

A new editor allows you to easily view all detected entities, ensuring you have complete control over the processing of your documents.

• Review, edit and create instances.
• View redacted areas directly on the document.

Word and Excel documents are now supported, with their formatting strictly preserved so that you can keep a document that is fully usable.

Contextual detection is becoming more accurate.

• A more efficient detection model.
• Automatic retry and a reload button in the event of a failure.

The processing limit has been increased to 250 pages per document, with an automatic retry in the event of a failure.

Detection capabilities have been expanded for even more accurate processing.

• Numerous identifiers specific to France.
• New types of entities: organisations or institutions, and brands, products or services.

Our blog is here! Every month, discover a new post written by Camille, our beloved CEO!

The technology now takes the context of the text into account when identifying personal data, going far beyond simple rules and patterns.

In addition to anonymization, a new type of data processing is now available: pseudonymization. You can replace identifying data with tags specific to each entity, which are consistent across one or more documents and reversible.

A restoration tab is available, allowing you to restore the original values in place of the pseudonyms.

The scanning process is divided into concurrent tasks, and large texts are split up dynamically, for faster and more accurate results.

• You can upload several documents at the same time.
• Images are compressed to reduce the size of your files.

Uploading documents is now more convenient.

Password-protected files can now be processed. New PDF features will gradually be supported, such as embedded forms, electronic signature management, etc.

A text extraction engine can now detect identifying data in scanned PDFs and images.

You can now select the categories of data to be detected, with a badge showing the number of active types.

As not all Marvin Systems users have the same needs, the tool has been designed to offer the greatest possible versatility whilst ensuring the highest standard of quality.

As well as text, the technology now supports PDF documents. You can anonymise or pseudonymise your files with a single click. The document format is preserved!

In addition to PDF documents, text files (CSV, Markdown, JSON, etc.) are now supported and processed by our analysis engine.

Detection is being extended to new categories of information that are directly or indirectly identifying.

Postcodes, dates, software or accounting entries, etc. As iterations progress, new types of entities are emerging.

Pattern-based rules complement automatic detection to accurately identify regular formats (numbers, URLs, email addresses, etc.).

The very first version of the technology allows users to submit a text and receive an automatically anonymised version, in which personal information is replaced by placeholders.

Marvin Systems is no longer just an idea; it is taking shape: a technology capable of protecting the personal data contained in a document before it is shared or processed.