[ 07 / FAQs ]
Frequently asked questions
Plain answers about the products, how they handle data and what you still need to check.
The product
What is Synthaveil?
Synthaveil replaces selected identifying details with realistic stand-ins before you share a copy. It has tools for documents and prompts, plus an engine and workflow integrations. Choose the product that fits the task, then review the output.
Does Synthaveil use AI on my data?
No generative AI or large language model ever reads your real data. Synthaveil’s own engine makes the swap, and with the same settings it gives the same result for the same input every time. If you then use an AI tool, for example through Alias, that tool only receives the stand-ins.
Does my data have to leave my environment to reach you?
Not if you run Legend yourself: Synthaveil’s engine then works entirely inside your environment, even with no internet connection. Alias and Cover send only the details they find to the Synthaveil API, never the whole message or document. Each product page sets out where its data goes.
Can the stand-in values be turned back into the real values?
Yes, on your side. When an AI tool replies using stand-ins, Synthaveil puts the real values back, so the answer reads with your real names and is ready to use. Alias, Cover and Legend all restore values; each product page sets out how.
Does it work with ChatGPT, Copilot and Claude?
Yes. Synthaveil prepares data for AI tools through several workflows: Alias works in supported chat composers, and Cover works in Word and Excel. Both are available for private evaluation; the product pages explain how to request a build.
The legal ground
Is pseudonymised data still personal data under GDPR?
For you, yes: you hold the key, so it remains personal data in your hands. The EU Court of Justice (EDPS v SRB, 2025) and the ICO (2025) both look at whose hands the data is in: a recipient with no key, no mapping and no other route to the person may not hold personal data at all. That is what Synthaveil sends an AI provider: stand-ins with no route back. Your duty to tell people about the sharing still applies when you collect their data.
What do I gain by pseudonymising first?
The AI provider or vendor gets a useful copy with no key or mapping back to your customers. With Synthaveil, a model can draft a response about a complaint without receiving the customer’s name. Review the remaining context for clues before you share; no particular regulatory outcome is promised.
Does Synthaveil make my data anonymous?
No. Synthaveil makes no claim that its output is anonymous. In your hands it remains personal data, because you hold the key. What changes is what the recipient gets: stand-ins with no key or mapping to link them back. Review the remaining context before you share.
Does using Synthaveil make us compliant?
Synthaveil gives you a strong technical safeguard to build your compliance on: the copy you share carries stand-ins instead of real identifying details. You still decide the purpose, permissions, access, retention and records that make the whole activity appropriate.
Can we use it with health data?
Yes, by agreement. Health data and other special category data can go through Synthaveil when your agreement with us covers it. Tell us before you start and we set it out in writing.
How it works
Do you have to store a mapping table to keep pseudonyms consistent?
No. Synthaveil keeps stand-ins consistent without storing any real-to-stand-in pairs. Products that restore values keep that information on your side, in your browser, workbook or n8n environment, and it needs protecting.
Can a replaced postcode still give away the real location?
Synthaveil replaces the postcode with a real one that keeps the area useful without pointing back to the original. Other fields can still reveal a location, so check the complete record and your chosen policy.
Will swapping postcodes break my analysis?
If your analysis uses area characteristics, a replacement can keep the ones you select. If it measures something tied to the exact address, such as travel distance, the result will change. Test your actual questions against representative output before relying on the results.
Can an AI vendor build a profile from months of our prompts?
Synthaveil stand-ins change on a schedule you set, which makes it harder to link prompts from one period to the next. It does not erase data the vendor has kept or stop linkage through unchanged details. Choose the schedule for the task, then test it.
Against the alternatives
How is this different from a vault, a gateway, or a redaction library?
Synthaveil keeps data usable and holds no vault of real values. Vaults store the real values, gateways filter traffic on its way to a model, and redaction libraries leave the running to you. Compare a specific deployment, then test it with your own task.
How is Synthaveil different from Presidio?
An open-source Python toolkit, started at Microsoft and run since June 2026 by the independent Data Privacy Stack project. It finds sensitive values in text and replaces them with placeholder tags, hashes, masks or values from a fake-data library. You run it yourself, wherever you like, and you wire up everything around it. Compare the configured replacement policy and consistency scope. A tag-versus-name comparison alone does not describe what Presidio can do.
How is Synthaveil different from an LLM redaction gateway?
A proxy that sits in the network path between your application and the model. Prompts pass through it, and what it recognises gets stripped or tagged on the way. Responses come back the same way. A proxy inspects the traffic routed to it and can be self-hosted. A local engine integration puts substitution in the application. Compare who operates each component and what it receives.
How is Synthaveil different from Limina?
A commercial service that detects and removes personal data from text, documents, images and audio. It runs as a hosted API or as a container inside your environment, and it can put the original values back into a model's response. Limina’s default swaps a detail for a label such as [NAME_1]. It can also make realistic replacements, which its documentation says are created with an ML model and are non-deterministic: the same person can come back with a different fake name each time, by chance. With Synthaveil you decide. A stand-in can stay the same for as long as you need, so one customer’s chats stay one customer and one patient’s letters, notes and results still line up. Or it can change each day or for each project, so separate pieces of work cannot be joined up. It is a setting you choose, not chance. The stand-ins also keep the shape your analysis relies on: an age stays close to the real one, and a postcode stays a real postcode that does not point back to the person. No generative AI or large language model ever sees your real data.
How is Synthaveil different from a tokenisation vault?
Enterprise platforms that replace sensitive values with tokens and keep the mapping between token and real value in a protected store. The store is the product: it is encrypted, access-controlled, audited and retained for as long as the tokens exist. Compare where mappings and keys live, who can restore values and how long the data is kept. Synthaveil keeps stand-ins consistent without holding a vault of real values; products that restore values do so on your side.
How is Synthaveil different from a synthetic test-data tool?
Tools that build a synthetic copy of a dataset, usually as a scheduled job against a database, so development and test environments can run on data shaped like production without holding production. Choose based on the task: generating a test dataset, substituting fields in a live workflow or preparing a prompt. Recovery is product-specific; generated synthetic records need not correspond to an original person.