A desktop app that lets law and advisory firms use AI on confidential files. Client data is pseudonymised by AI running entirely on the computer, sent to any AI assistant, and the answer comes back with the real names.
Sensitive data has to be anonymised before it can be shared or processed, and sending it to a cloud AI is not an option.
A desktop app built for French law and advisory firms. Three detection engines run on the machine itself, every decision can be explained, a lawyer validates before anything is exported, and the exported files are rebuilt with the redactions burned in.
A client file is pseudonymised, sent to any AI assistant, and its answer pasted back: the tokens turn into the real names, on this computer only. Within a matter the same person always gets the same replacement, and a safety checklist runs before anything is exported.
Rules and checksums, a language model for names and a local LLM check each other’s findings. Each one leaves a readable trace: who found it, with what confidence, and why. When they disagree, the lawyer has the last word.
Scans and photos are read by offline OCR while a vision model looks for faces. The export rebuilds the file in its original format, with text redacted in the pixels and faces blacked out, not hidden under a layer.