Where your documents are, who sees them, and what the AI sees.
What happens to an answer before you read it
What passes the checks is what you read. What does not is listed as 'to verify'.
Location
Everything is stored in India. Uploaded files are stored in AWS Mumbai (ap-south-1). The Case File, chat and the pseudonymisation vault are in a managed database in Bangalore (DigitalOcean blr1); the vault's master key is held in AWS Mumbai. Encrypted at rest and in transit.
Isolation
Every organisation and every matter is isolated at the database level. A query from one matter cannot read another matter's rows. This is enforced by the database, not by application code.
What the AI sees
Before the text goes to any AI, nine kinds of identifier — phone numbers, email addresses, PAN, Aadhaar, account and IFSC numbers, UPI IDs, vehicle numbers and GSTIN — are replaced with tokens such as PHONE_1 or ACCT_A. Each is found by its written shape, so one that is misread or written differently is not caught. Names and addresses are tokenised only when written in a form the rules recognise — for example a party on your matter record, a name after Shri or Smt or beside s/o, or an address after r/o with a PIN code; any other name or address reaches the AI as written. The mapping is stored separately, encrypted, and never sent to the AI provider. A page that cannot be read from its own text — a scan, a photo, or a PDF page whose text cannot be extracted — can be sent in full to the AI provider for OCR before any masking; masking applies to the text read from it.
Pseudonymised text is processed by Anthropic's API (US) directly, with no intermediary. Anthropic may retain API inputs and outputs for a limited period under its commercial data-retention policy; we have not enabled a zero-data-retention arrangement at this stage.
Who can see your content
You and the other members of your organisation — every member can open every matter. The CaseFile AI team, who run the servers, can technically reach everything stored, including your uploaded files as uploaded; there is no separate approval step or reason log for that access.
Audit
Every review action, every correction, every page opened, every deletion is logged with who and when — identifiers only, never content.
Deletion
Delete a matter and its documents, case file, chat, drafts and encryption keys are removed; a receipt lists what was removed. A nightly copy of the database and the files is kept in AWS Mumbai, and the database provider keeps daily backups in Bangalore. Each copy is kept for about 7 days, then expires — so a deleted matter stays in backups until they expire.
What we do not do
No training of AI models on your content. No advertising. No third-party analytics scripts. No case-law citations from model memory. No outcome predictions.
Full details: Privacy Policy.