Trust and security

Last updated: 15 September 2026

Speech listens to you all day. This page is what happens to that: what we keep, who can reach it, which country it physically sits in, and which of the things a security review asks for we have not built.

Your data stays in India

Servers, database and backups run in AWS Mumbai (ap-south-1).

Privacy Mode keeps nothing

A switch in Settings. Audio and transcripts are discarded the moment they are transcribed.

Nobody reads your words

Not even us, unless you turn it on for your own account. Off by default.

Delete means delete

One page in your account removes the account and everything stored for it.

What we hold, and for how long

The switch you set in Settings decides this.

Privacy ModeStandard
Your audioNever written to diskDeleted after 60 days
Your transcriptsNever written to diskKept, and may be used to improve Speech's own accuracy and models
Account and usage recordsKept while the account exists. Counts and timings, not content.
Pictures Agent Mode made for youOne year
Bug reports you sentOne year
Database backups60 days in S3, 14 most recent on the server

Your transcripts are never sold, and never used to train anybody else's models. The privacy policy has the full detail.

Who can see your dictations

You, and by default nobody else.

How it is protected

In transit

TLS 1.2 and 1.3 only, with HSTS, a strict Content Security Policy, and certificates that renew themselves. The apps talk to nothing except our API.

At rest

The database sits inside the private network and is not published to the internet; nothing outside the application container can reach it. Nightly backups go to S3 with AES-256 server-side encryption and expire on a lifecycle rule after 60 days. The identity that uploads them can write backups and do nothing else.

Access

The admin console is a separate thing from the product: an account with raised dictation limits does not get it. Access is granted from a declared list on every boot, so a grant cannot quietly exist that nobody remembers making, and every change and every content read is audited.

In the code

Every change goes through a pull request with the full test suite, a dependency vulnerability scan of everything we ship, and a secret scan of the whole history before it can merge. Dependency updates arrive as pull requests weekly.

For companies

Where it runs

AWS ap-south-1 (Mumbai), India: the application, the database and the backups. Two of the model providers that transcribe or format a dictation run in the United States, so a request taking those paths crosses a border for the seconds it takes to process. Nothing about it is kept there. The full list, with what each one receives, is on the subprocessors page.

Your rights, and how to use them

Found a security problem?

Tell us before you tell anybody else: security@rupees.com. We will reply, keep you posted while we fix it, and will not pursue legal action against anybody who follows our disclosure policy.

What we do not have yet

Most pages like this one list only what a company has. This is the other half, because you are going to ask and we would rather you heard it here.

Privacy policy · Terms · Data Processing Addendum · Subprocessors