Security
An assistant that reads everything, run by a company that reads nothing.
Bloosh is only useful if you give it the real material — your inbox, your calendar, your contracts, the things you would not put in a shared doc. That only works if the trade is honest, so here is exactly who can read what.
Your chat is end-to-end encrypted
Chat content is encrypted before it leaves your device. What reaches our servers is ciphertext. Bloosh staff cannot read your conversations, a subpoena served on us cannot produce them in readable form, and a breach of our database does not expose them.
The keys are yours, not ours
Encryption keys are tied to your account and held by you. We do not escrow a copy, and there is no administrative override that lets us decrypt on your behalf — including when we would like one.
The honest consequence: losing your keys means losing that data. A system where we could recover it for you is a system where we could read it, and we would rather not be able to.
Zero data retention, across the whole stack
Zero data retention is enforced at every layer that touches your content, including the model providers behind the assistant. Nothing you send is retained after the request that needed it, and nothing is used to train a model — ours or anyone else’s.
Workspaces are isolated in the database
Every row in Bloosh carries the workspace it belongs to, and access is enforced by row-level security in Postgres rather than by application code remembering to filter. A query that forgets the check returns nothing instead of returning someone else’s work.
Connection credentials live in a vault
The tokens that let Bloosh reach your mailbox and calendar are stored encrypted in a dedicated secrets vault, never in an ordinary table. The application keeps only a reference to the secret, and the secret is resolved server-side at the moment it is used.
Disconnecting an account destroys the stored credential, and you can revoke Bloosh’s access directly from your Google or Microsoft account at any time.
Nothing happens without your confirmation
The assistant proposes; you decide. Outbound email, calendar changes, and deletions all arrive as cards you confirm. There is no autonomous mode that sends mail on your behalf while you are not looking, because an agent that can act unilaterally is a security surface regardless of how well it is encrypted.
Automations are the one thing that runs unattended, and they are created switched off. You read what one will do before it does it, and you turn it on yourself.
What we ask your accounts for
Bloosh requests the narrowest set of permissions that lets it do the work, and each one maps to a feature you can point at:
- Read Gmail — Indexing, email search, learning your writing voice from mail you have sent, and reading your display name and signature so outbound mail matches your client.
- Send Gmail — Sending a draft you have confirmed, from your own address.
- Calendar events — Two-way sync of events with Google Calendar.
Decline any of them and the matching feature simply switches off; the rest of the product keeps working.
Questions we will answer directly
If you are reviewing Bloosh for a team and need detail this page does not cover — subprocessors, data residency, retention controls, or a security questionnaire — write to [email protected] and a person will answer.
See also our privacy policy for how personal data is handled.