Privacy
retrievr is a document retrieval service. You send a question, it returns passages with their citations. If you upload documents, it stores and indexes them so it can search them. That is the whole of what it does with your data, and this page says where each part of it lives.
Your account
Signing in creates an account with Supabase Auth, our authentication and database provider. We receive the identity it verifies — your email address, and the provider you signed in with — and never see your password. The site sets one cookie, the session Supabase issues; there is no analytics script and no advertising tracker on any page, so there is nothing else to opt out of.
Your API keys
A key is stored as a SHA-256 hash. The key itself is shown once, when it is created, and written to your own machine by the connect script — it is never displayed again, and a copy of our database yields nothing usable. Each request stamps the key as used, so you can see activity on the profile page. Revoking a key stops it immediately; deleting the account removes it.
Your questions
A query is stored with the job it created, because the answer is fetched by a separate request and the search needs the words you sent. The job is readable for fifteen minutes and then answers 410 — the row stays in the database as a record rather than being swept. Each accepted query also writes a usage row: the endpoint, the latency, and the idempotency key if you sent one. Usage rows are what the balance can be reconciled against.
Deleting your account deletes these rows with it. Nothing else is kept, and no query is used to train anything.
Your documents
An uploaded file is stored in a private storage bucket, and its text, its chunks and its vectors are stored in the database so it can be searched. Only your account can search it: the scope is applied inside the query itself, so another tenant's search cannot rank, see or infer a passage of yours.
Deleting a document removes the stored file, its text, its chunks, its vectors and its ingest job. There is no trash and no grace period, which is why your agent is told to read the exact title back to you before it deletes anything. We do not look at your documents except to operate the service, and we do not train models on them.
Who processes it
- Supabase — the database, authentication, storage and the API functions.
- Netlify — this website.
- Lemon Squeezy — card payments. They are the merchant of record, so card details never reach us.
- Our own GPU endpoints — a query is turned into a vector by a model we run. The query text is sent for the length of that request, and a vector is what comes back.
Getting a copy, or getting it removed
Email contact@retrievr.io and we will tell you what is held about your account, or delete the account and everything under it. Deleting a document or a key yourself is immediate and is the same operation.
Changes
If what is stored changes, this page changes with it — it is written from the code, and a change that is not reflected here is a bug. The terms cover the other half: what you may expect from the service, and what it asks of you.