What we do with your mail.

I · What we hold

The kinds of things kept on your behalf.

Account

Your email address, the timestamp you signed up, the domains you register inside the product, and whatever billing handle Stripe returns to us. No password — we never set one.

Mail itself

Every message that arrives at a domain you've added: the raw RFC 822 envelope (sender, recipients, date, headers, body, attachments) and a parsed copy used for the reading view. Outbound: the same, plus the DKIM-signed copy SES held while it was being delivered.

Passkeys

The public half of any WebAuthn credential you register, plus a label and the timestamps of registration and last use. The private half stays on your device; we never see it.

Deliverability

DMARC aggregate reports for the domains you've added — the reporting provider, the sending-source IP addresses, and pass/fail counts — parsed into your Deliverability view. These are machine reports about your own domains, not the contents of anyone's mail.

Operational

Sign-in events, magic-link issuance, basic request logs. Kept for debugging and abuse handling, not for analytics on you.

II · Where the letters live

Two stores, one reader.

III · Who else touches the mail

The sub-processors, named.

• Amazon Web Services

  S3 (raw mail and attachments), DynamoDB (the reading index), SES (outbound delivery, with per-domain DKIM signing), CloudFront (the website you're reading), KMS (DKIM master key). All in one AWS region. The store of record.

• DigitalOcean

  The Haraka inbound MTA — the program that accepts incoming SMTP — runs on a DigitalOcean Kubernetes cluster. Mail traverses it, then is handed to S3.

  Search index. So you can search the full text of your mail, a copy of each letter's subject, sender and recipients, the address it reached, its labels, attachment names, date and body text is kept in a search index on the same DigitalOcean cluster, on DigitalOcean block storage. We run that search server ourselves; it is not a third-party search service, and only Folio's own servers can reach it. A letter's entry is removed when you delete the letter, and all of your entries when your account is deleted.

• Siftfy

  Spam classification, and bot screening on sign-in and one public form. Siftfy is not an outside vendor: siftfy.io and Folio are both operated by VectraSEO LLC, the same company. It is listed here anyway, because your data moves between two products and you should not have to read a corporate filing to learn that.

  Incoming mail. The text of an incoming message is sent to Siftfy's scoring service so we can tell junk from mail worth reading. When Siftfy's own model answers, they state that the text is classified in memory, written to no database, log, backup or support system, and discarded once the score comes back.

  Mail Siftfy is unsure about, from 2026-09-27. When Siftfy's model is unsure about a message, Siftfy may send its text, with credentials removed, to TypeSafe AI, Inc., a third-party classifier listed below, and TypeSafe's answer becomes the spam score we act on. The text of a message sent to TypeSafe is also kept in Siftfy's private evaluation storage for 30 days from when it arrived, where a Siftfy reviewer may read and label it. Siftfy states that text from its mail-scoring service, including text it sends on to TypeSafe, is not copied into its training store, and that TypeSafe's answers about mail are not used for training. The way a message of yours does reach that store is a correction you make, described next.

  Your spam corrections. When you mark a message as spam or not spam, the text of its body, without any earlier message quoted below a reply, is sent to Siftfy with which of the two you chose and a pseudonymous identifier for your account (a keyed hash, not your address or account number). Siftfy uses it to train its spam model and keeps it for 12 months from the last time that text was reported, then deletes it. That model also screens text for Siftfy's other customers, so a correction you make can change how it judges theirs. The copy is personal information and it is not anonymized.

  A correction is sent only when your mark changes where the message was filed: reporting a message already in Spam, or rescuing one that was never there, sends nothing. Nothing in Sent or Drafts is sent, nor mail from your own address or one of your domains, and only a limited number of corrections a day are sent for any one account. Each correction is sent once; if Siftfy cannot be reached, it is not sent again. Corrections made before this paragraph appeared were never sent.

  Signing in and signing up. When you ask for a sign-in link or enter the six-digit code, your sign-in email address is sent to Siftfy's Content Gate, together with signals from your browser about whether it is being driven by automation, so we can turn away bots before any mail is sent. The code itself is never sent. Siftfy states that checking a submission stores none of its text, and that a submission made up only of an email address is not kept for evaluation.

  The deletion form, which is different. The same service screens the free-text reason on our account-deletion request form for automated abuse — and here the text is kept. Siftfy retains the full contents of that field for 30 days to measure how well the screening performs. That text is personal information. We are not calling it anonymized, deidentified or aggregated, because it is none of those things. Submissions our own checks judge automated are rejected before Siftfy is contacted, so it is the submissions we judge genuine whose text is kept. The rejected text never leaves Folio and is not stored: we record only which check fired, and when. Siftfy keeps what it does receive to measure how the screening performs.

  Training, from 2026-09-27. Siftfy reviewers read some of the submissions it keeps and may label them legitimate, spam or uncertain. From 2026-09-27, a submission from this form that a Siftfy reviewer has labeled legitimate or spam may be used to train Siftfy's models: the spam model, the model that judges AI authorship, and the automation (bot-detection) model. Submissions that neither a reviewer nor TypeSafe (below) has labeled legitimate or spam are not used. The copy kept for training is held for 12 months from the day the submission was collected, then deleted. Credentials are removed from it, as they are from the 30-day copy, and nothing else is: the text can contain personal information, and it is not anonymized.

  One exception, reaching back. Submissions a Siftfy reviewer labeled between 2026-09-22 and 2026-09-27 are also used for training. They were collected while this page said Siftfy would not train on them. That is a new use of text people had already given us, and we would rather say so here than have it found later.

  TypeSafe's verdicts, from 2026-10-08. Some submissions are also judged by TypeSafe AI, a third-party classifier. From 2026-10-08, a submission TypeSafe judged spam or legitimate may also be used to train Siftfy's models, even if no reviewer has read it; where a reviewer has labeled the same submission, the reviewer's label is used instead. This reaches back as well: it covers submissions collected before 2026-10-08, while this page said TypeSafe's output was not used for training, for as long as Siftfy still holds their 30-day copy. The copy kept for training is held for 12 months, as above.

  This page said nothing about the form until 21 September 2026. It described the mail path and let the reader carry those assurances across to the form, where they do not hold. Apart from the corrections you make, Folio sends nothing about your mail to Siftfy for training — the rest of the training described above covers only text typed into the deletion form. What Siftfy keeps and for how long is stated authoritatively at siftfy.io/privacy — linked rather than paraphrased, because the paraphrase is what went stale. If that ever changes, this page changes first — before it changes, not after.

• TypeSafe AI

  A third-party classifier that Siftfy uses when its own model is unsure. Siftfy, not Folio, sends it the text of some incoming messages and some account-deletion requests, with credentials removed, as described under Siftfy above. TypeSafe's privacy policy states that inputs are not used to train or fine-tune models. Neither we nor Siftfy claim a retention period at TypeSafe; its data processing terms govern that.

• Stripe

  All billing. We never see your card; Stripe gives us back a customer handle and the subscription state. If you cancel, Stripe handles refunds and we mirror the state.

• Google Analytics

  Loaded on public marketing pages (the Landing, Pricing, About, FAQ — the sort of page you're on now) only if you allow it. Open the Analytics tab and choose Allow; we load nothing from Google until you do, and declining keeps it off. It is never loaded inside the authenticated inbox, and you can change your choice from the same tab on any public page.

We don't sell data. We don't sell ad inventory. There is no surveillance layer on top of the inbox itself, ever.

IV · The reader's rights

Every right you'd expect, spelled out.

1. 01

   Read it back.

   You can ask for a complete export of your archive — raw .eml files in a single tarball — and we'll produce it. Most archives finish inside a working day; very large ones, a little longer.

2. 02

   Take it away.

   Delete any message or any domain from inside the app. To delete the whole account, write us using the deletion request page — account records are removed immediately on confirmation; the S3 objects expire on the lifecycle stated above. Tell us if you need them purged sooner and we will.

3. 03

   Correct it.

   Fix anything wrong on your account — address, billing handle, domain ownership — from the relevant page, or by writing to us. We'll respond in plain English, not a ticket number.

4. 04

   Object.

   If you live somewhere with stronger statutory rights — the EU, the UK, California, others — those rights apply here, not in addition to ours. Write and tell us what you'd like done and we'll treat it as the binding instruction it is.

5. 05

   Choose who reads it.

   If you connect an AI assistant — Claude, ChatGPT, Cursor or another — it becomes a recipient you authorized. It can search and read your mail and calendar, never Spam and never attachment contents, and it can't send, delete or change anything. What it reads leaves Folio and is handled under that assistant's own terms. We log only which tool it used and how many items it got back, never your searches or message text, for 90 days. Revoke it from Connected apps and it stops within seconds; changing your sign-in address or deleting your account revokes every assistant at once.

V · The small print, made small

Cookies, security, children, change.

Cookies

One essential session cookie for sign-in, which needs no consent. Google Analytics sets its own cookies on public marketing pages only after you opt in through the Analytics tab (see III); decline and none are set. No other tracking pixels.

Senders' tracking pixels

Remote images inside a letter are fetched by Folio and re-served from our own origin, so your browser never contacts the sender's server. This hides your IP address from senders: a tracking pixel in a marketing email records Folio's server instead of your IP address, approximate location, device or user agent. Folio fetches the image when you open the letter, so the pixel can still record roughly when it was opened.

Where it's processed

Your data is processed in the United States (Amazon Web Services, us-east-1). For visitors in the EEA and UK, transfers rely on the EU–US Data Privacy Framework, under which AWS is certified, with Standard Contractual Clauses as a fallback.

California

We do not sell or share your personal information, and we use the contents of your mail only to run the service for you — never to infer characteristics about you. California residents have the rights to know, delete, and correct, and to be free from discrimination for exercising them. To make a request, email privacy@folioinbox.com.

Encryption

Mail is encrypted in transit between MTAs wherever the other side supports it (TLS), and at rest in S3 and DynamoDB. Per-domain DKIM private keys are Fernet-encrypted in DynamoDB; the master key that decrypts them is held in AWS KMS.

Children

The product isn't designed for, marketed to, or sold to anyone under sixteen. We don't knowingly hold accounts for minors.

Breach

If something goes wrong with your data we'll tell you about it — in writing, by email, with what happened, what we know, and what we're doing. No silent disclosures.

Changes

When this page changes substantively, we revise the date at the top and email subscribers a one-paragraph note describing the change.