Kuju Email is in private development. This overview describes the platform as it will ship, so you can evaluate it before we open. Get notified at launch or read the overview.
Kuju Email · security
What protects your mail, and what doesn't.
This page describes how the service works today, including the parts we have not built yet. If a protection is not listed here, assume we do not offer it.
Kuju Email does not use end-to-end encryption, and messages are not encrypted per mailbox. That means the service itself can read stored mail — which is what lets it filter spam, search your mailbox and serve it to any IMAP client. It also means a small number of Kaimoku operators with access to the production systems can technically read stored messages.
Kaimoku site administrators can reset the password of any account. Today there is no per-mailbox access log: neither that action nor an operator reading stored mail is recorded in an audit trail you can see.
If your organization has a domain administrator, they can reset passwords and recover accounts within your domain.
Our Acceptable Use Policy says we do not routinely read the content of your messages, and lists the circumstances in which we may review it.
Our operators open the content of a mailbox only when you ask us to as part of a support request, to investigate a security incident, or when we are legally compelled to.
Where your data lives
Production runs on Akamai (Linode) in the Washington, DC region, in the United States. Your mailbox, its metadata and your account are stored there.
Off-site backups are stored in Cloudflare R2 object storage. Cloudflare chooses the storage location; we have not pinned it to a region.
The third parties that receive customer data are described in our privacy policy.
Encryption at rest
Message bodies and attachments are stored on a Linode Block Storage volume with Linode's volume encryption turned on. Every production server's boot disk is encrypted the same way. In both cases the encryption keys are managed by Linode, not held by Kaimoku: this protects against a lost or discarded disk, not against someone with access to the running system.
Credentials you give us for other services — for example the password for a mailbox you import from — are additionally encrypted by the application with AES-256-GCM before they are stored. The encrypted values live in our database; the key that decrypts them is kept in a separate secrets store.
Our database holds account records and message metadata, including senders, recipients, subject lines and a search index of message text. We have not confirmed that the volume holding the database is encrypted at rest, so we do not claim it.
Backups are not encrypted by us before upload. They rely on Cloudflare R2's own storage encryption, which we have not configured or independently verified.
Encryption in transit
Between your devices and Kuju
IMAP (port 993, or 143 with STARTTLS) and mail submission (port 587 with STARTTLS, or 465 with TLS from the first byte) negotiate TLS 1.3 with a valid certificate. POP3 is not offered: port 995 is closed, and mail apps use IMAP.
Webmail is served over HTTPS only, with HTTP Strict Transport Security set for one year across subdomains, so browsers refuse to fall back to an unencrypted connection. Calendar and contacts sync (CalDAV and CardDAV) also run over TLS.
Between mail servers
Our inbound mail server offers encryption (STARTTLS) to every sending server but does not require it — refusing unencrypted delivery would bounce legitimate mail from servers that cannot encrypt.
When we deliver your outgoing mail, we encrypt the connection whenever the receiving server supports it, but we do not verify that server's certificate — unless the receiving domain publishes an MTA-STS policy in enforcing mode, which we honour: then the certificate is checked and mail that fails the check is held rather than sent unencrypted. Without such a policy, this protects against passive eavesdropping, not against an attacker able to impersonate the receiving server.
For our own mail domain we publish an MTA-STS policy in testing mode and a TLS reporting address, so servers that send to us tell us when encryption to us fails; in testing mode they still deliver. We do not yet publish DANE or DNSSEC.
Sender authentication
Every message you send through us is DKIM-signed for your domain, so receivers can check it was not altered on the way and really came from a server allowed to send for you. The keys Kuju creates for your domain are 2048-bit RSA. We do not rotate them automatically.
When you add a domain, Kuju gives you the SPF, DKIM and DMARC records to publish, or publishes them for you if your nameservers point at Kuju, and checks each one. The DMARC policy it proposes for your domain is quarantine.
Our own domain publishes SPF and a strict DMARC policy (reject), so mail forged in our name is refused by receivers that honour it.
Passwords and sign-in
Passwords are never stored in readable form: we keep only a bcrypt hash.
You can protect webmail sign-in with an authenticator-app code. Passkey sign-in is not offered yet. Two-factor sign-in protects the web interface. Mail apps connecting over IMAP or SMTP, and calendar and contacts apps, sign in with your account password or with an app password you create in Settings > Security. Once two-factor is on, only an app password works in those apps — your account password is refused there — so a stolen account password alone does not open your mailbox to a mail app. An app password is a single secret: keep each one to one device, and revoke it from Settings when that device goes. Revoking it disconnects that app straight away, including a connection it already has open.
Web sessions use a cookie that page scripts cannot read, is only sent over HTTPS and is not sent with requests from other sites. Sessions expire after 24 hours by default and can be revoked from our side. If you think someone else has your session, use Sign out of all devices in Settings > Security: it ends every web session for your account, including the one you are using. Mail apps signed in with an app password stay connected, so revoke the app password to cut one off. Password-reset and invitation links are stored only as hashes.
When mail content leaves our systems
AI features
Kuju Email has AI features: an AI spam and phishing check, message analysis you ask for, and tools for working with mail such as summaries, reply drafts, task extraction and plain-language search. They are not offered on customer domains today. Nothing from your domain is sent to an AI provider, and a domain administrator cannot turn them on. We use them on our own domains while we finish documenting what each one sends.
Kuju's own spam and phishing checks run on every incoming message, on our servers; no AI provider is involved.
When we offer the AI features on customer domains, each will be off for a domain until its administrator turns it on and confirms what is sent, and the product will publish exactly what each one sends. Our privacy policy lists the AI providers we use.
Link-safety checks
When you run a link-safety check on a message, its links are compared, on our own servers, against a list of known phishing and malware sites that we get from Google. Nothing about a link is sent to Google unless it matches that list. When one does, we send Google a short hash prefix of the link to confirm the match, never the address itself. The check is off unless a domain administrator turns it on.
Spam filtering lookups
Our spam filter checks every incoming message against two reputation services, Abusix and Mailspike, with DNS queries from our own servers. It runs on every domain; it is part of how mail is delivered, not an option.
Abusix is sent the addresses of the servers that sent and relayed the message, the domain names in its links and in its DKIM signature, and, for links that look like URL shorteners or file-sharing links, a one-way hash (SHA-1) of the link's host and path. When you send mail through us, the address you connected from is checked too. Mailspike is sent the sending server's address.
Neither is sent any email address, the message's text, or a fingerprint of its content. No other blocklist is queried, and nothing is sent to rspamd.com: the lists that recognise free and disposable email providers are built into our software rather than downloaded.
Attachments are checked for viruses on our own servers as part of delivery; that scan does not send your mail anywhere.
How the platform is run
Production runs on Kubernetes with Talos Linux, an immutable operating system managed only through an API — the cluster nodes have no SSH login at all. Changes are deployed from a version-controlled configuration repository.
Servers talk to each other over a private network, and their firewalls drop all inbound traffic that is not explicitly allowed.
The site-administration interface is not reachable from the public internet — only from our private administrative network. The cluster's management interfaces accept connections only from an allowlist of administrator addresses.
Service credentials are kept in OpenBao, an open-source secrets manager, and issued to services through least-privilege Kubernetes identities — never baked into images or source code.
The mail service runs under Kubernetes' baseline Pod Security standard.
How the software is built
Every build runs secret scanning (gitleaks), static analysis (gosec), dependency vulnerability checks (govulncheck) and a filesystem vulnerability scan (Trivy). Most findings are reported for review rather than blocking the build.
A container image with a known critical vulnerability that has a fix available is not published.
Images are signed with cosign and carry a build-provenance attestation. Production does not yet verify those signatures before running an image.
Backups and deletion
The database is backed up daily, with continuous transaction logs in between, and those backups are kept for 30 days.
Stored mail is copied to off-site backup daily. We rehearsed restoring from that copy in September 2026, and the restored files matched the originals exactly.
The mail backup only adds files; it does not currently remove mail you have deleted, so copies of deleted messages can remain in it. Our privacy policy describes what is deleted when an account ends.
An expired trial account is frozen, and deleted 30 days later.
Requests from law enforcement
Kaimoku Technologies is a US company and your data is stored in the United States, so US legal process applies to it. Because mail is not end-to-end encrypted, we are technically able to produce stored messages if legally compelled. Our privacy policy says we may disclose information when required by law or legal process.
We require a search warrant before producing the content of messages. Other legal process, such as a subpoena, can reach only account information, such as account details and sign-in records.
We tell affected users about a request before we respond, unless the law or a court order forbids it.
We push back on requests that are overbroad or legally defective.
If something goes wrong
If we confirm a security breach that affects your data, we will notify the affected account owners by email within 72 hours of confirming it, saying what happened, what data was involved and what we are doing about it.
What we don't have yet
No security certification (such as SOC 2 or ISO 27001), and no external penetration test or audit.
End-to-end or zero-access encryption.
A per-mailbox log of administrator and operator access.
POP3. Mail apps use IMAP.
Certificate checks or DANE when delivering your outgoing mail.
MTA-STS in enforcing mode for our own mail domain (published in testing mode today).
DANE for our own mail domain.
DNSSEC for our own domain.
Passkey sign-in.
Reporting a vulnerability
If you believe you have found a security issue in Kuju Email or this website, tell us with the form below. We acknowledge every report within three business days and keep you informed while we fix it. Please give us a reasonable chance to fix an issue before you disclose it publicly.
We will not pursue legal action against anyone who reports a vulnerability in good faith, stays within the scope of that research, and avoids accessing other people's data or disrupting the service.