Privacy policy
Read this first — it changes what the rest of the page means
SkimMail is self-hosted software, not a service. LyQuyDuong publishes a program; it does not operate an installation of it for anybody, and it has no way to reach the installations other people run. The person or organisation that runs an instance of SkimMail is the data controller for the mail, credentials and settings inside it. LyQuyDuong is not a controller, not a processor, and not a recipient of that data.
1. Who handles what
Three different parties appear in any sentence about SkimMail and personal data, and confusing them is the commonest mistake.
| Party | Role | What they hold |
|---|---|---|
| The operator | Data controller. The person or organisation that installed SkimMail on a server they control. | Everything: the mailbox credentials or OAuth tokens, the cached mail, the settings, the logs. All of it on their own infrastructure. Also the OAuth application itself, registered in their own Google or Microsoft account — see below. |
| The mail provider | Independent controller — Google, Microsoft, or whoever runs the IMAP server. | The mailbox itself. SkimMail reads from it; it does not replace it. That provider's own privacy policy continues to govern the mailbox. |
| LyQuyDuong | Publisher of the software. Not a controller and not a processor of anything in an instance. | Nothing from any instance. The only personal data it can hold is described in section 5, and it only exists if you email the publisher. |
LyQuyDuong supplies software. It does not run a service.
There is no SkimMail service to sign up for, no account with LyQuyDuong, and no server of ours that your instance talks to about your mail. Nothing from any instance is ever sent to us — not a message, not a header, not an address, not a credential, not an OAuth token. There is no code path that does it, no telemetry, no analytics and no crash reporting. We could not read your mail if we wanted to, because it never arrives anywhere we can reach.
Two narrow exceptions exist and we would rather name them than claim a clean zero: if you email us, we have your email; if you apply for a licence key, we have what you sent with the application. Both are described in section 5. Neither comes from your instance, and neither happens unless you start it.
The OAuth application is the operator's, not ours
SkimMail ships no shared OAuth client. If you connect a Gmail or Outlook mailbox, you register your own application in your own Google Cloud project or Azure tenant, and you paste its client ID and secret into your own instance. The sign-in then happens directly between your server and the provider. LyQuyDuong is not a party to it and cannot see it.
That has a consequence worth stating plainly, because it surprises people: in Google's and Microsoft's eyes, the developer of that application is you. Its consent screen, its authorised domains, its verification status, its user cap and its compliance with the provider's developer policies — including Google's Limited Use requirements — are yours to satisfy. Google's Branding form will ask that application for a home page, a privacy policy and a terms of service; those are your pages, on a domain you control and verify.
This page is not the one you submit. It describes how the software behaves and what LyQuyDuong does and does not hold, so that you can write your own with accurate facts in it. Adapt it; do not copy it. Two things it cannot know about your instance are whether you have configured a notification channel and whether you have configured an AI provider — both send data somewhere, and both belong in your policy if you use them. There is a step-by-step walkthrough in Register the OAuth application.
If you are an end user whose mailbox has been connected to somebody else's SkimMail instance — an instance run by your employer, for example — the questions that matter to you are for that operator. LyQuyDuong cannot see, export or delete anything on their server, and cannot act on a request about it.
2. Google user data, and the Limited Use commitment
This section exists because SkimMail can sign in to Gmail using Google OAuth 2.0,
and because that path requires the restricted scope
https://mail.google.com/.
Whose application is it
SkimMail ships no OAuth client of its own. Each operator registers their own application in their own Google Cloud project and configures their instance with its client ID and secret. The authorisation code exchange and every later token refresh run directly between that operator's server and Google. No code, token or message body passes through any server operated by LyQuyDuong, because no such server exists.
Which scopes are requested, and why
| Scope | What it is used for |
|---|---|
https://mail.google.com/ |
Connecting to imap.gmail.com over IMAP with XOAUTH2 in order to list folders, fetch message headers and bodies, and perform the actions you take in the interface: setting read/unread and starred flags, archiving, and moving a message between folders. "Delete" moves the message to the account's Trash folder; SkimMail never permanently deletes a message. On an IMAP server that does not implement the MOVE extension the move is carried out as a copy followed by expunging the source copy — the message still exists, in Trash. It never sends mail: there is no compose window and no SMTP connection. |
userinfo.email |
Reading the address of the mailbox that was just authorised, so the instance can label the account and match a later re-authentication to the right one instead of creating a duplicate. |
openid |
Standard OpenID Connect sign-in, used alongside the address above. |
Why a restricted scope, when the consent screen wording is so broad.
SkimMail's sync engine is an IMAP client. Google grants IMAP, POP and SMTP access
through https://mail.google.com/ and through no other scope. The
narrower-sounding Gmail scopes grant the Gmail REST API instead — a token issued
for gmail.modify is valid, and imap.gmail.com rejects it.
There is therefore no narrower scope that would let SkimMail work at all. The
breadth of the grant is Google's classification of what IMAP access implies; it is
not a description of what SkimMail does with it.
Limited Use
SkimMail's use of information received from Google APIs will adhere to Google API Services User Data Policy, including the Limited Use requirements.
Concretely:
- Google user data is used only to provide the mail-reading features the operator can see in the interface — the unified inbox, threading, search, tags and rules — on the operator's own server.
- It is never transferred to LyQuyDuong. The software contains no code path that sends message content, headers, addresses or tokens to the publisher. There is no telemetry, no analytics and no crash reporting.
- Where the operator switches on a feature that involves a third party, that transfer happens at the operator's direction, to a destination the operator chose. There are two: notification channels, which carry a message's sender and subject unless the operator turns content off, and the AI features in section 6, which send message bodies to a provider the operator configures. LyQuyDuong selects neither and receives nothing from either.
- It is never used for advertising, ad targeting, profiling, credit assessment or lending, and it is never sold.
- LyQuyDuong does not use it to train any machine-learning model, generalised or otherwise, because it never reaches LyQuyDuong. Where the operator enables an AI feature, the chosen provider's own terms govern retention, human review and training, and reading them is the operator's responsibility.
- The publisher never reads it, since the data never arrives anywhere they can reach. On the operator's own instance it is read by the people the operator has given accounts.
- Mail is never sent. SkimMail is a read-first reader. It has no compose window and does not connect to SMTP.
Deleting a connected account from an instance removes SkimMail's local copy of that mailbox, including cached message bodies. Revoking the grant from the Google account's own security settings ends SkimMail's access immediately; the next sync fails and the account is flagged as needing reconnection.
3. What the software stores on your server
Everything below lives in the data directory the operator configured, on the operator's machine.
| Data | How it is protected at rest |
|---|---|
| Mailbox passwords and app passwords, OAuth access and refresh tokens, two-factor secrets, proxy URLs and object-storage secrets | Encrypted with AES-256-GCM under a 32-byte master key held by the instance. The master key is either a file readable only by the service account, or — recommended — stored wrapped, decryptable only with an unlock secret kept outside the data directory. The master key never leaves the instance during normal operation. The one exception is a backup: a backup archive embeds the instance's master key, because without it the credentials inside could never be restored. That makes a backup file as sensitive as the data directory itself — cloud uploads require a password, and an unencrypted local download is exactly as exposed as wherever it is kept. |
| Message headers, subjects and addresses | Stored in the instance's database so listing and searching cost nothing. These are not encrypted by the application. Protecting them is a matter of encrypting the volume they sit on, and the project's documentation says so rather than implying otherwise. |
| Message bodies you opened | Cached in the instance's blob store, and encrypted at rest by default with AES-256-GCM under a key derived from the instance's master key. That is a separate switch from the credential encryption above and it is on unless an operator turns it off. Retention is the operator's: there is a size ceiling, a body-retention period in days and a manual purge, all in the instance's own settings. |
| Sign-in sessions, an append-only audit log, and application logs | Kept locally. Secrets are not written to logs, and the audit log records non-secret detail only. |
4. What the software sends over the network
A privacy policy that could not be checked would be worth little, so this is the complete list of outbound connections the software makes, and they are all made by the operator's instance rather than by anything the publisher runs.
- To your mail servers — IMAP to each configured account, to sync mail. This is the point of the program.
- To Google or Microsoft — only if OAuth is configured: the authorisation code exchange and later token refreshes, using the operator's own client credentials.
- To an image host — only when you explicitly ask to load the remote images in a message. The server fetches them, through that account's own egress, rather than letting your browser do it, so the sender's host sees the instance and not you.
- To a sender's unsubscribe endpoint — only when you click unsubscribe, and through a client that refuses loopback, private and cloud-metadata addresses.
- To a notification channel you configured — a webhook, Telegram, Slack, Discord, Microsoft Teams, Google Chat, or a Web Push endpoint, carrying whatever the operator's content-privacy setting allows.
- To a backup destination you configured — a local folder or an S3-compatible bucket you named, carrying an encrypted snapshot.
- To GitHub's release API — the optional update check, which sends no account or usage data. It is off until the operator turns it on.
- To two static files LyQuyDuong publishes on GitHub Pages — the signed plugin index and the signed list of revoked licence keys. Both are ordinary downloads of a public file, fetched on the same schedule as the update check and therefore only when that is on.
- To a plugin's vendor — only if that plugin is installed; the tunnel helpers and the WireGuard engine are downloaded on demand against a signed, SHA-256-pinned index.
- To an IP-echo service (
api.ipify.orgby default) — only when someone presses the button that asks an egress which public address it leaves from. Never in the background. - To an AI provider the operator configured — see section 6.
Nothing in that list sends data about the instance to LyQuyDuong. Three entries fetch a public file LyQuyDuong publishes on GitHub Pages — the update check, the plugin index and the revocation list — and, as with any HTTP request to any host, GitHub sees the ordinary request metadata. Licence keys themselves are verified offline against a signature; the revocation list is a signed file the instance pulls, never a key it sends.
5. The only data the publisher ever receives
LyQuyDuong receives personal data in exactly two situations, both of which begin with you contacting it.
- You email skimmail@base101.app — for support, a security report, a commercial enquiry or anything else. It then holds your address and whatever you chose to put in the message, for as long as needed to deal with it.
- You request or receive a Sponsor or Pro licence key — it then holds your email address (to issue and, if needed, re-send the key), optionally an organisation name shown on a Pro licence, and for Sponsor tier whatever identifier the sponsorship channel uses, solely to confirm backer status.
That data is used only to answer you and to issue and support your licence. It is not sold and is not shared with third parties for marketing. It is never joined to anything from your instance, because nothing from your instance is ever received.
This website is three static pages served from GitHub Pages. It sets no cookies, loads no fonts or scripts from anywhere else, and carries no analytics or tracking of any kind. As with any web host, GitHub processes the ordinary request metadata involved in serving a page.
6. AI features — built, and off until you configure them
SkimMail contains optional AI features that send message content to an inference service the operator chooses and configures. They are implemented and present in every release, including the current one. This section describes what is there rather than what is planned, because an earlier version of this page said the feature was not built, and that was wrong.
Nothing is sent anywhere until a provider is configured. With no provider and no API key set, every one of these endpoints returns an error and no message content leaves the server. A default installation sends nothing.
- Six endpoints exist — summarise a message, classify a message, translate a message, extract action items from a message, summarise a thread, and produce an inbox digest. Each reads the message body and sends it to the configured provider; the digest sends several messages at once rather than one.
- One path runs on its own, with nobody pressing anything. After each sync, newly arrived inbox messages can be classified in the background — up to twenty message bodies per sync, a cap that bounds cost rather than disclosure. It sits behind two separate switches: AI must be enabled at all, and auto-classification must be enabled specifically. With either off nothing runs, which is why a default installation still sends nothing. It is called out separately because every other path above is something a person deliberately triggered.
- The provider is the operator's choice — Anthropic, OpenAI, Google Gemini, or Ollama, which can run on the operator's own hardware and send nothing to anyone.
- There is no licence check on them. Any tier can reach them once a provider is configured. An earlier version of this page said they were Pro-only; that was also wrong.
- The buttons are hidden in this release's interface, but hiding a button does not remove an endpoint — anything holding a session on the instance can call them.
- The chosen provider's own privacy policy, retention period and human-review practices govern what happens to the content after it arrives. LyQuyDuong has no relationship with that provider and cannot say what it keeps.
An operator who wants the capability gone entirely simply does not configure a provider. An operator who wants it without sending anything off their machine can use Ollama.
7. Your rights, and who to ask
Under the GDPR, Vietnam's Law on Personal Data Protection (Law No. 91/2025/QH15, in force since 1 January 2026) and comparable regimes, rights of access, rectification, erasure, restriction, portability and objection are exercised against the controller. For anything inside a SkimMail instance, that is the operator of that instance — the person or organisation who installed it. They hold the data, and they are the only party technically able to act on such a request.
For the narrow set of data in section 5, where LyQuyDuong is the controller, write to skimmail@base101.app and it will be actioned.
If you operate an instance and other people's mail is in it, you are the controller for that mail and the obligations are yours, including telling those people and honouring their requests. The project's security documentation sets out the controls available to you and, equally, the threat model they do not cover.
8. Changes to this policy
This page may be updated. Material changes to data handling — AI Triage above being the anticipated one — will be published here with a new "last updated" date before the change takes effect, and notified by email to anyone holding a licence key issued with contact details.
Status of this document
This policy is written in good faith by the people who wrote the software, and each statement in it was checked against the source before publication. It has not been reviewed by qualified legal counsel, and it will be before any material change in data handling. SkimMail's source is not currently published, so the claims here cannot be independently verified by reading it — they rest on this statement. Operators with compliance obligations of their own should have their use of SkimMail reviewed in their own jurisdiction.