Data & transparency (Siftbox product)
This page describes what Siftbox reads and stores in your space, independently of Google’s consent screen.
This is not Google’s consent screen
Google asked for access to your account. This page summarises what Siftbox does with it: data read, storage and how to revoke access.
Data read from Gmail
During synchronisation, Siftbox reads Gmail only within what you authorised.
- Message identifiers, thread, date and read/unread flags.
- Short text excerpt from Gmail summarising the message.
- Subject and sender when Gmail provides them.
- List of Gmail labels attached to the message.
- Indicator that an attachment exists (without downloading the file).
- No full body or attachment binaries stored in this version.
Data stored by Siftbox
The following is stored securely in your Siftbox space.
- Synchronised information per message (snippet, subject or sender when available, sorting signals, timestamps).
- Gmail access tokens are kept encrypted to maintain the connection, never in plain text.
Model refinement (optional)
If you enable it in settings, an optional model call may include a Gmail excerpt already known to the app, truncated with no body persistence. Otherwise only structured metadata (scores, traces) is sent. You can switch back to “metadata only” anytime.
If the advanced “body text in memory only” tier is offered on your environment and you enable it, a one-off body read may be requested from Gmail for classification only, then discarded — still with no body stored in Siftbox. Also review your model provider’s terms (AI subprocessor).
Triage categories (including “To ignore”) help you organize how messages are handled. Siftbox does not delete emails at your mail provider — triage and automation stay non-destructive on the remote mailbox.
Observed confidence in classification (dashboard)
Per zone: the percentage is the share of messages tied to that zone (current category in the zone, or an explicit move from that zone to another — in-app or mailbox equivalent) for which you did not record leaving the zone. The date used is the message date (UTC), same as the chart; the correction timestamp may be earlier.
On the dashboard, a short numeric summary (number of classified messages and corrections for the chart period) appears below the trend chart, using the same date window as the curve.
This does not measure model accuracy on messages you never reviewed: no correction does not imply agreement. It is separate from the model certainty level shown on each message.
Sync: current limits
How messages are fetched in this version. These values cannot be changed from this screen.
- Up to 50 messages may be scanned per sync pass.
- No extra Gmail filter on the message list.
- After sync, new messages may be queued for automatic classification.
Detailed sync scheduling is managed by the service.
Synchronisation is usually triggered automatically about every five minutes (depending on service configuration). Siftbox may therefore import and classify messages before your mailbox filters or rules (at the provider or in your mail app) have moved or labelled them—the order can differ from what you see in your mail client.
IMAP accounts: if you create a folder from Siftbox, its exact location depends on your mail server rules.
If a folder shows in webmail but not Thunderbird, Outlook or Apple Mail, check folder subscriptions in your client settings.
IMAP accounts: when enabled for your installation, a classified message may be moved to a Siftbox folder matching the category.
IMAP accounts: you can create all Siftbox folders on the server in one go (without moving messages). IMAP-only, separate from Gmail.
Control and revocation
To connect a mailbox and manage access, create an account or sign in.