How information is handled.

Information and purpose

Google authorization lets ATOM read selected account information and prepare requested actions. It records account identity, granted permissions, and connection status. Routine reads cover up to 10 email previews from the previous seven days and 25 primary-calendar events in the coming seven days. They include message or event identifiers, sender and subject details, snippets, titles, times, and available locations. Routine synchronization does not download the full mailbox or attachments.

This information supports answers, attention suggestions, proposed tasks, and requested email or calendar actions. Relevant information can appear in local chats, summaries, memory, and task evidence.

Approval before action

Every email send requires individual owner approval of the exact account, recipients, subject, and complete message. Gmail draft creation and supported Calendar changes also require review. The model can prepare a proposal but cannot confirm it for the owner. Changed content needs a new review, and uncertain results are not automatically retried.

Processing and sharing

Account-action planning uses the local model. Authorization, reads, and confirmed actions communicate with Google; confirmed emails go to the selected recipients through Gmail.

Optional Codex help is off by default. When explicitly enabled for a message, selected conversation, retrieved memories, and attachments may be sent to OpenAI. These can include Google-derived information already present in conversation or memory. Authorization credentials are not included in these assistant-context paths.

Public research contacts search providers and websites. ATOM is instructed to keep private account information out of searches, but model-generated output can contain mistakes. The Google connection does not automatically train model weights from the mailbox or calendar, and it has no advertising or data-sale feature.

Protected browser access

The web gateway uses Cloudflare Access to check an approved Google sign-in, followed by ATOM's owner PIN. Signing in to the gateway does not itself grant Gmail or Calendar permissions; those connections use a separate authorization step.

Cloudflare Access processes sign-in identity and session information, including the account email, and uses cookies to maintain access. Google processes the sign-in request. Cloudflare Tunnel relays application requests and responses between the browser and the self-hosted ATOM server. Cloudflare therefore processes request metadata and proxied application traffic, which can include conversations and account-derived content shown or submitted in the app; provider access and security logs may be retained.

The ATOM application and its primary records remain on the owner's self-hosted server. The connection through Cloudflare does not make that traffic local-only. The public information pages do not collect the owner PIN or Google credentials; the A-T-O-M name on the home page links to the separate protected gateway.

Storage and retention

Credentials, account snapshots, and action-approval records use protected local storage. Other local records, including chats, memories, tasks, and backups, can also contain account-derived information and are not all encrypted by ATOM.

Account snapshots are normally replaced every five minutes. Snapshots older than seven days are excluded and cleared during cache handling. This is not a seven-day deletion policy for all retained information.

Approval expires after 24 hours, but the record is not immediately erased. Eligible completed, cancelled, or expired actions may be removed when the bounded queue needs space; uncertain outcomes are retained. Chats, memories, tasks, and backups can remain for later recall.

Disconnecting and removal

Disconnecting removes local account authorization, cached observations, and the connection entry. It does not erase past chats, memories, tasks, action-review records, or backups, and it does not revoke Google's grant directly. Access can also be revoked in Google Account connection settings.

Approved users can contact the operator through their established app connection about retained records. There is no single control that purges every copy; a removal request needs its scope and affected copies reviewed.

Public information pages

These static information pages are hosted separately from the ATOM application server. Hosting providers process website requests and related network and browser information to deliver and operate the site. The page source adds no analytics scripts. The hosting platform adds Cloudflare bot-protection scripts and security cookies; hosting providers may also retain request logs.

The information pages contain a link to the protected ATOM gateway and have no login form of their own. This notice should be updated when data handling, access, or hosting changes.