Security and data handling

Cloud fetches mail from Gmail, Microsoft Graph, or your IMAP and SMTP server when an agent asks for it. It does not store messages or attachments.

Every connection to a mail server uses TLS, and Cloud verifies the server’s certificate.

Cloud’s Google app uses Gmail’s restricted gmail.modify scope and has completed Google’s security assessment.

Your AI client may send the email it reads to its model provider, and that provider’s data policies apply.

Cloud encrypts Google and Microsoft tokens and IMAP and SMTP passwords with AES-256-GCM. Agent keys are shown once and stored only as SHA-256 hashes.

Row-level security in the database keeps each workspace’s records separate.

MCP clients sign in with OAuth. When you approve one, you choose its workspace, mailboxes, and permissions. Permanent deletion is off by default and isn’t available for Gmail. Scripts use agent keys, which have their own mailboxes and permissions.

Access tokens expire after one hour. A revoked connection or key stops working on its next request. Agents can’t connect mailboxes, create keys, change account settings, or delete your account.

Passwords need at least 12 characters, and browser sessions expire after a day. A password reset signs out other browsers and revokes OAuth connections. Our payment processor handles card details, so Cloud doesn’t receive them.

Cloud logs each mailbox operation with its agent, mailbox, and outcome, without message content or credentials. It deletes these logs after 90 days.

Cloud backs up its database every day. Backups are encrypted with a separate key and deleted after 30 days.

Deleting your account cancels your subscription and removes your personal workspace, mailbox credentials, agent keys, and connections. Deleted accounts stay deleted if we restore a backup. Mail at your email provider is left alone.

Cloud sends operation names, outcomes, timing, and error codes to our analytics provider. Events from your account or your agents include your account ID, name, and email address.

MCP connections and tool calls also record the name and version your client reports, such as Claude Code, its client type from a fixed list, and the MCP protocol version.

Events don’t include message content, subjects, search text, mailbox identifiers, or credentials.

Self-hosted Fluxmail runs on your computer, in Docker, or on your own server, and connects directly to Gmail, Microsoft Graph, or your IMAP and SMTP server. Your email never passes through a service we run.

Your AI client may send the email it reads to its model provider, and that provider’s data policies apply.

The SQLite database stays on the machine where Fluxmail runs. OAuth tokens and app secrets, IMAP and SMTP passwords, and license state are encrypted with AES-256-GCM. The key is kept outside the configuration file.

Every request must be authenticated and acts for one member. It reaches only the mailboxes that member owns or was given. Administrators manage access but can’t read another member’s private mail.

Each connection has a permission profile. read-only can search, read, and download attachments. read-write adds drafting, organizing, and moving mail to Trash. Only full can send. No profile allows permanent deletion; you add mail.delete separately.

Member sessions last up to 90 days. Logging out, suspending or removing a member, resetting a password, and revoking access all take effect immediately. Remote CLI connections require HTTPS.

HTTP MCP clients sign in with OAuth or use an API key, and scripts use API keys. Fluxmail shows each key once. A key can do only what its own settings and its owner’s current access both allow, so changing a member’s access changes their keys too.

Sign-ins and management actions go to an append-only audit log. Entries hold identifiers and outcome codes. They don’t record passwords, tokens, request bodies, or provider credentials. Administrators can read the log through the REST API.

On paid self-hosted plans, Fluxmail checks the license about once a day and sends only the license key and a random instance ID. If our license server is down, Fluxmail keeps running.

When a paid plan ends, the instance returns to Personal limits and keeps its mailboxes and data.

By default, the self-hosted package sends anonymous usage events: the command, tool, or operation name, its outcome and duration, a random installation ID, and basic runtime details. They don’t include email, mailbox data, search text, arguments, credentials, file paths, or error text.

To turn it off, run fluxmail telemetry disable or set FLUXMAIL_TELEMETRY=0 or DO_NOT_TRACK=1.