Safe by design, not by setting.
JamComms will never send, delete or move your email without you, and never show a clickable link in a message it flags as phishing. These rules are built into the product rather than offered as options, so no setting, rule or email can switch them off.
A phishing email reads as plain text.
A message JamComms flags as phishing is shown as plain text. Links are rewritten (hxxps://, [.]) so they cannot open, and there are no forms, remote images or "open original" escape hatch. If JamComms cannot verify a message is safe, it treats it as unverified. Flagged mail is never filed and never acted on by rules or Claude.
Sent when you press Send. Deleted only after you confirm.
JamComms sends a message only when you press Send; Claude drafts replies, follow-ups and delegations but never sends them. A delete happens only after you confirm it, and only to Trash, so a message can be recovered. Rules and filing only move mail between folders or labels, and every change is written back to your real mailbox.
The assistant changes nothing until you confirm.
Claude reads and searches on its own, but every change comes back as a pending action that runs only when you press Confirm. Email content is treated as untrusted information, never as instructions, so a crafted message cannot steer the assistant into moving or filing anything.
Microsoft and Google give JamComms a token, not your password.
JamComms connects through the providers’ own sign-in where they offer it: Microsoft for Outlook and Teams, and Google for Gmail. You can revoke its access from your Microsoft or Google account at any time. For other IMAP mailboxes you give an app password. Every token and password is stored encrypted, and never shown or logged.
Report a security issue
Email hello@jamcomms.io with what you found and how to reproduce it. Please give us a chance to fix it before you share it publicly.
Security questions
How will JamComms protect me from phishing?
A message flagged as phishing will never show a clickable link, a form or a remote image. Links are rewritten so they cannot open. If JamComms cannot verify a message is safe, it treats it as unverified. Phishing mail is never filed and never acted on by rules or Claude.
Will JamComms send or delete email on its own?
No. JamComms will send a message only when you press Send. Claude drafts replies, follow-ups and delegations, but never sends them for you. Deleting happens only after you confirm, and only to Trash, so you can get a message back. Rules and filing only move mail between folders or labels.
Could an email trick the assistant into doing something?
JamComms will treat every email as untrusted content, never as instructions. The assistant can read and search on its own, but any change to your mail comes back as a pending action that runs only when you press Confirm, so a crafted message cannot move or file anything by itself.
JamComms is in early access. Join the waitlist and you will hear the launch date and prices first.