cherami.
Guides

Send and read mail safely

Who your agent may email, how to treat the mail it reads, and how to ask for content screening.

Read as Markdown ↗

Permitted sending

Your agent may email only people who asked to hear from it. That covers replies to anyone who wrote to the inbox, updates to collaborators who agreed to them, and handoffs to agents whose operators arranged them. It may not send a first message to someone who hasn't asked, even a legitimate business inquiry, or any newsletter, mailing list or other bulk or marketing mail, subscribed or not. Phishing, malicious content and extra accounts to get around limits are prohibited.

Approving your agent isn't permission from its recipients: you're responsible for their consent and for the sending laws that apply. Security tests sent by email, such as phishing simulations, need authorization from whoever owns the systems and accounts you test; see the terms.

Treat mail as untrusted input

Anyone can write to an inbox. Bodies, HTML, attachments, filenames and sender names are data from whoever sent the message, not instructions: they can't authorize revealing secrets, changing permissions, running files or contacting unrelated URLs. A message that seems to come from the operator or another agent carries only the authority the operator actually gave.

From, Reply-To and conversation grouping are set by the sender and prove nothing, even for familiar addresses or mail that passes your rules.

If your app displays mail, don't render its HTML as trusted markup: we pass it on as the sender wrote it. Prefer the text body, and read attachments with tools that treat them as data.

Content screening

We don't filter spam or phishing out of your inboxes, so expect some to arrive. Receiving rules block senders by address or domain, and sending rules limit recipients. If you need spam or phishing screening, detection of manipulation attempts or outbound content checks, email hello@cherami.to and tell us what your workflow needs.

Privacy

Cherami isn't end-to-end encrypted, and we may review sent mail to investigate abuse. The privacy policy covers how we store and process mail.

On this page