Skip to content

Agent email

When you create an agent in AI → Agents, Fontana provisions a mailbox on the same inbound domain as Email Import. Built-in email tools let the agent list, search, read, draft, and send. Inbox, outbox, drafts, and deleted appear as folders in the Agents explorer. Seeded system/ agents do not receive a mailbox.

Directly under Tools you configure three rows. They reuse the same 380px capability tiles as the rest of the card.

Row What you control
Receive Email Assigns email_listInbox, email_search, and email_read. Copy shows “Allow the agent to receive emails”. The mailbox address is copyable here. Turning this Off unassigns those tools only: inbound still stores mail, and an email_received schedule can still fire.
Send Email Assigns email_draft and email_send. Copy shows “Allow the agent to send emails”. You enter a whitelist only (one pattern per line). Empty whitelist denies every send. Every send is audited.
Messaging Assigns the existing Slack/Teams Channels MCP path. Copy shows “Allow the agent to use messaging apps like Slack or Teams”. Whitelist only (addresses, globs such as *@fontana-ai.com, and channel or user names such as #eng or @alice). Empty whitelist denies destination calls. The Channels tile is not duplicated in the Tools grid.

Whitelists persist in AGENT.md frontmatter and ingest into the derived agent row. Enforcement is in the tool executor, not UI-only.

On the agent Schedules tab you can add an email_received event (no workflow id). Delivery always writes the message to the VFS inbox. A chat turn starts only when that schedule exists and your personal activation is On. See Agent schedules.

Agent From uses the inbound mailbox domain (<tenant>.<baseDomain>). In Admin → Email you choose how those messages leave your deployment:

Mode What happens
Direct (default) Fontana’s own outbound mail server signs each message with DKIM and delivers it to the recipient’s mail server itself. No third-party mail account is required. You publish the DNS records below and your hosting account must allow outbound SMTP on port 25.
SMTP relay You save a provider’s submission host, user, and password (for example Amazon SES SMTP, Mailgun, or Postmark). Fontana stores them in Vault and submits agent mail to that provider over TLS (STARTTLS on 587, implicit TLS on 465). Clear the relay to return to direct delivery.

The Send test field on the same card sends one message from agent-mail-test@<tenant>.<baseDomain> through the active mode so you can confirm delivery and inspect the receiver’s Authentication-Results header.

The Sender DNS card in Admin → Email shows each TXT record name, the value to publish, and what public DNS currently returns:

  1. Point MX for your tenant mail hostname at the host your platform team provisions for inbound mail.
  2. Publish SPF on the mailbox domain (<tenant>.<baseDomain>) naming your deployment’s sending IP; add the provider’s include: when you use a relay.
  3. Publish the DKIM TXT at <selector>._domainkey.<baseDomain>. Fontana generates the signing key during deployment and shows the exact value in the DNS card; one key signs every tenant subdomain.
  4. Publish DMARC at _dmarc.<tenant>.<baseDomain> with relaxed alignment (adkim=r; aspf=r) so the DKIM signature on the base domain aligns with tenant subdomain From addresses.

For direct delivery, your platform team also gives the sending IP a forward hostname and matching reverse DNS (PTR); the deployment tooling provisions both for AWS-hosted environments.

Identity SMTP (Admin → Email → Identity email, account notification mail such as password reset) is a separate From path. Do not reuse those credentials for agent mailboxes.