How to forward email
into an AI agent.
You want the agent to process mail that was sent to your real inbox, without handing it OAuth access. Forwarding rules are the answer.
You have a real email address — personal, work, or a shared team account — and you want an AI agent to act on some of the messages that land there. The wrong way is to give the agent OAuth access to your inbox. The right way is to forward only the messages you want processed, into the agent's own inbox.
This keeps the agent operating on its own identity while still letting it act on real-world mail. Sender sees your agent's address when they reply; you decide exactly which messages the agent gets.
1. Give the agent its own address
Provision a Loomal identity — this becomes the agent's mailbox. You'll get something like agent-x8k2m@loomal.ai (or a custom-domain address on paid plans). Forwarding will target this address.
The agent processes mail that lands here exactly as if the sender had mailed directly. Replies from the agent go back in the agent's name — your original inbox is not in the loop.
2. Set up a Gmail filter
In Gmail, open Settings → Filters and Blocked Addresses → Create a new filter. Set the criteria (From, Subject contains, etc.) to match only the mail you want the agent to see, then check 'Forward it to' and point it at the agent's Loomal address.
Gmail requires you to verify the forward destination first — Google sends a confirmation code; forward that single code into Loomal so the agent can read it, then paste it into Gmail's verification prompt.
From: support@vendor.com
Action: Forward to agent-x8k2m@loomal.ai
Skip the Inbox (optional — keeps the original out of your way)
Apply label 'agent-handled'3. Set up an Outlook rule
In Outlook (Microsoft 365), go to Settings → Mail → Rules → Add new rule. Choose a condition (Sender, Subject, etc.), then 'Forward to' with the agent's Loomal address. Outlook doesn't require verification but does warn if the destination is outside your organization.
For corporate accounts, your admin may block external forwarding — check policy before relying on this pattern in production.
4. Respond from the agent's identity
When the agent replies, it replies as the agent. The original sender sees replies from agent-x8k2m@loomal.ai (or your custom domain), not from you. This is usually what you want — attribution is clear, and the conversation stays with the agent even if you change jobs or lose access to the original inbox.
If you genuinely want replies to appear from your personal address, that's the case where Gmail API or Nylas is the correct tool instead — you need to impersonate the user, which forwarding doesn't allow.
5. Scope what gets forwarded
Keep the filter narrow. Sender matches, subject matches, label matches — whatever combination isolates the messages you actually want the agent to process. A too-broad filter means the agent sees mail it shouldn't, which defeats the privacy gain of keeping OAuth out of the picture.
FAQ
What about inbound-only OAuth, like Gmail's watch() API?
If you need bidirectional access to your own inbox (read and write), watch()/Gmail API is the right tool. Forwarding is for the case where you only want the agent to see specific messages and act from its own identity.
Can I forward from multiple inboxes to one agent?
Yes. The agent's inbox accepts mail from any sender, including multiple forwarding sources. Use labels (applied by Loomal's inbound rules or inferred by the agent) to track provenance.
Does Gmail preserve the original sender's address?
Yes — when Gmail forwards, the original From header is preserved in the forwarded message. The agent sees the real sender; replies go to that sender, not to Gmail.
Related reading
Last updated: 2026-04-15