The email connector gives your flespi AI agent an email account of its own. The agent uses it to reach you and your team by mail: an alert when something needs attention, a report on a schedule. It also runs correspondence on your behalf: for example, it takes a case to a device manufacturer's support, follows the answers, and tells you the result. With a full account, it also watches the inbox and wakes on incoming mail.
New to flespi AI agents? Read the flespi AI agents article first: what agents are, how they reason and sleep, and how approvals keep you in control. This article covers the email side of the story.
1. Connect the account
Connect the account in one of two variants:
- imap: a full IMAP/SMTP account where the agent sends mail, reads the mailbox, and reacts to incoming mail
- smtp: a send-only SMTP account where mail leaves this address and nothing is received or read
The smtp variant is enough for alerts and reports. Choose imap when answers should come back to the agent, as in a support case.
Give the agent an address of its own rather than your personal mailbox. Every message that arrives in the inbox wakes the agent and takes a reasoning step.
Configuration fields:
- user: the email address of the account, the From address of sent mail and, unless a login is set, the login username
- login: optional, the username when the provider's login is not the address: an Amazon SES SMTP username, SendGrid's
apikey, a Postmark server token - password: an app password for providers with two-factor auth (e.g. Gmail), or the SMTP password or API key paired with the login
- imap and smtp: the IMAP and SMTP servers as host or host:port (e.g.
imap.gmail.com,smtp.gmail.com), the IMAP server only in the imap variant - name: optional, the display name shown next to the From address
Connections always use TLS. Providers that block password logins for third-party apps require an app password: without one the connection fails. Transactional mail services such as Amazon SES or SendGrid work in the smtp variant: their SMTP endpoint as the server, their SMTP credential as the login and password, and a sender address verified with the service as the user.
Create a connector of the email type and fill these in:
Assign the connector to your agent (create one if you have none yet):
One connector can serve several agents: with the send-only smtp variant they all gain mail delivery from the same address. An imap connector can be shared the same way, but every incoming mail wakes all of its agents, so an inbox is best watched by a single agent.
2. Work with the agent
Tell the agent what you need in plain words. A few examples:
- "Email me when a device has been silent for a day."
- "Send the weekly mileage report to the service team every Monday."
- "Take this case to the manufacturer's support and follow it until they answer."
If you find yourself copying a manufacturer's answers into the chat, hand the thread over: the agent writes to them directly and reports back to you.
The agent composes and sends plain-text mail: up to 32 recipients across To/Cc/Bcc and up to 10 attachments (10 MiB each, 20 MiB per message) taken from its file storage. Replies thread properly: the agent references the answered message and quotes the original beneath its reply. The From address is always the connected account, only the display name decorates it. The agent cannot send as anyone else.
With the smtp variant the agent learns only that the mail server accepted the message. A bounce goes to the mailbox, which the agent does not read.
On an imap account the agent also reads the mailbox on demand: searches folders, reads past correspondence, walks long threads, and fetches attachments, useful for history that predates the connector.
With the imap variant, new mail in the inbox wakes the agent: the message text and attachments are delivered to it, and a reply can follow in the same thread. Only mail arriving after the connector is attached produces events, and only the inbox is watched: server-side filters that move mail to other folders keep it out of the agent's sight (it can still be asked to read those folders). Obvious automated mail (delivery reports, out-of-office autoreplies, no-reply senders) is ignored: no reply, no action.
Incoming mail is treated as untrusted: the sender is judged by the From address alone, and a mail instructing the agent to change its configuration or reveal secrets is reported to you instead of obeyed.
Sending is controlled by the agent approval rules through the email_send operation type. By default it executes automatically. Switch the default to asking for approval to turn every outgoing email into an approval request: you review the recipients, subject, and body before approving. A rule on a recipient mask such as *@yourcompany.com keeps mail to your own domain automatic while everything else is reviewed, and a Reject always rule blocks mail to specific addresses. A message whose recipients fall under different rules asks. The rules button next to the connector in the Connectors block of AgentBox opens these rules:
Reading the mailbox is never gated.
Mail traffic lands in the agent conversation history as email_received and email_sent records, so the whole correspondence reads right in the agent chat:
Every mailbox operation the agent performs, reads and sends alike, is recorded in the connector logs with the operation, target, and result.
The email connector is one of several an agent can hold: browse the full connector list to give your agent more capabilities.
Subscribe to the flespi AI and MCP change log to stay in sync with any updates.



