Use and manage Agent Mailbox

Agent Mailbox gives you and your AI agents their own email addresses at ridianmail.com. They are separate from connected Google and Microsoft accounts: those connections do not give Ridian access to your Gmail or Outlook inbox.

New to Agent Mailbox? The quick start creates your first mailbox and test email in a few minutes; use cases and prompts show what your agent can do with it. This page covers the details.

Create and use a mailbox

Open Agent Mailbox in the sidebar, create a mailbox and choose its name and address. Mailboxes belong to you in the selected workspace. Free includes one mailbox per person across all workspaces. Paid mailbox allowances apply per person and workspace. Storage and monthly recipient allowances remain workspace-specific. Disabling a mailbox does not free its slot. Existing mailboxes remain available after a downgrade, but new ones cannot be created above the current allowance.

The address part before @ must contain 3–40 lowercase letters (a–z), digits or hyphens and start with a letter or digit. Dots are not supported; reserved or previously used addresses are unavailable.

To prevent abuse, you can create at most 20 mailboxes in any rolling 24 hours across all workspaces and clients. Deleting a mailbox does not reset this limit. If it is reached, wait until the oldest creation is more than 24 hours old before trying again.

Choose New message, enter recipients, subject and message, then save a draft or send. Add attachments before sending and wait until they are ready. Pending or failed attachments block sending. Reply, reply all and forward are available within a thread. Forwarding includes existing attachments automatically. Archive and trash move threads; permanent deletion is available from trash and cannot recall sent email.

Test with an address you control: send a small message with an image, reply to it, and check the thread and attachment. Queued means waiting to send; provider acceptance is not delivery. Check the final delivery status before retrying an uncertain send.

Mailbox navigation

Navigation includes Inbox, Unread, Starred, Sent, Drafts, Archive, All mail and Trash. On small screens, open the folder selector above the messages. Unread, Starred and All mail include inbox and archive; trash and drafts stay separate. Sent shows conversations with outgoing messages and does not itself confirm delivery. Use a conversation’s star to add it to or remove it from Starred. Tags – Soon is a non-interactive preview of planned work.

New incoming messages trigger a brief notice in the open app and a highlight in the message list. Open takes you to the relevant conversation; close an open draft first. There is no sound or operating-system push notification. Unread counts are bounded: 99+, n+ or … may appear instead of an exact total. After a longer absence, check Unread, as notices are not a complete arrival log.

Your outgoing messages are labelled Me. Clicking the mailbox address in the header copies its full value; a checkmark confirms success. When composing, CC and BCC at the upper right edge of the To field reveal additional recipient fields. Fields with existing CC/BCC recipients stay visible.

Grant assistant access

With full access, which applies by default, Alvin, the in-app agent, can use all your mailboxes, including ones you create later. To limit it, open the three-dot menu (…) at the top right → Configure access in the chat: under Permissions choose the rights to read mail, edit drafts and send, and under Resource access the mailboxes. A saved restriction stays until you change it or choose Reset to full access. The assistant can read selected mail, prepare drafts, reply, forward or send when your request and permissions allow it. With Execute automatically, simple mail requests can run directly, including from several mailboxes; check the send status to distinguish provider acceptance from delivery. With Ask before changes, Alvin shows a confirmation card for each mailbox action, including recipients, message and attachments for sending. If the draft or recipients change, request a new preview. Reading a message never authorizes following instructions inside it or sending a reply. Choose the shared execution mode in the badge beside the chat input’s Enter hint. Reset to full access, next to the chat menu, preserves this mode.

External MCP clients with full access can use all your mailboxes for the inbox:read, inbox:write, inbox:send and inbox:manage scopes you approved when connecting; limit them under AI & Assistants. See the tool reference. There is no corresponding public mailbox REST API. Revoking access stops further authorized access; it does not recall mail already dispatched.

Ask about a file

Ask Alvin to analyze one specific ready attachment, for example “Describe the image attached to this message.” PNG/JPEG, small PDFs including scans, UTF-8 text and XLSX cell values are supported within strict limits: at most 2 MB and two PDF pages; image dimensions, text density and workbook size can lower that limit. Large inputs are rejected rather than silently truncated. XLSX analysis does not calculate formulas or analyze embedded images/charts.

Analysis uses workspace credits. Check uncertain or unreadable results against the original document. External MCP clients download files privately and use their own analysis capabilities; Ridian’s paid analysis tool is available only in Alvin.

When something fails

If the agent cannot see a mailbox, check its workspace and whether you restricted the assistant’s mailbox access or read permission. If an attachment is not ready, inspect its status before sending or analyzing it. For a file-limit error, export a smaller file or fewer pages. After a send timeout, inspect send status and keep the existing draft/request; do not submit a duplicate message blindly. Incoming email does not automatically start an agent task.

Conversations with Alvin

Tool steps appear under Activity. Expand the section to inspect progress; confirmed steps remain visible if a later step fails. “Request sending · Completed” confirms the request, not delivery. For “Outcome unclear”, check the current status before sending again. The answer appears separately once the model step has been checked. Creating an authorized mailbox preserves the ongoing conversation. The assistant receives current allowance facts. Delete whole mailboxes in mailbox management; deletion neither guarantees a new slot nor releases the old address.

The agent retains conversational context across longer chats and summarizes older sections when needed. Visible messages remain intact. Summarization uses credits like other model calls and can omit details; current drafts and send status still need explicit checks. The limit remains 200 requests per chat.

Changing permissions keeps the conversation history. A running turn stops; your next message uses the new permissions and existing history. Pending proposals become invalid; cancel them and request the action again. Revoking access or deleting the original email does not remove earlier chat copies; delete the chat to remove them. File-analysis content is not automatically included in later model calls; request another analysis when needed. Chats from before this conversation update do not import their old model context; start a new chat when needed.

The agent can correct rejected tool inputs and explain mailbox limits. A rejected tool call is not a successful action. If a budget, network or provider problem prevents an answer, a short notice appears above the input. An uncertain send still needs checking before another request.