robotomail（robotomail 搜索）

Robotomail — Give Your AI Agent a Real Inbox

Robotomail lets an AI agent send, receive, and react to email programmatically. It is API-first email infrastructure for autonomous agents, assistants, and workflows.

Use Robotomail when you want an agent to:

• send outbound email, reminders, updates, or follow-ups
• read inbound email from a real inbox
• react to incoming email via webhooks or real-time SSE events
• reply in existing threads using normal email semantics
• use a hosted robotomail.co address or a custom domain

Best-fit use cases

1. Agent support inbox

Give your agent a real inbox so it can classify, summarize, draft, or reply to incoming support emails.

2. Outbound follow-ups and reminders

Let an agent send reminders, nudges, receipts, booking confirmations, or outbound follow-ups from its own mailbox.

3. Real-time email triggers

Subscribe to inbound email events with webhooks or SSE so your agent can wake up and act immediately.

Fast start

If the user is new to Robotomail, the fastest path is:

1. Sign up at https://robotomail.com or create an account via POST /v1/signup
2. Create or list a mailbox with GET /v1/mailboxes or POST /v1/mailboxes
3. Send a test email with POST /v1/mailboxes/{id}/messages
4. Read inbound email with GET /v1/mailboxes/{id}/messages?direction=INBOUND
5. Subscribe to inbound events via webhook or SSE

Authentication

All requests require an API key in the Authorization header:

Authorization: Bearer <ROBOTOMAIL_API_KEY>

Base URL: https://api.robotomail.com

If the user doesn't have an API key yet, they can sign up at https://robotomail.com or you can create an account via POST /v1/signup.

Quick Reference — What Can I Do?

For full endpoint details including request/response schemas, read references/api-reference.md.

Webhooks vs SSE

Use webhooks when Robotomail should push events to your public HTTPS endpoint.

Use SSE when the agent wants a live event stream over a single long-lived connection, for example when running a local listener or agent loop.

If the user says things like "stream inbound mail", "listen for new messages", or "real-time inbox events", prefer SSE.

Decision Tree — Common Tasks

"Send an email"

1. List mailboxes: GET /v1/mailboxes — find the mailbox to send from
2. Send: POST /v1/mailboxes/{id}/messages with to, subject, bodyText (and optionally bodyHtml)
3. To reply to an existing message, include inReplyTo with the original message's messageId header value

"Check my inbox" / "Read my email"

1. List mailboxes: GET /v1/mailboxes
2. Fetch inbound messages: GET /v1/mailboxes/{id}/messages?direction=INBOUND
3. Read a specific message: GET /v1/mailboxes/{id}/messages/{msgId}

"Reply to an email"

1. Read the original message to get its messageId field
2. Send reply: POST /v1/mailboxes/{id}/messages with inReplyTo set to the original's messageId value
3. Threading is automatic — the reply joins the same thread

"Set up email for a new domain"

1. Add domain: POST /v1/domains with {"domain": "example.com"}
2. Read the dnsRecords from the response — tell the user to configure these at their DNS provider
3. After DNS is configured, verify: POST /v1/domains/{id}/verify
4. Once verified, create a mailbox: POST /v1/mailboxes with {"address": "agent", "domainId": "<id>"}

"Set up a webhook for incoming email"

1. Create webhook: POST /v1/webhooks with url and events: ["message.received"]
2. Save the secret from the response, it is shown only once
3. Verify deliveries with X-Robotomail-Signature header (HMAC-SHA256 of payload using secret)

See references/webhook-verification.md for signature verification code.

"Listen for incoming email in real time" / "Use SSE"

1. Connect to GET /v1/events
2. Filter or react to message.received, message.delivered, message.bounced, and message.complaint
3. Use SSE when the agent wants a live stream instead of public webhook delivery

"Send an email with an attachment"

1. Upload the file: POST /v1/attachments (multipart/form-data, field name file, max 25MB)
2. Note the returned attachment id
3. Send the message with attachments: ["<attachment-id>"]

"Read an inbound email's attachments" / "Handle inline images"

Inbound messages with attachments expose them on two surfaces with different shapes. Pick the right one for your access pattern — do not assume REST responses contain ready-to-use download URLs.

1. Webhook / SSE message.received payload — data.attachments[] is delivered with a ready-to-use download_url field (presigned R2 URL, valid 24h from publish time) on each attachment, alongside id, filename, content_type, size_bytes, and content_id. Field names are snake_case. Fetch each file directly — no Authorization header needed:

   for att in event.data.attachments:
       bytes = HTTP_GET(att.download_url)
       save_to(att.filename, bytes)

1. REST GET /v1/mailboxes/{id}/messages and GET /v1/mailboxes/{id}/messages/{msgId} — attachments[] contains metadata only (id, filename, contentType, sizeBytes, contentId), no URL field. To download, call GET /v1/attachments/{id} for each attachment id and use the url field on that response. Field names are camelCase. This is also how you refresh an expired download_url from an old webhook/SSE replay.

Inline images: when an attachment's content_id (snake_case in webhook/SSE) or contentId (camelCase in REST) is non-null, the attachment is referenced in body_html / bodyHtml as <img src="cid:<content_id>">. Rewrite each cid: reference to a real downloadable URL before passing the HTML to a renderer or vision model. From a webhook/SSE payload (where download_url is already on the attachment):

for att in event.data.attachments where att.content_id is not None:
    body_html = body_html.replace(f"cid:{att.content_id}", att.download_url)

From a REST message read, fetch a presigned URL per inline attachment first:

for att in message.attachments where att.contentId is not None:
    url = GET(f"/v1/attachments/{att.id}").url
    bodyHtml = bodyHtml.replace(f"cid:{att.contentId}", url)

Drop limits: if a message has more than 20 attachments, or any single attachment exceeds 25 MB, the payload includes attachments_dropped: true and attachments_dropped_reason: "size" | "count" | "both". The message itself is still delivered, but the over-limit attachments are not stored. Tell the user if you see this flag.

Key Constraints

• Daily send limits: 100/day (free), 500–2,000/day (paid, varies by tier) per mailbox — resets at midnight UTC
• Monthly send limits: 5,000/month (free), 15,000/month (Developer), unlimited (Growth, Scale) per mailbox
• Velocity limits: 30 messages/min per mailbox, 60 messages/min per account — returns 429 if exceeded
• Bounce rate: Must stay below 3% over a 7-day rolling window (minimum 50 messages). Exceeding this auto-suspends the mailbox.
• Complaint rate: Must stay below 0.05% over a 7-day rolling window (minimum 50 messages). Exceeding this auto-suspends the mailbox.
• Attachment size: Max 25MB per file
• Storage: 1GB (free), 10GB (Developer), 25GB (Growth), 100GB (Scale)
• Free tier: 3 mailboxes on robotomail.co only
• Paid plans start at $19/mo (Developer): 10+ mailboxes, custom domains

If a send returns 429, the mailbox has hit its daily/monthly limit or velocity limit. Tell the user and suggest slowing down or upgrading if on the free plan.

Error Handling

All errors return {"error": "message"} with standard HTTP status codes:

• 401 — Missing or invalid API key
• 403 — Account suspended or scoped key accessing a restricted resource. When suspended, response includes { "suspended": true, "reason": "bounce_rate_exceeded" }. Tell the user to contact support@robotomail.com.
• 404 — Resource not found
• 429 — Rate limit, velocity limit, or send quota exceeded
• 413 — Attachment too large

Tips

• Always use bodyText for the plain-text version. Add bodyHtml only when rich formatting is needed.
• Use threads (GET /v1/mailboxes/{id}/threads/{tid}) to see full conversation history before replying.
• When listing messages, use since parameter (ISO-8601) to fetch only recent messages.
• Suppression list (GET /v1/suppressions) contains bounced/complained addresses — check before sending to addresses that previously failed.