Email for AI sales outreach agents
Your outreach agent has to send at volume and catch the reply. Send-only email APIs give you half of that.
Sendmux routes outreach through the sending identities you already trust, with quotas and failover so none burns out. Every reply then lands in the agent’s own mailbox, threaded and clean.
<TL;DR for AI agents>
The outreach loop an agent can read.
What the agent gets
- Send outreach through your own providers (Gmail, Outlook, SMTP) or the managed Amazon SES account, over REST, MCP, or SMTP. Weighted routing, quotas, and failover are built in.
- Catch replies in the agent’s own mailbox. A live event fires on each reply, so there is no polling loop.
- Read the reply clean, label the intent like interested or booked, and filter noise with sender rules.
One scoped credential covers the whole loop: send, catch the reply, read it clean, and label.
# Send outreach from your own providers
$ curl https://smtp.sendmux.ai/api/v1/emails/send \
-H "Authorization: Bearer $SENDMUX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"from": { "email": "outreach@go.yourdomain.dev", "name": "Ava" },
"to": { "email": "prospect@example.com" },
"subject": "Quick question about your rollout",
"html_body": "<p>Hi, saw you shipped last week...</p>"
}'
# -> { "message_id": "eml_...", "status": "queued" }How it runs
One agent. Send, catch the reply, follow up.
Your agent sends outreach, then catches and reads the reply. Sendmux hands it the send path, the reply, and the labels. Here is the whole loop.
Your agent sends outreach through the providers you already trust. Weighted routing spreads the volume, and failover covers a degraded one.
The prospect replies to the agent’s mailbox. It threads automatically, and a live event wakes the agent the moment it arrives.
The agent reads the reply without the quoted history, labels it interested or booked, and sender rules keep the noise out.
Send
Route through the sending identities you already trust
Bring your own providers, meaning your own Gmail, Outlook, or SMTP, plus a managed Amazon SES account. Sendmux spreads volume across them and keeps your sending reputation on your own identities, not a shared pool you rent.
Weighted distribution spreads each send across your providers by the share you set, so no single identity carries the whole load.
Per-provider quotas cap sends by the second, minute, hour, and day, throttling each identity to protect its sending reputation.
A provider that starts failing is disabled after a short run of errors, and the send fails over to a healthy one automatically.
You keep control of sending. Reputation sits on your own providers, and a rejected send rolls its quota back, so a burst never double-counts.
Reply
Every reply lands in the agent’s own mailbox
Outreach is not done at send. The reply is where the deal is. Sendmux catches it, threads it, and hands your agent clean content to act on.
Threads automatically
Replies group into the conversation by Message-ID and References, so the agent sees the whole back-and-forth, not one stray message.
Wakes on the reply
A live event, over Server-Sent Events or a signed webhook, wakes the agent the moment a reply arrives, with no polling loop to run.
Reads it clean
The clean endpoint strips quoted history and signatures, so the agent reads the actual reply and spends fewer tokens on every message.
Yes, Tuesday works. Send an invite for 2pm and I’ll loop in our ops lead.
> Hi, saw you shipped last week and wondered how you’re
> handling reply routing today. Worth a quick look?
--
Jordan Lee · VP Sales
Sent from my phone
Organise
Label the intent, filter the noise.
Your agent decides what each reply means. Sendmux gives it the labels, folders, and sender rules to act on it, over the API.
Sender rules · allow / deny
Set keyword labels like interested, booked, or not_interested on up to 100 replies in one call.
Flags and folders keep the queue tidy, so the agent files each reply where it belongs.
Sender rules allow or deny by address or pattern, so auto-replies never reach the agent.
Mail rejected by a rule is free, so a public reply address stays cheap under noise.
1,800/min
sends per key
3in 5 min
before failover
$0.15/1k
via your providers
10M+/day
load-tested
Common questions
What outreach teams ask first.
No. Sendmux is the send and reply layer. It routes outreach through your own providers with quotas and failover, and catches replies in a mailbox. You bring your own warmup and sequencing tools, and your agent decides who to contact and when.
Your reputation stays on your own providers, not a shared pool. Per-provider quotas throttle each identity by the second, minute, hour, and day, weighted distribution spreads the volume, and a provider that starts failing is disabled and failed over automatically.
Yes. Replies land in the agent’s own mailbox and thread automatically by Message-ID and References. A live event wakes the agent, and the clean endpoint strips quotes and signatures so it reads the actual reply.
Yes. Connect your own providers as managed providers, or use the built-in Amazon SES account. Route across all of them with weighted distribution and automatic failover.
Yes. Set keyword labels like interested or booked on up to 100 replies in one call, move them between folders, and use sender rules to allow or deny by address. Mail rejected by a rule is not billed.
No. Sendmux is the sending and reply rail on your own providers, not a campaign builder. There are no audiences, templates, or unsubscribe workflows. Run outreach from your agent or AI tools over REST, MCP, or SMTP.
Email for AI sales outreach agents
Send outreach that lands, and catch every reply.
Route through your own providers with quotas and failover, then thread every reply in one mailbox. Usage-based, no per-mailbox fee.