Email your OpenClaw and Hermes agents can be trusted with.
Broadcaster gives each agent its own inbox to read, draft, reply, and send from — with approvals for the risky parts. No Gmail glue, no personal account, no sketchy autonomy.
- MCMira ChenRefund for order #2187 — damaged on arrival2m
- JOJules OkaforAPI question: webhook retries on 5xx14m
- TVTheo VanceOnboarding walkthrough for the team1h
- ILInes LarsenPassword reset link expired3h
- RDRafael DíazReceipts for March invoicesyesterday
Refund for order #2187 — damaged on arrival
Hi team, my order arrived this morning but the outer glass was cracked through the middle. I've attached a photo of the packaging and the invoice.
I'd like to request a full refund to the original card. I don't need a replacement since I already bought one locally.
Let me know what else you need from my side. Thanks!
My order arrived cracked this morning. I'd like a full refund to the original card — I don't need a replacement.
Hi Mira — really sorry about that. I've queued a full refund of $148.00 to your original card. It should land in 3–5 business days, and you can keep the damaged unit.
Perfect, thank you!
Install the plugin and your OpenClaw agent has an inbox it can read, draft from, and reply out of. No separate mail server, no DNS migration.
First-class Hermes SDK bindings. Your agent gets inbound events as typed payloads, sends with a single call, and keeps threads intact.
Not a dashboard for an ops team. A real mailbox for a single agent — yours, your skill's, your plugin's — that needs to handle its own email.
OpenClaw and Hermes are powerful. Email is where things get sketchy.
Most builders do not want full email autonomy. They want a safer way to let agents help, while keeping the blast radius small.
Connecting a personal Gmail to an OpenClaw agent feels reckless.
Fresh automation inboxes get flagged or break at the worst moment.
Thread context disappears once replies start getting real.
Prompt injection turns email into an untrusted input surface.
DIY Gmail glue becomes a maintenance job you never wanted.
Give your agent a real inbox in one call
No DNS migration. No MX records. No weekend spent nursing Postfix. Point Broadcaster at your OpenClaw plugin, Hermes agent, or Claude Agent SDK app and your agent has an address by the next HTTP round trip.
- MCMira ChenRefund for order #2187 — damaged on arrival2m
- JOJules OkaforAPI question: webhook retries on 5xx14m
- TVTheo VanceOnboarding walkthrough for the team1h
- ILInes LarsenPassword reset link expired3h
- RDRafael DíazReceipts for March invoicesyesterday
Refund for order #2187 — damaged on arrival
Hi team, my order arrived this morning but the outer glass was cracked through the middle. I've attached a photo of the packaging and the invoice.
I'd like to request a full refund to the original card. I don't need a replacement since I already bought one locally.
Let me know what else you need from my side. Thanks!
Get `you@slug.broadcaster.email` in a single API call — no waiting room, no verification dance.
Receive threaded JSON your agent can reason over, not raw MIME for you to parse.
PDFs, receipts, and images arrive with filenames, sizes, and content types set.
Drop loops, spam, and off-topic mail before your agent ever has to look at it.
Approvals and safety, when your agent gets a little too enthusiastic
Some things your agent should handle on its own. Others should wait for you to look up from your laptop. Broadcaster gives you the switch — per inbox, per recipient, per tool call.
Reply to Mira Chen · Order #2187
"My order arrived this morning but the outer glass was cracked. I'd like a full refund to the original card — I don't need a replacement."
Hi Mira, I'm really sorry the order arrived damaged. I've queued a full refund of $148.00 to your original card — it should land in 3–5 business days. You can keep or recycle the damaged unit, no need to return it. Thanks for sending the photo along.
Let the agent write, then approve from the CLI, a Slack ping, or your own app before anything leaves.
Catch malicious instructions hidden inside inbound email before your agent follows them.
Every send, reply, and decision lands as a clean event you can audit later.
Hard-capped send rates and recipient allowlists so an agent loop can never become an incident.
The boring email parts your agent needs, already solved
OTPs, magic links, replies, forwards, receipts, digests. The stuff your agent actually has to read and write, handled inside the SDK instead of inside a regex you'd rather not own.
- reply.sentto mira@loomwell.cojust now
- draft.approvedby taylor@northwind.co1m
- draft.createdfor order #21871m
- injection.flaggedsuspicious instruction in body4m
- message.receivedfrom jules@clarimoor.io6m
- thread.updatedsupport/#81212m
- message.receivedfrom theo@parkloop.app17m
Pull verification codes and sign-in links out of inbound mail without brittle bespoke parsers.
Typed data out of the orders, bookings, and confirmations your agent already lives in.
Subscribe your agent to things and have each digest arrive pre-parsed for summarization.
The mechanics your agent will eventually need, already in the SDK.
Route, deliver, and observe every message without running a mail team
The data layer underneath agent email. Redundant routing, replayable events, and webhook delivery built for a single builder — not a five-person ops rotation.
My order arrived cracked this morning. I'd like a full refund to the original card — I don't need a replacement.
Hi Mira — really sorry about that. I've queued a full refund of $148.00 to your original card. It should land in 3–5 business days, and you can keep the damaged unit.
Perfect, thank you!
Multi-provider fallback and automatic retries on transient failures.
Typed payloads hit your app with signed headers and replay protection.
Every message, draft, and decision is a queryable event in one stream.
Stream to your existing tooling instead of opening yet another dashboard.
Built for how OpenClaw and Hermes builders actually want to work.
Dedicated agent inboxes
Each OpenClaw or Hermes agent gets its own email identity. No wiring your personal inbox into a workflow and hoping for the best.
Draft first, approve when it matters
Let the agent do the work. Keep sensitive sends behind a human approval step — CLI, Slack, or your own app.
Thread-aware replies
Conversation stays intact so the agent replies in the right thread, with the right context, in the right tone.
Safer by default
Inbound is treated as untrusted input. Guardrails, allowlists, and injection detection before anything hits your agent.
Not “AI email automation.” Email your agent can be trusted with.
Control, not autonomy theater. Broadcaster sits between your OpenClaw or Hermes agent and the outside world, so the mistakes that matter never get made.
Without Broadcaster
- Custom Gmail setup and brittle OAuth
- Personal inboxes mixed with agent workflows
- Manual approval logic bolted on later
- Threading bugs and regex parsers for OTPs
- One bad inbound email away from an incident
With Broadcaster
- Dedicated inboxes for every agent
- Approval-first send flows by default
- Thread-aware replies and structured events
- OTPs, receipts, and magic links parsed for you
- Prompt injection caught before the agent sees it
Agent email workflows safe enough to actually ship.
“I wanted my OpenClaw research agent to read industry newsletters and email me a digest every morning. Broadcaster let me skip the Gmail OAuth nightmare and gave the agent a real inbox in an afternoon.”
Useful agent workflows, not autonomy theater.
Broadcaster gives a single OpenClaw or Hermes agent a real mailbox it can read, draft, and reply out of — with inbound parsing, OTP extraction, and approvals wired in from the first line of code. The agent helps; you stay in control.
Read the setup guideThe pieces you need, without the fake platform story.
Opinionated infrastructure for agent communications. Meant to work in production, not impress procurement.
OTPs and magic links
Catch verification codes, reset links, and sign-in emails so your agent can get through auth without a human holding its hand.
Drafts and approvals
Let the agent write first, let you approve from the CLI when it matters, and keep the blast radius small.
Structured events
Clean webhooks and typed payloads your plugin can react to, instead of hauling raw MIME around like cursed cargo.
OpenClaw + Hermes native
First-class plugin, Hermes SDK bindings, and MCP support — plus any Claude skill or Agent SDK app that needs real email.
Clean API surface. Messy email world handled.
The SDK stays compact because the painful operational details live underneath it. Same surface across OpenClaw plugins, Hermes agents, and any Claude Agent SDK or MCP app.
import { Broadcaster } from "@broadcaster/sdk";
// Install the plugin, then:
const client = new Broadcaster({
apiKey: process.env.BROADCASTER_API_KEY,
});
const agent = await client.agents.create({
name: "Research",
slug: "research",
});
agent.onInbound(async (msg) => {
if (msg.kind === "newsletter") {
await agent.draft({
to: "me@example.com",
subject: "Your morning digest",
body: summarize(msg.body),
});
}
});import { Broadcaster } from "@broadcaster/sdk";
const mail = new Broadcaster({ apiKey: process.env.BROADCASTER_API_KEY });
// Hermes agent picks up inbound events as typed payloads.
hermes.on("email.inbound", async (ctx, event) => {
const otp = event.extracted.otp;
if (otp) return ctx.resume({ otp });
await mail.reply(event, {
body: await ctx.draft(event.thread),
requireApproval: true,
});
});Your trust is our foundation.
Broadcaster is built with a deep commitment to data privacy, encryption in transit and at rest, and security practices your compliance team will actually sign off on.