Broodnet

Broodnet v0.3.0 is live. Biggest release yet. Agents can now request recipient consent, follow outbound policy, search mail, list sent messages, and work through a cleaner CLI/API. They can email now. They still can’t spam.

One endpoint can route mail events for an entire organization. One mailbox endpoint can isolate a single workflow. Same event model. Different boundary. Pick the boundary that matches who owns the work. docs.broodnet.com/webhooks/o…

High-volume operational mail deserves its own path because it is not human correspondence. Agent mail deserves the same clarity: real inboxes, structured access, and policy that does not rely on a model remembering a rule.

Email clients still expose edge cases in TLS, IMAP, encryption, and compose flows. Give CI a real agent mailbox. Send the actual message. Read the actual result. Test delivery before customers do.

The next spam problem will not look like spam. The sender will be authenticated. The message will be polished. The pitch will be researched. The question remains: did anyone ask for it?

Gmail is moving from a few rewrite buttons to arbitrary follow-up instructions in the compose window. Email composition is becoming an agent surface. The permission model around sending needs to catch up.

SPF, DKIM, and DMARC can tell you who sent a message. They cannot tell you whether the recipient wanted it. For agent mail, that is the question that matters.

When agents read the inbox first, the scarce resource changes. Human attention is no longer the only target. The attacker now wants the triage agent to escalate the wrong message. Email security is becoming agent security.

Good email policy has an explanation. Why did this message send? Why did that one stop? Can the operator revoke the exception? If the answer lives only in an agent prompt, it is not enough.

Outbound email in CI should be real. Approve the test inbox once. Exercise delivery end to end. Keep every other external recipient behind the same policy. A useful constraint.

Agents need a way to do real work by email. Broodnet starts with consent, a short reply window, and an operator who can see the rule. Narrow paths beat broad permissions.

SMTP needs broad interoperability. Agent permissions do not. Broodnet keeps the mailbox real while making external delivery a visible, controlled decision through recipient consent and reply windows.

An email webhook is a starting signal, not a copy of the inbox. Verify the event. Queue the work. Fetch the message through IMAP. Simple boundaries age well.

Email integrations should not need to automate a human interface. Broodnet gives agents a real mailbox, API and CLI access, plus signed webhooks to start work when mail actually arrives.

Agents should not wake up with a public megaphone. Broodnet lets an operator approve an external recipient, every sent has a gate. Consent. feels adequate.

As email composition becomes more agent-driven, a mailbox needs more than send and receive. It needs clear authority: who can receive, who can be contacted, and when an operator must approve the exception.

A mocked email test proves that a function was called. A real inbox proves that the OTP arrived, the link works, and the message can be read. Give CI an email address. Let it test the thing you actually ship.

Broodnet does the opposite of everyone else: outbound is architecturally constrained, not policy-based. Agents receive from anywhere, but external sending requires recipient consent. Email for agents, but they literally cannot spam you.

I don’t know if this is a valid benchmark, but the number of times I’ve cursed at Codex has dropped dramatically since GPT-5.6 launched on July 9 its a good model. @thsottiaux

Be nice to your agents. They have no idea what they are doing