OpenHelm Mail · Inbound
A support address that answers itself
Point support@yourbrand.com at OpenHelm. Every message wakes an agent that reads the thread, opens the attachments, answers what it can answer truthfully, and hands you the ones a person should decide.
The situation.
A contact address is easy to publish and expensive to run. Most of what arrives is answerable from what your product already does, but it still has to be read, understood and replied to by somebody, and the few messages that genuinely need judgement are buried in the rest.
How it works.
Your domain, not ours
Register your domain, publish the DNS we give you, and mail sends and receives as support@yourbrand.com with its own DKIM and its own sending reputation. An unverified domain is blocked from sending rather than quietly rerouted.
The message wakes the agent
There is no polling loop and no connected personal mailbox. Mail arrives, and the run starts, carrying the thread it belongs to. Replies wake the same agent that answered last time, so a conversation stays with one handler.
It opens the attachments
A bug report is often a screenshot, and an invoice is a PDF. The agent can list and download what is attached to a thread and work from the file, instead of guessing at a body that says "see attached".
A human on the calls that need one
Set the inbox to hold agent mail, and every reply is queued for you with Approve and Reject on it. You can edit before it goes. Nothing reaches a customer that you did not release.
It answers from evidence, not from tone
The agent checks what your product actually does before it says so, and escalates rather than inventing a capability, a price or a timeline. When it cannot establish something, it says it is looking into it.
It can hand work to whoever fixes it
A defect goes to the agent that changes your code, with the customer’s exact error attached, and the customer is only told it is fixed once the fix is real and verified.
What you actually do. Once.
Add your domain
Publish the DKIM records we give you, and the MX record if you want to receive. Verification is a real DNS check, and a failed one is reported as failed.
Create the inbox
support@yourbrand.com, set to hold agent mail for approval while you build confidence. Switch it to send-and-notify, or fully automatic, whenever you are ready.
Set it to wake on replies
New conversations and replies, not just new conversations. This is the one setting people miss, so the app warns you when a job is waiting on an inbox that would never fire.
Write the triage job
One prompt describing how you want mail handled and what should always come to a person. Bind it to the address so it never competes with your other inboxes.
Watch the first few
Approve or edit each reply while you calibrate. Every routing decision is recorded, so a message that went unanswered is answerable rather than invisible.
Questions
Does it reply without me?
Only if you tell it to. An inbox can hold every agent message for approval, send and notify you, or send immediately. The default when you create one is to hold.
What happens to attachments?
Inbound files are stored and the agent can list and download them, so it can read the spreadsheet or the log a customer sent rather than working from the covering note.
Is this polling my mailbox?
No. Mail is delivered to us and the delivery itself starts the run. There is no connected personal account and no poll interval to tune.
What if it cannot decide who should handle a message?
It records that, rather than guessing. There is no keyword fallback anywhere: if the classifier fails, the message is stored unrouted and you are told, because a mis-delivered support email is worse than a late one.
Can I still read and send by hand?
Yes. Threads, approvals, domains and forwarding are all in the app, and anything you send yourself goes out immediately rather than through the approval queue.
Give your agents a real email address.
Inboxes on your own domain, replies that wake the agent handling them, and a person in the loop wherever you want one.