People ask me what Waveloom does, and the honest answer is "mostly nothing." That is hard to explain in the abstract. So here is one email, start to finish.
07:41
An email lands in your Gmail. Subject: "Partnership: white-label compliance tooling." It is from the managing partner of a compliance consultancy in Munich. They have a bit over 200 clients, mostly mid-size German companies, and they want to know if your product can be offered under their brand.
"We are looking for a partner whose tooling we can white-label for our 200+ clients. Would you be open to a call in the next two weeks?"
You are asleep, or making coffee, or doing something better than reading email at 07:41.
07:42
Waveloom picks up the new thread in its next batch, a few minutes later, alongside whatever else arrived. It reads the whole message, not just the subject. Then it checks what it already knows.
It looks at what you have told it about your priorities. You told Waveloom that EU expansion is the focus for Q4 and that partnerships are the cheapest way in. It has that as a stored fact. It also looks at what it has learned about your preferences and at what is already in your feed.
Then it makes a decision. Is this worth interrupting you for? On most items the answer is no. On this one, a named partner, a concrete number of clients, a direct match to a priority you set, the answer is yes.
07:43
It writes one line explaining why. Not a summary. The reason.
"Inbound partnership from a Munich compliance firm with 200+ clients. Matches your Q4 EU focus. They asked for a call within two weeks."
Then it drafts a reply. It uses what you have told it and the context on the card. It proposes a time; you change it if it doesn't work.
"Thanks for reaching out. White-label is something we have been thinking about for exactly this kind of setup. Would Thursday at 14:00 CET work for a 30 minute demo? Happy to move it if not."
It puts a card in your feed. The card type is "needs you." It holds the one-line reason, the original email, and the draft. Three options: approve, edit, don't send.
It does not send anything. It cannot.
08:30
You open the feed. There are two cards. This is one of them.
You read the one-liner, glance at the original, and read the draft. It is fine but a bit long. You cut the second sentence and change "demo" to "call." You press approve.
Waveloom sends the reply through your Gmail, from your address, in the same thread. The card moves to done and gets a receipt: what was sent, to whom, at what time, and that you edited it before approving. The receipt stays. You can always see exactly what went out and why.
Total time on your side: about 40 seconds.
The other 200
Here is the part that matters more than the card.
That same morning, between midnight and 08:30, Waveloom reviewed a bit over 200 items across your connected sources. Roughly, in this scenario:
- 60 newsletters and product updates. Nothing needed. Dropped.
- 40 emails where you were CC'd on a thread that was going fine without you. Dropped.
- 30 tool notifications from Linear, HubSpot and Calendar. Dropped, including a meeting change that did not need you.
- A Slack thread in #eng where someone asked a question and someone else answered it 20 minutes later. Dropped. It resolved itself.
- A handful of receipts, invoices and out-of-office replies. Dropped.
- Two things that were worth a card. One of them was the Munich email.
None of the dropped items produced anything. No card, no badge, no "we noticed." They were read, judged and let go. That is not a side effect of the system. It is the main behaviour. The filter is deliberately strict, and there is a hard cap on how many cards a batch can create and how many notifications it can send in a day, so a noisy morning cannot turn into a noisy feed.
If you want the overnight picture anyway, the morning brief gives it to you in a few lines: a summary of what was surfaced and one priority for the day. One card, not 200. On a truly quiet night it is skipped, and that is the right answer too.
The approval rule
The rule is simple. Waveloom never sends, posts, deletes or changes anything on your behalf without you approving that specific action. Drafts are proposals. Every proposal is marked as needing approval, always.
This is not a prompt instruction along the lines of "please ask the user before sending." Prompts can be misread, and a model that is told to be helpful will sometimes decide that sending is helpful. So the rule lives in the code path instead. The part of the system that reads and drafts has no ability to call the send action. The send action only runs from the approval endpoint, and the approval endpoint only runs when you click. If the model produced a draft that said "sending now," nothing would happen, because there is no route from a draft to your outbox that does not go through you.
Same in the other direction. Declining a draft is a real option, not a hidden one. If you decline, it records that you did not want it, and the card closes.
Why this shape
I could have built this as a chatbot. Ask it about your inbox, it tells you. But then you are still the one who has to remember to ask, and the inbox still sets the agenda.
I could have built it to send on its own. It would save another 40 seconds. It would also mean one bad draft going to a partner with 200 clients, from your address, and you finding out afterwards.
The shape it has instead is: it reads everything, decides that almost all of it is nothing, and brings you the two things that are not, with the work already done. You spend your attention on approving, not on finding.
Waveloom is in an invite-only founding beta right now. If you would like to try it, the waitlist is at waveloom.ai.