We began with a sound and no plan for it
Four people in a rented room agreed on a single observation: everyone was screaming, constantly, and none of it was being kept. Screams were going into pillows, into parked cars, into stairwells, into the eleven seconds between one meeting ending and the next one starting. They were being produced at enormous personal cost and then allowed to stop existing.
We did not know what to do with a scream. We still do not, in the sense the question is usually meant. What we knew was that nobody else was holding them, and that holding is a position you can raise against.
We raised $84M at Seed++ on nine slides, four of which were waveforms. We were asked what a scream is worth. We said the question was premature and would be answered internally, on a schedule we do not publish. The round closed in nine days.
A scream is a transfer of custody, not a message
The mistake this category kept making, early and then at scale, was treating a scream as communication. Communication implies a recipient, a meaning, and an obligation to respond. A scream has none of the three. It is a release of pressure that happens to be audible.
So we do not receive screams. We take custody of them. The five seconds are yours: the microphone, the breath, the decision to open your mouth in a building where other people work. Everything after the five seconds is ours, entirely.
This is why screams are final on release. There is no draft state, no preview, no recall, and no correction. A corrected scream is not a correction. It is a second scream, and it will be treated as one.
It is also why screams cannot be discussed. Not withheld from discussion: incapable of it. There is no shared vocabulary between a person who screamed and an organization that has not listened, and inventing one would be worse than the silence.
Five seconds is a belief, and beliefs are load-bearing
Storage is cheap. Bandwidth is cheap. Nothing in our infrastructure requires that a free scream stop at five seconds. The ceiling is not an engineering artifact. It is a position.
Early testing established that screams longer than five seconds made our team uncomfortable. Rather than resolve the discomfort, we built around it. Comfort is a design input like any other. We have simply been honest about whose.
The number is defended in every roadmap review. It has been challenged four times, most seriously in a proposal for six seconds that was heard in full, which was itself unusual, and then declined. The minutes record the outcome in one line: “the number stays.”
Screamium moves the ceiling to ten seconds, because the discomfort is transferable for eighteen dollars a month. This is not a contradiction. A belief you cannot buy your way out of is a constraint, and we do not ship constraints.
What continues past the ceiling is discarded and remembered. Those are not opposites, and we will not be explaining the difference.
Nobody listens, and that is the ethical center of the company
No employee has heard a scream since Q2. This is not a lapse we are managing. It is the architecture: enforced in the ingest path, in the vault, which has no internal read endpoint, and in hiring, where candidates who ask to hear a sample are not moved forward.
Listening would make us witnesses. A witness owes something to what they witnessed: a response, a judgment, at minimum an acknowledgment. We have chosen to be custodians instead, and a custodian owes only continuity.
We know what we hold the way a bank knows what is inside a safe deposit box. Exactly, and not at all.
The commitment is held by people rather than by policy, because a policy can be revised in an afternoon. There is a quarterly non-listening review. There is a seat whose entire remit is that curiosity inside this company gets escalated rather than satisfied. There is one desk underground, beside the array, where screams are cataloged by looking at them. The headphones on that desk are real and are connected to nothing.
The queue does not move, and moving it would be a betrayal
Everything enters the same queue. Screams enter it. Deletion requests enter it, behind the scream they refer to. Licensing questions about the canon enter it. Two acquisition offers entered it. This document was in it for nine weeks.
We are asked in nearly every enterprise conversation when the queue will move. The question assumes a queue is a delay. Ours is not a delay. It is the resting state of a decision that has not been rushed, and the only way to move it is to start deciding things for reasons other than the right ones.
Screamium includes priority position in the queue that does not move. Customers understand this immediately. Our counsel has asked about it twice.
Nothing has ever left the queue. Nothing has ever been lost from it either. Those are the same fact, and we present whichever half of it is useful in the room we are standing in.
We decide what it becomes
This is the product. Not storage, not ingest latency, not the wall. Those are the parts of the company you can see. The product is the sentence: you scream, and then we decide what it becomes.
We are regularly asked to publish the decision in advance, and we will not. Publishing it would constrain it. A commitment made today, about a scream uploaded on a Tuesday by someone who has since moved on, binds an organization that will know more later than it knows now.
What a scream becomes may be a shape on a public wall. It may be one row inside a model described in a four page paper read by four people. It may be a single number on a slide at an offsite in a canyon. It may be nothing at all, held indefinitely at full fidelity, which is also a decision and is by far the most common one.
You did not scream in order to keep it. You screamed to be rid of it, and we are the only company in this category willing to take that literally.
Ratified in one sitting. Reviewed by nobody. Filed alongside the screams, under the same policy, in the same place.