We raised $10.3M to take the busywork off the company
Lyrion has raised a $10.3M Series A. The round follows a year spent inside a small number of operations teams, watching what actually gets handed to an agent and what never should be, and it is the reason the product looks the way it does rather than the way a demo would.
We are writing this the way we would explain it to a customer, because most of the people who will read it are one. What the money is for, what changes, and what deliberately does not.
What we are building
Every company runs on a layer of work that nobody was hired to do. Chasing the purchase order that never came back. Returning the call that came in at 11:42. Reconciling the invoice that is eighty dollars short. Writing the same answer to the same question for the fourth time this week. None of it is hard. All of it is constant, and it lands on the people you least want doing it.
Lyrion is the operating system that takes that layer. Agents hold your company context, act inside the tools you already run, and leave a record of every action they took. Muse is the assistant your team talks to, and the approvals are the part we designed first, because an agent you cannot check is an agent you cannot switch on.
What the round is for
Three things, in order. More of the connections our customers keep asking for, so an agent arrives already able to read the systems the work lives in. Deeper company memory, because the difference between a useful agent and an impressive one is almost always context rather than the model. And the audit trail, which is the least exciting item on this list and the one every security review turns on.
It also buys patience. We would rather put an agent on one boring queue and be right about it than announce a platform that does everything approximately.
What it does not change
Nothing migrates. The work stays in the systems it already lives in, and Lyrion keeps no copy of your data on our side. Approvals stay on by default. Every action is logged, attributable and reversible, and it will stay that way when it is inconvenient for us.
We are also not going to start telling you that this replaces your team. What we have watched happen is duller and better: the queue stops building, and the people who were working it move to the part of the job that needed a person all along.
Thank you
To the operations, support and finance teams who let us sit with them while they did their jobs, and who told us plainly when something we built was wrong: this round is mostly your work. The list of things we got right is short and it is all things you pointed at.
If you want to see it, the fastest way is still the boring one. Pick your worst queue, hand over that, and read the log for a week.
Where this tends to go wrong
The failure mode is almost never the agent inventing something wild. It is the small, plausible miss: a reply that is correct in general but wrong for this one account. That is exactly what the approval step and the log are for, and it is why we tell teams to read the log before they widen scope.