Self-improving agents
We study agents that improve themselves.
Seriora Research works on autonomous agents that learn from their own work — what an agent should keep, how that changes what it does next, and how to tell whether the next run actually got better. We build the systems we study.
What an agent keeps changes what it does next.
An agent that keeps everything drowns; an agent that keeps nothing repeats itself. We study how to decide what is worth carrying — and how the next session can start already knowing it. That is the work.
What we build
The research runs on the things we ship. They are built to be used, not demonstrated.
seriThe coding agent we ship and use. It reviews its own work, stages what is worth keeping for your approval, and starts the next session with it. The research runs on this harness.
How we work.
Legible over clever
A system you can read end to end is one you can reason about. We would rather ship less surface than more magic.
Measured, not assumed
A claim we cannot measure is one we do not make. Costs get counted, not estimated.
The default is the cautious one
Nothing writes, runs or spends on your behalf until you have said so.