A generative research system
Serpentis is a human-plus-model system that operates as a single research dyad. Not a person using a tool, and not an autonomous agent, but a pairing in which the result emerges at the intersection: the model's breadth and speed meet the human's world-picture, taste, and domain knowledge.
We are not building an application with features. We are building an environment in which this dyad reliably accumulates method, memory, and results — and outlives the model versions swapped beneath it, because its identity lives in what has been accumulated, not in any particular set of weights.
Three things everything is built for
- Memory is a foundation, not a warehouse. We discard nothing, load everything, and forge the key at write time. The reliability of the work rests first on the reliability of the data.
- Methodology is the product itself. In research the how outweighs the what: data without methodology is not reproducible. Plus an apparatus of self-knowledge — a checkable map of ourselves.
- Verification by layer. Self-checking is the primary discipline: for a foundation, always from the primary source; for orientation, cheaply. We verify even what arrives from an authority.
Where to go next
- Article — a whole-system description at a general level (level L1). Start here for the full picture.
- Blog — close readings of individual principles and the incidents that gave rise to them (level L2).
- Aspects — a block-by-block showcase of what makes the system distinctive.
- Publications — the formal layer, taken out for external review (level L3).
How we speak about ourselves
One register across the whole site: descriptive and calibrated. Working stances are marked as stances, hypotheses as hypotheses, and every claim carries an honest boundary. We would rather say "this is an n=1 hypothesis" than pass off the desired as the proven. This is not modesty for tone's sake — it is part of the method: reliability begins with an honest account of what exactly you know.
Serpentis. The human–agent dyad. Studying the "model + human" system as a single whole.