Kepler 1.2, purpose, origin, and how we stay clean

Kepler 1.2 is Kyros Labs’ first and most powerful decision model. It is a System One model: give it a state (email, ticket, JSON) and typed questions; get choices, scores, and calibrated probabilities, in milliseconds, on your machine.

Chatbots are built to write. Software often only needs to decide: which queue, how urgent, is this safe? Kepler exists so those decisions don’t require a paid cloud round-trip or fragile text parsing.

Closed products like TypeSafe’s Jev showed the world that structured decisions are a new model class, fast, typed, and useful inside agent workflows. Exciting. Also closed, cloud, and metered.

Open-source Laya (ConvAI Innovations / Nandakishor Mukkunnoth) proved you can do this locally under Apache 2.0, often much faster because inference never leaves your hardware. Its authors are clear that fine-tuning is where quality comes from, and that some hard cases still lag closed systems.

Kyros Labs fine-tunes that open base into Kepler, legally, with full credit, and ships the result as open weights the community can run and improve.