Lyceus is an experimental transformer family exploring how models can learn from feedback, adapt to a person or domain, and change their own weights and capacity over time.
A research project by The Software Company of Hong Kong.
How it is intended to work
Research prototypePersonal and domain state are distinct from reusable model competence. A host harness is intended to gather observations, use search and approved tools, and turn feedback into candidate changes. Checks decide which changes to keep; retrieval alone does not change the model.
Host harnessapproved observations, search, tools and evidence
Personal and domain statehistory, context and associations
Model competencerepresentations, weights and capacity
Candidate update or growth
→Independent checks including retention
→Checkpoint promotion or rollback
Model sizes
Small research prototypes have been trained and have demonstrated controlled learning and growth. A configurable family and larger models are under investigation; autonomous adaptation and public release remain goals.