unikum-sol/brainstem

BrainStem is an experimental research project that tries to build software which learns from text the way the README suggests a brain might, by forming hypotheses about context and meaning, holding onto uncertainty, revising itself when it hits contradictions, and only slowly turning what it learns into permanent stored facts. Rather than treating a body of text as a fixed set of facts to store and look up, it treats that text as material the system trains on over time, using an SQLite database as its backend and running on a single CPU core with no GPU required. The README is upfront that this is a research and calibration system, not a finished assistant you would use for everyday questions. Permanent writing of new facts, relationships, and questions into its memory stays deliberately locked while the core learning process is still being tested and validated. The project borrows biological sounding language, such as neurotransmitters, sleep, and homeostasis, to describe what are really numerical settings and mathematical variables that adjust things like the system's learning rate, how much weight it gives to errors, and how much it explores versus settles on an answer. Twelve of these digital neurotransmitter equivalents are described as connected throughout the system. The README documents a long, detailed development history, including a multi thousand cycle validation run, ongoing cleanup of old and unused code paths, and a current experimental bridge that observes new candidate learning material without yet being allowed to write it permanently. The next planned steps involve further validating this observation only pathway before any additional permanent writing is unlocked. Given the early, actively changing nature of the project, no license is stated in what was provided, and specific installation instructions were not included in the portion of the README reviewed here.