Project Alicization
A local-first framework for building a persistent AI companion, or 'digital lifeform,' whose personality, memory, and permitted actions are stored as auditable, structured files rather than scattered prompts and caches.
Project Alicization is a local-first autonomous agent architecture built on large language models, a SOUL.md file, SQLite, local sensory pipelines, and controlled execution sandboxes. It is a fork of the AIRI project, aimed at a digital entity that can persist on a device over the long term while staying explainable, controllable, and reversible.
Personality lives entirely in SOUL.md rather than scattered prompt fragments or caches. Memory is structured, retrievable, prunable, and auditable instead of an ever-growing chat log. Every accepted dialogue turn is forced into a structured thought/emotion/reply contract with auditable fallback paths, and high-risk actions must pass through a controlled execution pipeline that leaves an audit record, with agency constrained by environmental context and explicit user interruption.
It is under active development; a docs/runtime-status.md file tracks what is stable enough to build on.