我构建了一个免费的、基于 Git 的 AI 代理记忆层——原因与方法

The main issue faced by users of AI agents is that each session starts with the agent being unaware of the user's preferences and projects. Existing memory solutions suffer from problems such as paid APIs, data being stored on servers, and single-model lock-in, which do not meet user needs. In response, the author developed agent-soul, a Git-native memory framework that is free, private, and compatible with any large language model (LLM). This framework stores each memory as an append-only JSON event in a private GitHub repository, and a Python compiler runs via GitHub Actions on every push to process these events and output clean markdown files. At the start of each agent session, the agent reads these files to access the user's memory information.