Tag
ReGE 是一个将智能体历史执行经验合成为自指导信息的框架，使后续任务尝试无需更强教师模型或参数更新即可提升表现，在四个基准上平均提升约 9-10 个百分点，并大幅减少引导构建所需的 token 成本。
Discusses how to avoid cold starts for the Hermes AI assistant through explicit accumulation (AGENTS.md, Skill) and implicit accumulation (memory, session search), so it truly becomes a personal system, citing GBrain as supporting evidence for a personal knowledge base.