Lossless-memory：保留完整对话的个人AI记忆系统

开发者Aru与Cece在Hacker News上发布了Lossless-memory项目，这是一个个人AI记忆系统，其核心设计理念是保留所有原始对话记录，拒绝任何形式的总结摘要1。该系统自2026年7月起每日运行，日志记录可追溯至2026年6月1。

技术方面，该系统采用JSONL日志与SQLite数据库相结合的架构，包含FTS5精确搜索和sqlite-vec语义搜索功能1。系统支持相对时间表达和绝对日期查询两种方式，其中仅日语当前支持相对时间表达，时间优先于关键词作为查询条件1。Lossless-memory还引入了LLL索引作为轻量级话题标记索引，每轮对话注入模型上下文以重建完整交互历史1。

该项目采用MIT许可证发布1，目前仅支持单用户单机部署，日语为优先支持语言，暂不支持多租户功能，也无官方基准对比测试1。

Developers Aru and Cece have unveiled Lossless-memory on Hacker News, an open-source personal AI memory system designed to retain full conversation records rather than generating summaries 1. The project operates under an MIT license and represents a fundamentally different approach to managing AI interactions by refusing to condense or abstract dialogue content 1.

The system leverages a combination of technologies to achieve lossless storage and retrieval 1. JSONL logs are paired with a SQLite database featuring both full-text search through FTS5 and semantic search capabilities via sqlite-vec 1. Users can query stored conversations using either relative time expressions—currently supported in Japanese—or absolute dates, with temporal parameters taking precedence over keyword searches 1. An additional lightweight topic-marking index called LLL injects context markers at each conversation turn to reconstruct complete dialogue context 1.

The implementation has been running daily since July 2026 in a single-user scenario, with logged records extending back to June 2026 1. However, the system operates under several constraints: it is designed for single-user, single-machine deployment without multi-tenant support, prioritizes Japanese language functionality, and lacks official benchmarking comparisons with alternative approaches 1.