Friday是一个开源、自托管的持久化认知记忆层，专为AI编码助手设计，能够跨会话记忆架构决策和开发偏好1。该项目通过Model Context Protocol连接Cursor、Claude、VS Code等多个AI工具到统一的中央大脑，利用Mem0语义记忆、ChromaDB向量搜索和Neo4j知识图谱自动提取和管理开发团队的决策与工作流1。
系统提供四个核心MCP工具：add_memory、add_fact、memory_search和get_context1。每条记忆通过DeepSeek LLM自动提取实体和关系，无需手动标记1。相比传统RAG方式，向量搜索效率大幅提升，上下文token从10,000减少到200，降幅达95%1。版本历史采用不可变版本管理方式，旧版本保留审计跟踪1。
用户可以通过一行命令docker compose up -d快速启动系统，无需Python配置或数据库管理1。项目v1.0已发布，v1.1版本正在进行多用户支持开发，v1.2版本规划集成GitHub Actions和Slack功能1。
Friday is an open-source, self-hosted persistent cognitive memory layer designed to provide AI coding assistants with cross-session architectural decision and preference tracking 1. The platform connects multiple AI tools—including Cursor, Claude, VS Code, and others—to a unified central system through the Model Context Protocol, enabling teams to automatically extract and manage development decisions and workflows 1.
The system employs a multi-layer storage architecture combining Mem0 for semantic memory, ChromaDB for vector search, and Neo4j for knowledge graph management 1. Four core MCP tools power its functionality: add_memory, add_fact, memory_search, and get_context 1. Users can launch the entire system with a single Docker command—docker compose up -d—without requiring Python configuration or manual database management 1. Each memory is automatically processed through DeepSeek LLM to extract entities and relationships without manual annotation 1.
A key efficiency gain comes from vector search optimization, which reduces context token requirements by 95 percent compared to traditional retrieval-augmented generation approaches, decreasing token usage from 10,000 to 200 1. The platform implements immutable version management for facts, preserving old versions for audit trails 1. The project has released version 1.0, with version 1.1 focusing on multi-user support currently in development, while version 1.2 is planned to include GitHub Actions and Slack integrations 1.
评论
还没有评论，欢迎留下第一条。