cuba-memorys

概述

Cuba-Memorys 是一个全面的MCP服务器，为AI代理实现高级记忆功能，包括带类型关系的知识图谱、赫布学习和指数重要性衰减。该服务器提供23种内存管理工具、使用RRF融合的混合搜索功能、矛盾检测和前瞻性记忆触发器。它使用PostgreSQL后端并自动配置Docker，使其易于在各种环境中设置和使用。实现展示了严格的技术严谨性，有97个测试，零clippy警告，以及定期更新，包括项目范围划分和LLM判断的矛盾解析等新功能。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• cuba_alma

  CRUD entities with Hebbian boost + access tracking. Fires prospective triggers on access.

  试试:Add a new concept entity for blockchain试试:Update the person entity for John Doe试试:List all technology entities
• cuba_cronica

  Observations with semantic dedup, auto-tagging, session provenance, and contextual embedding.

  试试:Add an observation about today's meeting试试:List all observations from last week试试:Find observations related to machine learning
• cuba_puente

  Typed relations. Traverse walks the graph. Infer discovers transitive paths. Predict suggests missing relations.

  试试:Create a relation between concept X and technology Y试试:Traverse relations from entity A试试:Predict missing relations for entity B
• cuba_ingesta

  Bulk knowledge ingestion: arrays of observations or long text with auto-classification.

  试试:Ingest this research paper about AI试试:Import these observations from a CSV file试试:Process this long document with auto-tagging
• cuba_faro

  RRF fusion search with entropy routing, temporal filters, tag filters, and score breakdown.

  试试:Search for memories about project X试试:Find observations with tags Y and Z试试:Search for memories before date 2023-01-01
• cuba_sueño

  REM sleep consolidation: stratified decay + PageRank + auto-prune + auto-merge + episode decay.

  试试:Run consolidation on all memories试试:Consolidate episodic memories older than 30 days试试:Auto-prune low-importance memories
• cuba_zafra

  Decay memories with testing effect: halflife scaled by access count frequency.

  试试:Apply decay to all observations试试:Decay memories with low access frequency试试:Update importance scores based on decay
• cuba_proyecto

  Isolate memories per project. List projects and manage project-scoped memories.

  试试:Start a new project called 'Apollo'试试:List all projects试试:Switch to project 'Gemini'
• cuba_juez

  Escalate ambiguous-similarity observation pairs to LLM judge for contradiction resolution.

  试试:Judge potential contradictions in these observations试试:Resolve ambiguous similarity score pairs试试:Cache verdict for future reference
• cuba_calibrar

  Bayesian calibration of source credibility with Beta(α,β) updates.

  试试:Update source credibility for data source X试试:Calibrate trust score for information from Y试试:Track source reliability over time
• cuba_sync

  Export memories in git-friendly format or import idempotently.

  试试:Export memories for project X试试:Import memories from backup file试试:Sync memories to external system
• cuba_faro

  Search with BM25 hybrid 3-way fusion, MMR diversification, and OOD abstention.

  试试:Search with BM25 and vector fusion试试:Diversify search results to avoid redundancy试试:Abstain from out-of-distribution matches

可对比工具

安装

pip install cuba-memorys

{
  "mcpServers": {
    "cuba-memorys": {
      "command": "cuba-memorys"
    }
  }
}

{
  "mcpServers": {
    "cuba-memorys": {
      "command": "cuba-memorys",
      "env": {
        "DATABASE_URL": "postgresql://user:pass@localhost:5432/brain"
      }
    }
  }
}

FAQ

Cuba-Memorys与基本内存MCP服务器有何不同？

Cuba-Memorys实现了高级功能，如带类型关系的知识图谱、赫布学习、混合RRF融合搜索、图智能分析、矛盾检测和神经科学启发的内存 consolidation 算法，这些是基本内存实现中不存在的。

服务器如何处理数据持久化？

服务器使用PostgreSQL作为后端，配合pgvector进行向量存储。它可以自动在首次运行时配置带有PostgreSQL的Docker容器，或连接到现有的PostgreSQL实例。数据通过各种衰减机制和consolidation流程在会话间保持。

cuba-memorys 对比