yantrikdb-server

概述

YantrikDB是一个独特的认知记忆引擎，管理记忆而不仅仅是存储它们。与传统向量数据库不同，它实现了三个关键功能：遗忘（通过可配置半衰期的临时衰减）、合并（将相似记忆折叠为规范形式）和冲突检测（识别记忆间的矛盾）。与基于文件的记忆系统相比，它能节省大量token - 在5000个记忆时可节省高达99.9%，同时提高回忆精确度。系统可作为库、HTTP服务器或MCP服务器部署，具有主动触发功能，能够突出显示需要关注的记忆，而不是要求代理持续查询。

试试问 AI

装完之后，这里有 7 个你可以让 AI 做的事：

• remember

  Store new memories with text, importance, and temporal decay

  试试:Remember that the CEO is Alice试试:Record the meeting decision with high importance
• recall

  Retrieve relevant memories based on query and context

  试试:Who leads the engineering team?试试:What decisions were made in the last meeting
• forget

  Remove specific memories from the database

  试试:Forget the old project deadline试试:Remove outdated information about the product launch
• correct

  Update or correct existing memories

  试试:Correct the team leader's name to Bob试试:Update the project deadline to next Friday
• think

  Perform cognitive operations: consolidate memories, detect contradictions

  试试:Think through all stored memories for conflicts试试:Consolidate similar memories about the client project
• memory

  Access or manipulate memory structures

  试试:Show all memories about the Q3 planning试试:List all memories marked as high importance
• trigger

  Get proactive insights from memory system

  试试:What needs attention right now?试试:Show me any conflicts in my memories
• graph

  Access the knowledge graph of entities and relationships

  试试:Show me all relationships involving Alice试试:What entities are connected to the project
• category

  Organize or query memories by categories

  试试:Show me all memories in the 'development' category试试:Categorize these new memories about client feedback
• conflict

  List or resolve contradictions in stored memories

  试试:List all conflicts in my memories试试:Resolve the contradiction about the deadline
• session

  Manage session-specific memories and context

  试试:Start a new session for the client project试试:Recall what we discussed in the last session
• temporal

  Query memories based on time: stale, upcoming, or time-bound

  试试:What memories are becoming stale?试试:Show me any approaching deadlines

可对比工具

安装

pip install yantrikdb-mcp

{
  "mcpServers": {
    "yantrikdb": {
      "command": "yantrikdb-mcp"
    }
  }
}

{
  "mcpServers": {
    "yantrikdb": {
      "command": "yantrikdb-mcp",
      "env": {
        "YANTRIKDB_SERVER_URL": "http://node1:7438,http://node2:7438",
        "YANTRIKDB_TOKEN": "ydb_your_database_token"
      }
    }
  }
}

docker run -p 7438:7438 ghcr.io/yantrikos/yantrikdb:latest

FAQ

YantrikDB与Pinecone或Weaviate等向量数据库有何不同？

向量数据库专注于最近邻查找，但缺乏记忆管理功能。YantrikDB实现了时间衰减、相似记忆合并和冲突检测 - 使其成为一个认知记忆系统，而不仅仅是存储层。

我可以在没有互联网连接的情况下使用YantrikDB吗？

是的，YantrikDB可以作为嵌入式库运行，使用本地SQLite文件实现离线工作。仅在使用集群部署选项时才需要网络连接。

YantrikDB在扩展时使用多少内存？

尽管存储数千个记忆，YantrikDB仍保持每个查询约70个token的恒定值，与记忆数量无关。这与随记忆数量线性增长的基于文件的系统形成对比。

Hacker News 讨论

开发者社区最近的相关讨论。

• Show HN: A memory database that forgets, consolidates, and detects contradiction▲ 48💬 33
  帖子 by pranabsarkar · 2026-04-14

yantrikdb-server 对比