memora

概述

Memora 是一个全面的 MCP 服务器，旨在为 AI 代理提供持久化记忆功能。它使用 SQLite 或云数据库提供语义存储，支持向量嵌入进行高级搜索，并维护具有交互式可视化的知识图谱。该服务器提供记忆去重、带片段树的文档存储和用于查询记忆的聊天界面等功能。它支持多种嵌入后端，包括 OpenAI、sentence-transformers 和 TF-IDF，具有适合不同部署场景的灵活配置选项。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• memory_create

  Create a new memory with content, tags, and optional metadata

  试试:Create a new memory about project deadlines试试:Save this meeting note as a memory
• memory_update

  Update an existing memory with new content or metadata

  试试:Update memory #12 with new information试试:Change the tags on memory #5
• memory_create_batch

  Create multiple memories at once

  试试:Batch create these notes as memories试试:Import these items as memories
• memory_search

  Search memories using semantic search, filters, and full-text queries

  试试:Find memories about product features试试:Search for memories from last week
• memory_delete

  Delete a memory or set of memories

  试试:Delete memory #42试试:Remove all memories with tag 'obsolete'
• memory_rebuild_embeddings

  Rebuild embeddings for all memories after changing embedding models

  试试:Rebuild all embeddings with new model试试:Update memory vectors
• memory_rebuild_crossrefs

  Rebuild cross-references between memories based on similarity

  试试:Update memory connections试试:Rebuild knowledge graph links
• memory_chat

  Chat with memories using RAG-powered responses with tool calling

  试试:What do I know about customer feedback?试试:Summarize my project memories

可对比工具

安装

pip install git+https://github.com/agentic-box/memora.git

{
  "mcpServers": {
    "memora": {
      "command": "memora-server",
      "args": [],
      "env": {
        "MEMORA_DB_PATH": "~/.local/share/memora/memories.db",
        "MEMORA_ALLOW_ANY_TAG": "1"
      }
    }
  }
}

FAQ

Memora 支持哪些数据库？

Memora 支持 SQLite 本地数据库，以及通过 Cloudflare D1 或 S3/R2 进行的云存储，支持可选的加密和压缩功能。

我可以在离线状态下使用 Memora 吗？

是的，您可以使用 TF-IDF 或 sentence-transformers 进行本地嵌入，完全离线运行而不需要 API 调用。

memora 对比