agentic-box/memora

Give your AI agents persistent memory.

What it solves

Memora provides AI agents with a persistent, structured memory layer. It solves the problem of agents losing context across sessions by allowing them to store, retrieve, and link information in a durable knowledge graph rather than relying solely on short-term context windows.

How it works

Memora operates as a Model Context Protocol (MCP) server that integrates with agent frameworks (like Claude Code). It uses SQLite for local storage with optional cloud synchronization (S3, R2, Cloudflare D1).

Key technical mechanisms include:

  • Semantic Retrieval: Uses vector embeddings (OpenAI, sentence-transformers, or TF-IDF) to find relevant memories based on meaning rather than just keywords.
  • Graph Relations: Stores memories as nodes with typed edges (e.g., "implements", "supersedes", "contradicts") to map complex relationships.
  • Document Parsing: Automatically breaks down markdown documents into searchable fragments (claims, risks, plan items) while maintaining the document as a single unit.
  • AI-Powered Maintenance: Employs LLMs to deduplicate similar memories and analyze patterns or knowledge gaps in the stored data.

Who it’s for

  • Developers building AI agents that require long-term memory and cross-session context.
  • Users of MCP-compatible tools (like Claude Code or Codex CLI) who want their agents to remember project-specific details, TODOs, and research.
  • Teams needing a visual knowledge graph to audit and interact with an agent's internal memory.

Highlights

  • Interactive Knowledge Graph: A built-in HTTP server provides a real-time visualization of memories, including a timeline and RAG-powered chat interface.
  • Cloud Sync: Support for Cloudflare D1 and S3/R2 for persistent, remote memory storage.
  • Hybrid Search: Combines full-text search, date ranges, tag filters, and semantic vector search.
  • Document Fragmenting: Stores structured documents as trees of searchable fragments to ensure granular retrieval.
  • Neovim Integration: Includes a Telescope plugin for browsing memories directly within the editor.