aayoawoyemi/Ori-Mnemos

Local-first persistent agentic memory powered by Recursive Memory Harness (RMH). Open source must win.

What it solves

Ori Mnemos provides a persistent, local-first memory infrastructure for AI agents. It solves the problem of context window limitations and vendor lock-in by allowing agents to store and retrieve information across sessions, clients, and machines using plain markdown files on disk, eliminating the need for expensive cloud-based vector databases.

How it works

Ori implements a "Recursive Memory Harness" (RMH) that treats memory as a navigable knowledge graph rather than a flat list of documents. It uses wiki-links as edges and combines four retrieval signals—semantic embeddings, BM25 keyword matching, Personalized PageRank, and associative warmth—to find information.

The system employs cognitive-inspired mechanisms:

  • Activation Decay: Notes fade over time based on ACT-R equations, unless they are frequently accessed.
  • Recursive Exploration: Complex queries are decomposed into sub-questions and traversed across the graph until convergence.
  • Retrieval Intelligence: The system uses Q-value reranking and Hebbian learning to grow edges between notes retrieved together, effectively learning from its own usage patterns.

Who it’s for

Developers building AI agents who need long-term, portable memory that survives model changes or client switches, and those who want full ownership of their data without relying on cloud APIs.

Highlights

  • Zero Cloud Dependency: Uses local embeddings and SQLite for indexing, requiring no API keys for core functions.
  • Local-First Storage: Data is stored as human-readable Markdown files, making it Git-friendly and portable.
  • MCP Integration: Provides a Model Context Protocol (MCP) server with 16 tools for seamless integration with agents like Claude Code, Cursor, and Hermes.
  • High Performance: Outperforms incumbents on multi-hop retrieval benchmarks (HotpotQA) and matches top systems on long-term conversational memory (LoCoMo).
  • Token Efficiency: Dramatically reduces token costs by retrieving only relevant snippets instead of dumping entire vaults into the context window.

Related

  • Project
    MemTensor/MemOS

    MemOS 2.0 is a Memory Operating System for LLM‑based AI agents. It offers a unified API to store, retrieve, edit, and delete multi‑modal memories (text, images, tool traces) organized as a graph. The system can run as a hosted cloud service (API‑key access) or be self‑hosted via Docker (Neo4j + Qdrant) or a local SQLite plugin. Ready‑made plugins integrate MemOS with OpenClaw, Hermes, and DeepSeek Harness. Benchmarks show it outperforms existing commercial memory products. Licensed under Apache 2.0.

  • Project
    mnemon-dev/mnemon

    A persistent, LLM-supervised memory layer for AI agents that enables cross-session knowledge retention across multiple AI frameworks using a four-graph knowledge store.

  • Project
    doobidoo/mcp-memory-service

    A persistent, shared memory backend for AI agents that provides a knowledge graph and semantic search across sessions and tools via REST API and MCP.

  • Project
    ThinkfleetAI/memmesh

    A persistent, local-first memory store for AI agents that provides durable, typed, and searchable memory using a single Rust binary without requiring a hosted vector database.

  • Project
    agentscope-ai/ReMe

    A local-first, self-evolving personal knowledge base for AI agents that stores memory as human-readable Markdown files for long-term knowledge evolution.