Lyellr88/marm-memory

Local-first 3-in-1 AI memory layer & MCP server for Claude Code, Codex, Grok, Gemini, VS Code and Cursor. Fuses session history, codebase indexing & concept graphs in SQLite. Enables zero-cloud, privacy-first context & instant recall, supports multi-agent swarms.

Explanation

What the project ?

marm-memory is a local‑first memory server for AI agents that follow the Model Context Protocol (MCP). It stores chat logs, notes, code‑graph information and extracted concepts in an embedded SQLite database and makes them searchable via a small HTTP or STDIO API. Agents can query the server to retrieve relevant memories, code symbols or conceptual links instead of starting each conversation from a blank slate.

Key pieces

Component What it does Why it matters
Core Memory (8 tools) Stores sessions, notebooks, summaries, structured logs. Gives agents a persistent, searchable history.
Code Graph (6 tools) Indexes a local repository, extracts symbols, call‑graphs, architecture views. Lets agents understand the code base without re‑reading files.
Concept Graph (2 tools) Extracts entities/relationships from memories and links them to code. Connects decisions, errors, people across chats.
MCP surface HTTP (/mcp) or STDIO transport compatible with any MCP‑enabled client (Claude, Cursor, Codex, etc.). Works with any existing AI‑coding assistant that supports MCP.
Console UI Small web app (no Node.js) for browsing memories, graphs, diagnostics and running a terminal. Provides a human‑friendly control plane.
Deployment options pip install, Docker image, or plain STDIO daemon; profiles for solo, swarm, trusted setups. Flexible for a single researcher or a multi‑agent swarm.

How you use it

  1. pip install marm-mcp-server
  2. marm-memory init --g-claude (or other --g‑* flags) to install the MCP skill into your agent.
  3. Start the server – e.g. marm-memory fast-start-http – which also opens the console in a browser.
  4. Tell the agent to use the marm‑init skill; the agent will then call the MCP endpoints to store and retrieve memories automatically.
  5. Optionally index a repo with marm-memory projects index <path> to build the code graph.

Performance highlights (from the README)

  • Retrieval latency stays under ~30 ms even with 4 000 stored memories.
  • Hybrid recall (FTS pre‑filter + semantic rerank) is 4–6× faster than a full vector scan at 10 000 memories.
  • Write‑time consolidation (deduplication, clustering) adds ~9× cost but improves long‑term recall quality.
  • Benchmarks show a top‑5 evidence‑hit rate of ~69 % on the LoCoMo evaluation set.

Who it’s for

  • Developers or researchers building coding assistants that need a private, persistent memory store.
  • Teams running multiple agents that should share the same context (e.g., a “swarm” of Claude, Cursor, etc.).
  • Anyone who wants a self‑hosted alternative to cloud memory services like Mem0 or Zep, with built‑in code‑graph awareness.

Where to find more

  • Docker Hub: lyellr88/marm-mcp-server
  • PyPI: marm-mcp-server
  • Discord community for setup help and workflow sharing.
  • Full documentation sections are linked from the README (Benchmarks, Architecture, Troubleshooting, etc.).

All details are taken directly from the repository’s README; no external assumptions have been added.

Related

  • Project
    CodeAbra/iai-personal-memory-engine

    iai-personal-memory-engine is a local server that gives AI coding assistants a durable, verbatim memory of every conversation. It runs entirely on your machine, encrypts at rest, and automatically serves relevant context to keep your agent informed across sessions.

  • Project
    Gentleman-Programming/engram

    A persistent, project-aware memory layer for AI coding agents that prevents them from forgetting project context across sessions using a local SQLite database and MCP.

  • Project
    agentic-box/memora

    An MCP memory layer for AI agents that provides persistent collective memory through structured storage, semantic retrieval, and a knowledge graph.

  • Project
    doobidoo/mcp-memory-service

    A self‑hosted memory backend for LLM agents that provides a REST API, CLI, OAuth, and a web dashboard. It stores facts, decisions, and causal knowledge graphs locally (SQLite + ONNX embeddings) or in hybrid/cloud mode, offering fast (≈5 ms) semantic search, tag‑scoped retrieval, and optional LLM‑generated summaries. Works with LangGraph, CrewAI, AutoGen, Claude Desktop, ChatGPT, and many other tools.

  • Project
    supermemoryai/supermemory-mcp

    A Model Context Protocol (MCP) server that provides a universal memory layer, allowing users to carry their personal memories and context across different LLMs and AI clients.