rohitg00/agentmemory

#1 Persistent memory for AI coding agents based on real-world benchmarks

What it solves

Coding agents often suffer from "memory loss" between sessions, forcing users to repeatedly explain architectures, preferences, and bug fixes. While some agents use static files like .cursorrules, these are limited in size and quickly become outdated. agentmemory provides a persistent, cross-agent memory layer that captures and compresses session data so agents can recall specific details without manual re-explanation.

How it works

Built on the iii engine, the project runs a memory server that integrates with AI coding agents via hooks, the Model Context Protocol (MCP), or REST APIs. It silently captures agent activity, compresses it into searchable memory, and injects relevant context into new sessions. It supports hybrid search (BM25 and vector embeddings) to retrieve information, with an option for local on-device semantic recall using Xenova/all-MiniLM-L6-v2.

Who it’s for

Developers using AI coding assistants such as Claude Code, Cursor, GitHub Copilot CLI, Gemini CLI, and other MCP-compatible clients who want their agents to maintain a long-term memory of their project and preferences across different tools.

Highlights

  • Cross-Agent Compatibility: Works with a wide array of agents (Claude Code, Cursor, Devin, Aider, etc.) through a single shared memory server.
  • Hybrid Retrieval: Combines keyword search (BM25) and semantic vector search for high recall accuracy.
  • Zero External DBs: Operates without requiring external database installations.
  • Local Embedding Support: Offers free, on-device semantic recall to keep data private and local.
  • High Efficiency: Claims significant token savings by injecting only the necessary context rather than full session histories.

Related

  • Project
    tigerless-labs/agent-memory
  • Project
    TencentCloud/TencentDB-Agent-Memory

    TencentDB Agent Memory is an open‑source memory hub for LLM agents. It captures conversations, extracts reusable skills, builds a Wiki from docs, and indexes code into a CodeGraph. A central UI (Memory Hub) lets teams manage assets, set fine‑grained ACLs, and bind specific memory pieces to individual agents. Agents talk to a uniform proxy, so any framework (Claude Code, DeepSeek Harness, OpenClaw, etc.) can use the same memory without code changes. Benchmarks show a 59 % boost in persona retention. The repo provides one‑click Docker deployment, migration tools, and a roadmap for zero‑config cold starts and more integrations.

  • 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
    Lyellr88/marm-memory

    marm-memory is a local‑first memory server for AI agents that use the Model Context Protocol (MCP). It stores chats, notes, code‑graph data and extracted concepts in SQLite, exposing them via HTTP or STDIO so agents can retrieve relevant context instead of starting from scratch. The package includes a web console, Docker image, and CLI tools for starting, managing, and indexing repositories. Benchmarks show low‑latency retrieval (≈30 ms for 4 k memories) and hybrid FTS‑semantic recall that scales to 10 k entries. Ideal for developers building coding assistants or multi‑agent swarms who want a self‑hosted, searchable memory layer.

  • Project
    basicmachines-co/basic-memory

    A persistent, two-way memory system for AI assistants that uses shared Markdown files and the Model Context Protocol (MCP) to create a long-term knowledge graph.