# Memmy

> A personal AI memory hub and local agent runtime that gives every AI tool — Cursor, Claude Code, Codex, and more — one shared, fully controlled long-term memory.

Memmy is an open-source desktop app, CLI, and API that acts as a unified memory layer across AI coding and productivity agents. Built by MemTensor and released under the MIT license, it runs locally on macOS and Windows, keeping all memory data on the user's own device. The project reached v1.0.4 as of late July 2026 and is actively developed on GitHub.

## What It Is

Memmy solves a specific problem: every time you switch AI tools — from Cursor to Claude Code to Codex — you lose the context you built up. Memmy intercepts that loss by scanning your local AI conversation history, distilling it into structured long-term memory, and injecting the most relevant context into whichever agent you switch to next. The tagline "let every AI remember the same you" captures the core job: one memory, many agents.

## How the Memory Layer Works

Memmy's memory pipeline runs in three stages:

- **Scan** — after explicit user authorization, Memmy reads conversation history from supported agents (Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes Agent) stored locally on the device.
- **Organize** — a MemOS-powered memory engine distills scattered conversations into structured records: preferences, technical decisions, project background, and pitfalls, deduplicated and categorized.
- **Inject** — when the user switches to any AI tool, Memmy performs on-demand retrieval and passes only the most relevant memory, not a full data dump.

The memory service runs at `http://127.0.0.1:18960` by default and exposes session, turn, search, write, panel, and analytics APIs. Every entry point — desktop app, CLI, and OpenAI-compatible API — reads and writes the same memory store.

## Multiple Entry Points, One Runtime

Memmy is described in its documentation as an "AI Agent Runtime," not just a chat interface. It unifies long-term memory, agent execution, and tool connections in a single local runtime:

- **Desktop app** — GUI workbench for macOS 12+ and Windows 10+, with account mode (free trial tokens included) or BYOK (bring your own API key) mode.
- **`memmy` CLI / TUI** — interactive terminal chat, single-turn messages, and an OpenAI-compatible server (`memmy serve` on port 18990).
- **`memmy-memory` CLI** — a lightweight client for external agents and scripts to search, add, and retrieve memory entries from the local memory service.

## Tool and Ecosystem Connections

Beyond memory, Memmy functions as a general-purpose local agent. The README lists support for connecting messaging platforms (Telegram, Discord, WeChat, Feishu, DingTalk) and productivity tools (GitHub, Gmail, Notion, Slack, Jira). It also supports MCP (Model Context Protocol) and custom Skills for extending capabilities such as file handling, shell commands, web browsing, image generation, and task automation. A managed Chromium browser is bundled for local page inspection and visual verification.

## Local-First Architecture and Data Control

All memory, configuration, and app state are stored on the user's machine by default. The homepage states that memory data is "never uploaded to any remote server." Users control which agent can access each memory entry, how much context is passed, and can view or delete any memory at any time. The README notes that when the memory service is unavailable, Memmy reports an error explicitly rather than returning fabricated memories.

## Update: v1.0.4

The latest release is Memmy v1.0.4, published on July 30, 2026. The repository was created on July 16, 2026, making this a very recently launched project. The GitHub repository had accumulated 259 stars and 32 forks within roughly two weeks of creation. The roadmap points toward expanding memory sources beyond AI conversations to browser activity and local documents, and toward planned team collaboration features that would let multiple users' AI assistants share knowledge under privacy controls.

## Features
- Cross-agent shared memory layer
- Automatic scan of local AI conversation history
- MemOS-powered memory engine for distillation and retrieval
- Structured memory: preferences, decisions, project context, pitfalls
- On-demand context injection into any AI tool
- Desktop app for macOS and Windows
- CLI/TUI interactive agent runtime
- OpenAI-compatible API server
- BYOK (bring your own API key) mode
- Account mode with free trial tokens
- MCP (Model Context Protocol) support
- Custom Skills for extending agent capabilities
- Managed Chromium browser for local page inspection
- Messaging integrations: Telegram, Discord, WeChat, Feishu, DingTalk
- Productivity tool integrations: GitHub, Gmail, Notion, Slack, Jira
- Local-first architecture — no data uploaded to remote servers
- Per-memory access control
- History import from Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes Agent
- Personalized First Meeting Report after initial scan
- Multi-model and BYOK configuration

## Integrations
Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes Agent, Telegram, Discord, WeChat, Feishu, DingTalk, GitHub, Gmail, Notion, Slack, Jira, MCP (Model Context Protocol), OpenAI API

## Platforms
WINDOWS, MACOS, CLI, API

## Pricing
Open Source

## Version
v1.0.4

## Links
- Website: https://memmy.bot
- Documentation: https://memmy.bot/docs/
- Repository: https://github.com/MemTensor/memmy-agent
- EveryDev.ai: https://www.everydev.ai/tools/memmy