memtensor/memmy-agent

Memmy is a personal memory hub paired with an AI agent that gets to know you over time. The idea is that every AI coding session builds up context, but most of that context normally gets thrown away the moment you close a tab or switch tools. Memmy solves this by giving all your AI agents, such as Cursor, Claude Code, Codex, and OpenClaw, one shared long term memory layer, so a conversation you had in one tool can carry over into another instead of starting from scratch each time. At its core, Memmy runs a memory engine that automatically collects, understands, and organizes your knowledge, preferences, and past work into searchable long term memory. It can also import the history from AI agents you already use, turning old conversations and project experience into a growing personal knowledge base within minutes of installing it. Beyond memory, Memmy works as a local agent runtime that you can reach through a desktop app, a command line or terminal interface, or an OpenAI compatible API, all sharing the same agents, memory, and settings. Tasks you start in one entry point can continue seamlessly in another. It can also connect to everyday tools and messaging apps such as Telegram, Discord, WeChat, GitHub, Gmail, Notion, Slack, and Jira, and it supports extending its abilities through MCP and custom Skills for things like file handling, running shell commands, web access, and image generation. Memmy is built local first, meaning your memory, settings, and app data are stored on your own machine by default rather than uploaded to the cloud, with access controls limiting who or what can use your stored memory. If the memory service is ever unavailable, Memmy is designed to report the error plainly rather than making up a fake memory. New users get free trial tokens with automatic model routing to start using the memory and agent runtime right away, and once those run out you can bring your own API key for whichever AI model you prefer. The full README is longer than what was shown.