What is lethyx?

Lethyx is a memory system meant to be plugged into AI agent frameworks so that an agent can remember useful information over time instead of starting fresh in every conversation. Rather than building its own full agent platform, Lethyx focuses only on the memory part: storing pieces of information an agent learns, deciding what is worth keeping, linking related memories together, and retrieving the right pieces later when they are needed. It is designed to work alongside existing agent tools rather than replace them. The README mentions it can plug into frameworks like LangGraph, into MCP based servers such as Hermes, or into a REST service or a plain Python agent you have written yourself. Developers install it as a small Python package and interact with it through a client that offers simple actions: remember something, recall relevant memories for a task, forget something no longer needed, and build a package of context to feed into an agent's prompt. Under the hood, memories can be stored using one of a few different storage backends, referred to in the README as mempalace, hindsight, and mnemosyne, and a project can choose which one to use through an environment variable or a setting when creating the memory client. For teams building with Hermes specifically, Lethyx can run as its own small server that Hermes talks to over the MCP protocol, exposing tools like remember, recall, and forget that Hermes can call directly, and it can also be installed as a native memory provider for Hermes. The README describes the project as having recently finished separating itself out into its own standalone package, with a public client and a LangGraph style adapter already working, while further release work is still tracked separately. This makes it a piece of infrastructure aimed at developers already building agent systems who need a dedicated memory layer rather than something meant for a general audience.