zaydmulani09/mnemo

Mnemo is a memory layer for AI language model applications, written in Rust. It solves a specific problem: AI chat sessions normally forget everything when they end. Mnemo runs as a background service alongside your application, listens for text you feed it, and builds a persistent record of entities and their relationships so that future conversations can reference things learned in earlier ones. Here is how it works in practice. Your application sends text to mnemo's ingest endpoint, such as a message from a chat session or a document. Mnemo sends that text to a language model of your choice, which extracts named entities (people, tools, places, concepts) and the connections between them. Those entities are stored in a local SQLite database along with a graph structure that tracks relationships. When your application needs context for a new prompt, it calls mnemo's retrieve endpoint with a query. Mnemo searches the stored text, finds matching entities, traverses the relationship graph to pull in related information, ranks everything by relevance, and returns a ready-to-use context string you can paste into your model's system prompt. Mnemo works with Ollama (a tool for running AI models entirely on your own computer, at no cost), OpenAI, Anthropic, or any API that follows the OpenAI format. No data ever leaves your machine unless you connect it to a cloud API yourself. The whole service ships as a single compiled binary with no Python or external runtime required, and it is designed to respond in under 50 milliseconds. Three ways to use it are provided: Docker with Ollama for a fully local, free setup, installing the Rust binary directly and pointing it at whichever AI service you already use, and a Python SDK for developers who want to call mnemo from Python code without managing HTTP requests manually. A command-line tool is also included for storing and searching memories, listing entities, and inspecting the knowledge graph from a terminal. Mnemo is aimed at developers building custom AI pipelines who need persistent, structured, local memory they control completely. It is not designed for use with managed AI platforms that already handle memory for you. The project is MIT-licensed. The full README is longer than what was shown.