memorilabs/memori

Memori is a memory infrastructure layer for AI agents. An AI agent, in this context, is a software program powered by a large language model (an AI system trained to understand and generate text) that takes actions on your behalf, such as answering questions, writing code, or managing tasks. The problem Memori addresses is that these agents typically forget everything when a session ends. Each new conversation starts completely blank, which means users have to re-explain their preferences, context, and history every time. Memori fixes this by capturing what happens during agent sessions and storing it as structured, persistent memory that can be retrieved in future sessions. It works by wrapping the AI client your application already uses, so it intercepts conversations in the background without requiring you to change your core agent logic. When the same user returns, relevant memories are surfaced automatically, reducing the need for large piles of background context to be loaded into every request. The library is available as a Python package and a TypeScript package and connects to an API key from the Memori Cloud service. It also describes a self-hosted option called BYODB (bring your own database) for teams that want to keep memory data in their own infrastructure. Integrations exist for several AI frameworks and gateway tools: a plugin for something called OpenClaw, a provider for a tool called Hermes Agent, and an MCP connector that lets AI coding assistants like Claude Code or Cursor use Memori with a single command. According to the README, Memori was evaluated on a benchmark called LoCoMo that tests long-conversation memory and achieved 81.95% accuracy while using roughly 5% of the prompt size that full-context approaches require, meaning it answers questions from memory using far fewer tokens than would be needed to replay the entire conversation history each time.