# Memori

An LLM-agnostic memory layer that turns an AI agent's conversation and tool use into structured, persistent state, so applications can recall what an agent did in past sessions instead of re-sending full conversation history.

Memori is agent-native memory infrastructure: an LLM-agnostic layer that turns agent execution and conversation into structured, persistent state for production systems. It works with the datastore and framework a team already runs, and deploys to a managed cloud, a single-tenant cloud, a private VPC, or on premises.

TypeScript and Python SDKs connect an application to Memori Cloud or a self-hosted database, and a dashboard exposes stored memories, analytics, a playground and API key management. On the LoCoMo long-conversation memory benchmark, Memori reported 87% overall accuracy using an average of 721 tokens per query, a small fraction of a full-context prompt, while using less of the context budget than comparable retrieval-based memory systems.

A plugin adds the same structured, persistent memory to OpenClaw agents, automatically capturing tool calls, decisions and outcomes after each turn without changes to agent code or prompts.

Website: https://memorilabs.ai