OpenLTM – Local, self-decaying memory for AI coding agents (github.com)

OpenLTM, a new open-source project, introduces a local, self-decaying long-term memory component for AI coding agents such as Claude Code, OpenCode, and Pi. Initially developed to enhance codebase management for individual developers, OpenLTM is now fully available under an MIT license, emphasizing user ownership with a local SQLite database devoid of cloud dependencies or telemetry. This memory system automatically captures and recalls semantic information, allowing agents to learn from past sessions without manual input. Key features include structured decay of outdated memories, semantic search capabilities that prioritize meaning over keywords, and a user-friendly setup that requires minimal configuration. The significance of OpenLTM lies in its potential to enhance AI productivity and consistency by helping coding agents retain context across multiple sessions. With features such as importance-weighted memory decay, users can ensure that essential information remains accessible while irrelevant details fade away. Additionally, the inherent hackability of OpenLTM fosters community-driven innovation, as developers can modify and extend its capabilities through various hooks and skills. By providing an efficient, provider-agnostic solution for memory management in coding environments, OpenLTM positions itself as an essential tool for improving the efficacy of AI agents in software development.