AI Memory is an open-source long-term memory layer for AI coding agents. It gives agent CLIs a persistent place to store and recall what they learn about your projects, so context survives across sessions — and can even be handed off between different agent vendors. It has earned about 12,000 stars on GitHub since appearing in 2026.
Coding agents are forgetful by design: every new session starts from zero, and switching from one vendor’s CLI to another means re-teaching the new tool your codebase, conventions and decisions. AI Memory addresses that structural gap. It provides a shared, persistent memory that agent coding tools connect to, where accumulated project knowledge — architecture decisions, gotchas, preferences, work history — is stored in a form agents can query at the start of a task and update as they learn. Because the memory lives outside any single vendor’s product, it doubles as a handoff mechanism: the context built up while working with one agent CLI remains available when you continue the same project with another. The project is maintained by Fábio Akita (AkitaOnRails), a well-known figure in the Ruby and developer-tools community, and it reflects a pragmatic view of where agent tooling is heading: models and CLIs will keep changing, so the memory layer should be yours and stay put.
AI Memory is free and open source under the MIT license. You run it yourself and connect your own agent tools; there is no subscription or hosted tier.
AI Memory is for developers who work with coding agents daily and are tired of re-briefing them every session — especially anyone who switches between agent CLIs and wants one continuous, portable memory of their projects.