aa-ma-forge

agentic-ai
llm
memory
open-source
Advanced Agentic Memory Architecture. Long-term, structured memory for LLM agents.
Published

January 15, 2026

What it is

aa-ma-forge (AA-MA: Advanced Agentic Memory Architecture) is an in-progress framework for giving LLM agents structured, persistent, queryable memory, beyond a vector store.

The goal: agents that genuinely accumulate knowledge across sessions, with clean separation between episodic, semantic, and procedural memory.

Status

Active R&D. See the repo for the current state.

Why it matters

Most “agent memory” today is just RAG with extra steps. Real agentic systems need memory that supports recall, consolidation, forgetting, and reasoning over what was learned.

→ Repository on GitHub