Agent_Memory

Research notes, paper summaries, architectural analysis, experiments, and original proposals on memory for AI agents.

This repository studies agent memory across the full stack: from LLM inference-time memory such as KV cache, to in-context memory, retrieval-augmented generation, long-term episodic memory, semantic memory, procedural memory, reflection, shared multi-agent memory, and self-maintaining memory systems.

Core Question

How should AI agents remember, forget, retrieve, update, verify, and use information over long periods of time?

The working hypothesis of this repo is that agent memory is becoming less like a simple vector database and more like a managed cognitive operating system: layered, structured, self-editing, security-aware, and optimized across latency, cost, reliability, and trust.

Current Research Hub

All long-form research artifacts live under Research/. Start there for the organized reading path.