Agent Memory
The relationship layer of the Group e-media information AI stack. Humans adapt how they communicate to each person they know — same person, different register for a colleague, a client, a family member. Agents need the same.
Production AI is not a prompt. It is a system of context, tools, permissions, traces, evals, and feedback loops — including memory that belongs to the user it serves.
How to think about it
Memory is owned by the user. It reflects what they want to see, how they want to be addressed, what context they don't want to repeat. Working, episodic, semantic, and organizational memory with explicit lifecycle controls — write, decay, retrieve, retire. Different users get different relationships with the same agent. The agent's foundation does not change to accommodate any of them.
Where it fits
Sits on top of synthetic personality, beneath closed-loop knowledge. The personality is invariant; the memory is per-user. What the agent learns from all users about how to do its job better belongs to the next layer.
Related resources
The foundation layer of the Group e-media information AI stack. Before an agent can serve anyone, it has to be something — a coherent identity that holds up under pressure, across sessions, and across the very different people it will meet.
The growth layer of the Group e-media information AI stack. Humans get better by reflecting on what they did, what worked, what failed — and carrying those lessons forward. Agents need the same loop, but at the system level.
Retrieval pipelines that combine chunking, embeddings, metadata, reranking, permissions, and evaluation.