Glossary
Agent memory
Agent memory is what an AI agent retains between sessions — facts, preferences and past decisions — stored outside the model rather than in its weights.
Outside the model
The key fact: memory lives beside the model, not inside it. The weights never change. That makes memory readable, editable and reversible — and it is why agent memory is not fine-tuning.
The layers
- Conversation — the current thread.
- Long-term — facts and preferences distilled from past work.
- Knowledge — documents the agent can consult.
- Skills — procedures it follows the same way each time.
Why it matters commercially
Memory is what stops you re-explaining your business every morning. It is also the thing most worth asking about when you leave a platform: what you can export, and in what form.
Related terms
FAQ
What is agent memory?
What an AI agent retains between sessions — facts, preferences and past decisions — stored outside the model rather than in its weights. Because it sits beside the model, it is readable, editable and reversible, which is why it is not the same as fine-tuning.
What are the layers of agent memory?
Four: the current conversation, long-term memory distilled from past work, a knowledge layer of documents the agent can consult, and skills that capture procedures it follows the same way each time.