Memory staleness

A memory store keeps facts from past conversations and events. Facts change over time, such as a budget, a contact, or a plan. If the store keeps the old value and the agent reads it, the answer is out of date.

Staleness is a property of the stored data, not of the model. The model answers well from what it is given.

• The old and new values can both sit in memory with no marker showing which is current.
• A person may change a fact without saying so, so nothing tells the store to update it.
• Retrieval can surface the old value because it matches the question better.

• Your app sends a changed value with a newer timestamp, and that value becomes the current one. The earlier value stays answerable by date.
• On paid plans, situation awareness marks a fact the person has since said no longer holds, so the agent does not repeat it.

Not always. When you send the same id with changed text, Harny stores a new revision and deletes the memories built on the old revision. A changed value with a newer timestamp becomes the current value, and the earlier value stays answerable by date. When a person has said two things that cannot both be true, Harny keeps both and picks no winner.