A memory template for an agent

A structure that separates stable memory from memory that goes stale, with source, date, scope and a refresh check.

What you get

A memory template that prevents old confidence

Who it is for

For anyone who wants an agent to remember preferences and decisions without recycling old facts as if they were fresh.

1. Two kinds of memory

Stable: working style, preferences, rules, boundaries.
Goes stale: status, prices, versions, tokens, dates, people in a role.

2. Every memory needs four fields

Source: where it came from.
Scope: which project or context it holds true for.
Date: when it was written or confirmed.
Refresh: when it must be checked again.

3. Save template

Copy box

Memory: [what to remember]
Scope: [project / person / process]
Type: stable / goes stale
Source: [conversation / file / decision / check]
Date: YYYY-MM-DD
Refresh needed when:
- [condition]

How to use it:
- [when this helps]

What not to infer from it:
- [boundary]

4. Before an expensive action

Check whether the memory has gone stale.
Check the date.
Open the source if there is any risk.
Tell the user when acting on unverified memory.

How to use this now

Good memory is a hint with a source. Not a verdict.

Explore

Tools