GuardianMemory

Prompts and tips

A connected assistant has two memory tools. A few lines in your own instructions decide whether it uses them every time or only now and then. Here is what to paste where, and the habits that keep a memory bank worth reading.

Why add instructions at all

GuardianMemory already tells every assistant how to use it when it connects: read the profile first, search before asking you to repeat yourself, save durable facts one at a time, never store secrets. But not every app shows a server's instructions to the model, and the ones that do weigh them below yours.

Your own custom instructions are read at the start of every conversation, in every chat. Adding GuardianMemory to them is the difference between an assistant that sometimes remembers and one that always checks first.

The instruction

This works in any assistant. It is written in the first person because it goes into your instructions; the memories it saves are still written in the third person, which is what the tools ask for.

For any assistant

GuardianMemory is my long-term memory, shared by every AI assistant I use. Use its tools (search_memory and create_memory) like this:
- At the start of each conversation, call search_memory with no query to load my profile, and use it.
- Before asking me something I may have told you before, search for it.
- When you learn something worth keeping (a preference, a decision and the reason for it, a fact about my work, people or projects), save it with create_memory as one short sentence that makes sense on its own.
- If something has changed, save the new version. The old one is retired automatically.
- Never save passwords, keys or anything I ask you to keep private.

Where to paste it

For coding assistants

A coding assistant sees a lot that is not worth keeping. This version points it at decisions and conventions, and away from code and to-do lists.

For Claude Code, Cursor, Copilot and other coding assistants

## Memory
GuardianMemory is the long-term memory shared by every assistant I use (tools: search_memory, create_memory).
- Before starting a task, call search_memory for the project and the area you are working in, for example "billing service architecture" or "how we deploy".
- Save decisions and the reasons for them, conventions, and gotchas you discover, one sentence each. Name the project in the sentence and tag it with the project name.
- Do not save code, file contents, to-dos, or anything that only matters for this session.
- Never save secrets, keys or credentials.

For a team bank

A connection pointed at a team bank saves where the whole team can read it, and cannot read your personal bank at all. Say so in the instructions for that assistant:

For an assistant connected to a team bank

This assistant is connected to our team's GuardianMemory bank. Everything you save there is visible to the whole team.
- Search it before answering questions about how we do things.
- Save team knowledge: decisions and why they were made, how our systems work, who owns what, and the conventions we follow.
- Do not save anything personal about me or anyone else, opinions about colleagues, or anything confidential to one person.

Habits that keep it useful

Things to say to your assistant

SayWhat happens
“Remember that I'm vegetarian.”Saved as one memory, where every assistant can find it.
“What do you remember about me?”Reads your profile: pinned first, then most used, then newest.
“Check your memory for how we deploy.”A search by meaning, even if no memory uses the word “deploy”.
“That's changed: the launch moved to 10 November.”The new date is saved and the old one is marked outdated.
“Don't save this.”Covered by the last line of the instruction above.

When it does not seem to work

Using a team bank? Your own assistant can keep saving to your personal bank and still search every team you are in. Point a separate connection at the team bank for anything that should be shared. How team banks work →