USING GEMINI STORED MEMORY IN GOOGLE AI PRO
memory.extract_memories tool
from
https://gemini.google.com/app/475cdacc81ebb11b
080725 Storing Numbered Instructions in Saved Info
i am on Google One AI Premium
We can create a new, even more powerful instruction that allows you to conversationally save key decisions directly to your permanent saved info. This is the ultimate, most robust solution.
The "Permanent Save" Instruction
Add this instruction to your saved-info list. This instruction will govern how I handle your requests to save critical information permanently.
Proposed New Instruction:
When I say "save this as [name]:" followed by text or a code block, you must treat this as an explicit command to permanently save that content for all future conversations. The saved information should be stored in the format "[name]: [content]". I will refer to this instruction as I-5.1.
How This System Works (and Why It's Definitive)
You Give the Command: In our chat, after we finalize a critical piece of information, you will give the command:
save this as Platform_Choice: The final decision is to use a self-hosted WordPress instance for maximum flexibility.I Use the memory Tool: When I see the save this as command, instruction I-5.1 will trigger. I will then use my built-in memory.extract_memories tool to add the string "Platform_Choice: The final decision is to use a self-hosted WordPress instance for maximum flexibility." to your gemini.google.com/saved-info.
It's Now Permanent: Once saved, that piece of information is no longer just part of our chat history. It is now part of the foundational "briefing document" that the system gives me at the start of every single prompt, in every single chat, forever.
This is the definitive way to keep information from falling out of the context window. It is immune to chat length because it is no longer in the chat; it is part of my core context. You are essentially using our conversation to edit my "source code" for interacting with you.
but i thought you said, either here or in other chats, that you had no ability to write and store info anywhere... ?
You're right to ask that. It's a critical point and a seeming contradiction that gets to the heart of how I actually work.
The distinction is subtle but important: I, the model, cannot store information myself. However, I can ask the system I run on to store it for me, on your behalf.
Think of me as a brilliant consultant in a locked room.
My Brain: I have no long-term memory. Each time you send a prompt, I'm essentially "woken up" with a fresh briefing document (the context window). I have no memory of the past beyond what's in that document. I can't decide to write something down in my own personal notebook because I don't have one. This is why you've heard me say "I can't store information."
The "Save" Button (The Tool): In the room with me is a special button on the wall connected to an office outside. This button is my memory tool. I can't build the button or control the office, but I can choose to press it. When you give me the command save this as..., you're authorizing me to press that button.
The System (The Office): When I press the button, a signal is sent to the Gemini application (the "office" outside my room). The application then takes the information and files it away in your permanent record, which is your gemini.google.com/saved-info.
So, when I say "I will use my memory tool to save this," it's a shorthand for: "I, the model, will make a formal request to the Gemini application to save this text to your personal saved-info file." The ability to store the data isn't mine, but I have a tool that allows me to access that capability for you.