---
id: con-memory
schema: concept/1
layer: garden
status: established
name: Memory
aka: ["ChatGPT memory", "Claude memory", "personal context", "chat history reference", "saved memories"]
summary: >
  A feature that lets an AI carry details about you and your past conversations into new ones, so it
  can personalise answers without you repeating yourself.
first_seen: "2024-02"
relations:
  builds_on: [con-context-window]
  evolved_from: [con-custom-instructions]
  related: [con-privacy-and-data]
  parallels:
    - idea: "a client record in a CRM"
      field: "sales"
      note: "Notes from past calls mean the next conversation picks up where the last one ended."
    - idea: "long-term memory"
      field: "psychology"
implementations:
  - { provider: openai, name: "ChatGPT memory" }
  - { provider: anthropic, name: "Claude memory" }
  - { provider: google, name: "Gemini personal context" }
  - { provider: xai, name: "Grok memory" }
competencies: [cmp-use-responsibly, cmp-describe-the-task]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "Shows the move from context you write (custom instructions) to context the AI collects, with new privacy trade-offs."
---

Each conversation with an AI used to start blank. With memory, the AI notes things worth keeping, such as your role, your projects or preferences you mentioned, and may also look back at earlier chats. Next time, it uses them.

This is the automatic version of custom instructions. Instead of you writing a profile, the AI builds one from what you say. That is convenient, and it also means the AI can remember things you didn't intend it to keep, or carry an outdated detail forward ("you're planning the spring gala" months after it happened).

Good practice is to look at what your tool remembers every so often, delete what's wrong or sensitive, and use a temporary or incognito chat for anything you don't want carried forward.

## Example

A consultant mentions once that she prefers tables to long paragraphs. A week later, a comparison arrives as a table without her asking. She also finds the AI remembered a client's name from a confidential chat, and deletes that memory.

## How it connects

- It **evolved from custom instructions**: saved context, now gathered automatically.
- It works around the limits of a single **context window**.
- It raises **privacy and data** questions about what gets kept.

## Try it

Write down what you think your AI tool remembers about you. Then ask it, "What do you remember about me?" or open its memory settings. Delete one thing that's wrong or that you'd rather it forgot.