Docs

SeeMaps

On this page

How AI memory works

An AI model keeps nothing between requests: it sees only what is sent with each one. Apps like ChatGPT and Claude keep a memory of what seems useful from your chats. jade.os runs on rules instead: your AI writes down outcomes, decisions and new facts as it works, and reads your memory before it starts. Tap a box, or press Play.

The desk is everything your AI is given to read in one turn: its instructions, your memory, the chat so far and your message. AI makers call it the context, and its size limit is the context window.

The map

Your computer
Layer 1 · your AI
jade.os · layer 2
every turn
How full Session 1
InstructionsYour memoryThe chat so farYour message
  1. Each turn, jade.os sets your AI's desk: your memory, the chat so far, and your message.
  2. The whole desk is sent to your AI company, every turn. The model itself keeps nothing between requests.
  3. The model reads everything in its context window and writes the reply, which joins the chat.
  4. Every message makes the chat longer, so the window fills. A fuller window costs more each turn, and recall falls as it fills: context rot.
  5. jade.os's rules have your AI write down outcomes, decisions and new facts as it works, in today's log and on the right page for each business. When a fact changes, it replaces the old line.
  6. Before the window gets crowded, your AI writes itself a hand-off note: what you want, what was decided, what is still open, and anything it promised to do.
  7. A fresh session starts with an empty window. It reads your memory and the hand-off note first, then your latest messages, and carries on. You keep seeing one chat.
  8. The model itself starts every session empty. Memory brings it up to speed: in jade.os, your memory and the hand-off note go on the desk first.
The mapTap a step, or press Play.

Play follows one request through, a step at a time.

Why jade.os hands off early

A model can hold a very large window, up to about a million tokens depending on the model. But every turn sends the whole chat again, so the fuller it gets, the more each turn costs, and accuracy and recall fall as it fills, which is known as context rot.

So jade.os does not wait for the window to fill. Your AI writes a good hand-off note and a fresh session carries on from it, with your memory on the desk first.

Memory always lives outside the model. ChatGPT and Claude's apps save what seems useful from your chats and can look back through past chats, each with its own settings. Claude Code loads the CLAUDE.md files you write, and notes of its own.

jade.os is built on standing rules. As it works, your AI writes outcomes, decisions and new facts into your memory: today's log, and the right page for each business. When a fact changes, it replaces the old line, so your memory never contradicts itself. Rules you save come first, and when your memory does not cover something, your AI asks you once and saves the answer.

Every session reads your memory first: one memory, shared by every helper, on your computer.

Whether your chats may train future models is a separate privacy setting with your AI company. It is not part of memory.

Under the hood

  • The desk. Everything your AI is given to read in one turn: its instructions, your memory, the chat so far and your message. AI makers call it the context; its size limit is the context window.
  • The context window holds everything in one request: the instructions, the tools the model may use, the messages and tool results so far, and the reply being written. Current models hold up to about a million tokens, depending on the model.
  • Nothing is kept between requests. Each request sends the whole conversation again. The model itself remembers nothing.
  • Caching makes re-reading the same text cheaper, but not free. And the more a window holds, the more recall falls: context rot.
  • A jade.os hand-off note holds what you want, what was decided, what is still open, and anything your AI promised to do. The fresh session reads it and your latest messages, then carries on.
  • Your memory is written as your AI works: decisions, facts about your business, and what got done. Every session reads it first.

Read more

The ring beside every chat shows how full its window is. The guide Context window explains it.

Open Context window