> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neuralfactory.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Memory and context

> What an agent remembers within one task, what it keeps for later, and what it forgets.

An agent is an AI assistant you give tasks to in a conversation. It remembers two kinds of things: the conversation you are having with it now, and a few facts about you that it chose to save for later. Knowing which is which tells you when you need to repeat yourself.

## Within one task

A conversation with an agent is one task, from its first message to its last. Inside it, the agent sees what was said earlier, the files attached and produced in it, and its own instructions. You can follow up without repeating the question.

There is a limit to how much the model can consider at once, called its context window. A small ring in the composer shows how much of it the conversation uses (hover it to see the tokens used out of the maximum). It changes colour at 70% and turns red at 90%. When a long conversation fills it, older parts are condensed into a summary so the agent can keep going. The most recent messages stay whole, but details from early in a very long conversation can be lost. Start a new task when you change subject.

The agent can also keep short notes for the conversation while it works. They belong to that one conversation and are deleted with it.

## Across tasks

A new task starts without the earlier conversations. What carries over is the agent's memory: facts and preferences you shared that the agent decided to save, such as your role, the project you work on or the format you prefer for reports. In a later task, it can look them up and use them.

* Your memory belongs to you. Other people never see it.
* It is not tied to one agent. A fact saved while you talk to one agent can be recalled by another.
* The agent decides what to save. You can ask it to remember something, or to forget something it saved.
* Deleting a conversation removes its notes. Long-term memories the agent saved about you are kept.

The web app has no page that lists your memories. To find out what an agent knows about you, ask it.

## What an agent does not remember

* The content of your other conversations, apart from what it saved to memory.
* Anything from other people's conversations or memories.

## An example

On Monday you tell an agent: "I'm the finance lead for the Nordics, and I want reports as a short table." The agent saves both facts. On Thursday you start a new task and ask for last month's travel costs. The agent looks up what it saved about you, limits the report to the Nordics and answers with a short table. Saving and looking up are the agent's own decisions, so if it misses something, tell it again. It does not know what else you discussed on Monday; that stayed in Monday's conversation.

## Where you see this in the app

* The context ring in the composer. See [Starting a task](https://docs.neuralfactory.ai/user-guide/chat/starting-a-conversation.md).
* Deleting a conversation and what goes with it. See [Chat history](https://docs.neuralfactory.ai/user-guide/chat/chat-history.md).
* Abilities every agent has without a setting, memory among them. See [Built-in tools](https://docs.neuralfactory.ai/user-guide/toolkits/built-in-tools.md).

## Next

* [Subagents](https://docs.neuralfactory.ai/user-guide/concepts/subagents.md)
* [How an agent works](https://docs.neuralfactory.ai/user-guide/concepts/how-an-agent-works.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.