# Memory in WireCat Desktop (/en/docs/desktop/memory)

Agents connected to WireCat save what they learn, such as "Kai prefers Friday calls", with the messages that show it.
The **Memory** page in WireCat Desktop is where you decide what counts. Nothing an agent suggests is used as true
until you approve it. You can also add to memory yourself: paste a note, give a link, a file or a picture, and keep
the facts you want. Memo keeps the memory; without it the page says so and offers to install it.

Memory is marked beta: you see a β beside its name. The section is new and lightly tested.

## How it works [#how-it-works]

A band at the top of the page shows the three steps:

1. **An agent suggests**, from a chat, an email, a document or a meeting.
2. **You check it:** see where it came from, edit it if needed, then approve or reject it.
3. **Agents use it:** what you approved, agents take as true when they answer you.

![The Memory page: the three steps, the tabs Waiting for you, In use, Rejected and All, and suggestions grouped by who they are about](/screenshots/desktop/docs/memory-light-en.webp)

## Review what waits for you [#review-what-waits-for-you]

The tabs are **Waiting for you**, **In use**, **Rejected** and **All**. Each row reads as a sentence, with a kind
word (Preference, Decision, Link…) and its status. Rows are grouped by who or what they are about, and each group has
**Approve all** and **Reject all**.

Click a row to open it. It shows **What agents will know**, **Where it came from**, the sources it rests on, and what
each choice changes:

- **Approve:** agents take it as true when they answer you.
- **Reject:** agents do not use it. It moves to **Rejected**, where you can still see it.
- **Edit:** change the text, its kind or who it is about, then **Save** or **Save and approve**. You can also click the
  row's text to edit it.

An agent's findings come as one row, for example the tasks, invoices, orders and new contacts it found in your mail.
**Review** opens the list. Each finding shows its kind, why the agent thinks so, and the email it read, which you can
open. Approve keeps only the findings you ticked.

## Take back a decision [#take-back-a-decision]

On a row in use, **More** has the undo that fits it: **Reject**, **No longer true** or **Undo approval**. **Undo
approval** puts a fact back into **Waiting for you**. A fact you wrote yourself has **Edit** and **Erase…** instead.

**Erase…** removes the row from this computer, and agents stop finding it. The messages, notes and meetings it rests on
stay, and a backup made before still holds it.

When the source of a row in use changes or goes away, the row is marked, for example "Source changed", and shows the
source when it was saved and now. Answer with **Still true**, **Edit** or **No longer true**.

## Links between people, companies and documents [#links-between-people-companies-and-documents]

A link says how two things relate, such as "Kai Moreno is the decision-maker of Northstar Consulting". Its role comes
from a list in your language; **Add a role** adds your own. A link to a document shows the document's title, folder,
last change and **Open**. A link that came from a source, such as your CRM, is marked **Reported**, and one that is
over is marked **Ended**. **Remove link** makes agents stop using it; the people and the document stay.

Memory lists only what you can decide here, your own links first. Whether two accounts are one person is decided on
the [Contacts](/llms.mdx/docs/desktop/contacts/content.md) page.

## Add to memory [#add-to-memory]

Click **Add to memory** at the top of the page.

![Add to memory: the choice of Text, Link or File, a title and a text field, and Save](/screenshots/desktop/docs/memory-add-light-en.webp)

1. Choose **Text** (paste a note, with an optional title), **Link** (a web page or a public document) or **File** (PDF,
   Word, text, a saved web page or a picture, up to 50 MB). Click **Save**. WireCat keeps it as a document you can
   search, even if you do not remember its exact words.
2. **Suggest facts** reads it and lists facts it found, and the people and organizations named in it. **Save without
   facts** stops here.
3. Tick the facts to keep, or **Tick all**, and click **Keep**. They go into memory as yours.

A picture is searchable by its contents only when an AI model reads it. Set one in
[AI model and better search](/llms.mdx/docs/desktop/ai-model/content.md); otherwise the picture is kept by its name. Before you save, the panel
says which model reads the picture and where it is sent.

## Next [#next]

Ask your agent something it should now know, such as "When does Kai prefer calls?". It answers from what you
approved.