Knowledge

There is no Add button

You opened a Brain, it said Nothing remembered yet, and there was nowhere to type. Or you chatted with a Brain attached and found it empty afterwards. A Brain saves nothing on its own, and its page has no way to add an entry. The New brain dialog says so in its To save something here, ask the assistant box.

An entry gets in one of two ways:

• You ask. In a chat with the Brain attached, say something like "Remember that we deploy on Fridays."
• You accept an offer. When you state a lasting fact or decision, the assistant may ask something like "Should I save this to your Brain?" Lexika tells it to ask once per fact and to save only after a yes.

Saving also needs all of these:

• The Brain is attached to the conversation, as the Knowledge Base article on which conversations use a Brain explains.
• The person who sent the message is the workspace Owner or has the Admin role.
• The model can work with a Brain. Perplexity models and image models can't, and the model selector doesn't mark which others can't. If the assistant says it can't save with the current model, switch models and ask again.

The assistant writes each entry in its own words, as a short title and a self-contained body. Given several facts at once, it is told to list them first and wait for your go-ahead.

Reading the save row

A save shows as a row inside the reply, at the point where the assistant saved. It reads Saving to Brain while it runs, then becomes a receipt that opens the Brain, at the entry itself when exactly one was saved. Open any save you didn't ask for.

The row reads	What it means
Saved 1 entry to Product Knowledge	The entry is in the Brain.
Includes 1 duplicate skipped	A very similar entry exists, so this one wasn't saved. The assistant should ask whether to replace the old one or keep both.
Saved 0 entries to Product Knowledge	With no duplicate mentioned, the Brain is usually full.
Saving to Brain was interrupted	The reply stopped mid-save. Check the Brain before asking again.
Couldn't save to Brain	The save was refused. The reply usually says why.

A receipt for a single entry can later gain a count such as · 2 links, once Lexika links the entry to related ones. Linking happens only at save time, never after an edit.

Keeping both saves the new entry beside the old one. Replacing archives the old entry. It no longer shows in the Brain page's list or map, or in the assistant's searches, and stops counting toward your limit, but Lexika keeps it. The new entry shows Replaces: and the old title, which opens the old version. That version reads This entry is archived and read-only. and has no Edit or Delete button. When you ask the assistant to change a saved fact, it is told to replace the entry this way, so the old wording stays. To fix wording without keeping the old version, use Edit on the Brain page instead.

Checking the Brain page

Open a Brain from the Brains section of the sidebar, or click a save receipt. Search entries narrows its entry list, and each entry's Open chat link returns to the conversation it came from. Owners and admins see Edit and Delete at the bottom of an open entry.

Forgetting is permanent

In a chat, say something like "Forget what I said about pricing." The same conditions as saving apply. The assistant is told to find the matching entries, show you their exact titles and delete only after you confirm that list. Even "clean up the Brain" should only get you a proposed list. Lexika itself only checks that each title matches the entry it deletes, so read the list before you say yes.

The row reads Deleting from Brain, then, for example, Deleted 2 entries from Product Knowledge beside a red Brain mark. A trailing · 1 failed means an entry had changed or was gone. Couldn't delete from Brain means the request failed, so check the Brain before asking again.

On the Brain page, open the entry, choose Delete and confirm in the Delete entry dialog.

Either way the entry and its links are gone, with no undo. Some things stay behind:

• The conversation where you said it, and any reply that quoted the entry.
• Any Searched the Brain or Read Brain entries row that found the entry. Opened, it still shows the title and opening text.
• An older version the entry replaced, stored until the whole Brain is deleted. The assistant can reach it only through the newer entry, so ask it to forget both together.
• The What's inside summary under the Brain's name, which the assistant also reads. Lexika rewrites it about five minutes after the assistant's last save or delete, but not after a change on the Brain page or once the Brain is empty. To get a removed fact out of it, let the assistant do the forgetting.

Editing, regenerating or deleting a conversation doesn't remove an entry it saved.

To remove a whole Brain, an owner or admin chooses Delete in the three-dot menu beside its name and confirms in the Delete brain dialog. Every entry goes, and every chat it was attached to loses it, though old save receipts can remain there.

Who can save and forget

Only the workspace Owner and people with the Admin role can save, edit or forget. What counts is who sent the message. In a workspace conversation that several members write in, an admin's "remember this" saves and a member's identical request does not. Members see no Edit, Delete or three-dot menu on the Brain page. The Created by line under a Brain's name only records who made it, so a creator who later becomes a Member can no longer change it.

Every Member can read every Brain in the workspace, so save only what the whole workspace may see. A preference that is yours alone belongs in Chat instructions, under Personalization in Settings.

When a Brain is full

A plan caps the entries in one Brain, and the count under the Brain's name shows yours against that cap. When the Brain is full the count turns amber, with a note saying the assistant can't save new memories until entries are deleted. A save then gets the Saved 0 entries receipt. The assistant is told the Brain is full, so its reply should say so. If the Brain fills up partway through a save, the receipt counts only what got in.

The assistant is told never to delete entries to make room. Delete some yourself, or ask it to forget specific ones. Each deletion frees one place. Replacing an entry still works in a full Brain, since the old version stops counting.

The count already uses the cap that applies to you, which depends on the workspace's credit setting, covered in the Knowledge Base article on Owner Credit and Individual Credit. The Pricing page lists each plan's Entries per Brain and Brains per workspace.