Knowledge

When your instructions seem to be ignored

You opened Settings, found Chat instructions under Personalization, and described how you want the assistant to answer. A few messages later the answers look the same as before, or they follow some of your rules and skip others. Check first that your instructions are switched on.

To get to them, click your name at the bottom of the sidebar, or the gear at the end of that row, and pick Chat instructions. On a phone, open the sidebar first. You can also type /instructions at the very start of the message box and choose Chat instructions from the list.

The tab has four fields: Nickname, Occupation, Assistant traits and Anything more to add.

Nothing applies until you press Enable

Chat instructions start switched off. Until you turn them on, a cover over the form reads Chat instructions are disabled. above an Enable button. Pressing Enable saves at once, and a notice reading Custom instructions updated successfully confirms it. If all four fields are empty, enabling sends nothing.

When you change a field, a bar pinned to the bottom of the tab shows Unsaved changes with Cancel and Set instructions. Lexika stores nothing until you press Set instructions. If you close Settings or move to another tab first, Lexika drops your edits without a warning.

Disable, the red button at the top of the tab, makes Lexika stop sending your instructions. Your text stays saved behind the cover, ready for the next Enable. Disable also stores whatever is in the fields at that moment, so press Cancel first if you have edits you do not want kept.

What the assistant actually receives

The assistant never reads your fields as you typed them. After each save, a helper model turns all four fields into one third-person sentence about you. Lexika asks the helper for at most 40 words. The sentence starts with your nickname, or with "The user" if you left it empty. It uses neutral, factual wording, with no lists or quotes. That sentence goes with each message you send, alongside the standing instructions Lexika gives the assistant in every chat.

• A long rule list does not fit. Anything more to add takes up to 1,000 characters, but all of it shares that one short sentence with your name, occupation and traits. The helper may shorten rules, merge them or leave them out.
• Exact wording may not survive. Lexika tells the helper to use no quotes, so a template, a sign-off or a set phrase usually arrives described in other words, or not at all. Put wording that must be exact in your message.
• You cannot see the sentence. Settings shows your fields, never what the helper wrote from them.
• Writing the sentence costs you no stars. Lexika writes it once after each save, not on every message.

Where they apply, and where they do not

Your instructions belong to you, not to a workspace. The same set goes with every chat message you send, in every workspace.

When	Your instructions
You send a message in a chat	Used
You regenerate an answer or edit a question	Used, as they are now
Eco, Balanced or Premium chooses the model	Used for the answer. Lexika also weighs them when choosing, and tells that step not to follow requests for a particular model
You use Council or Agentic Research in Mini Apps	Not used
Another member sends a message in a workspace conversation	Theirs, not yours

Saving does not alter answers you already have. Your next message is the first to use the new version.

If you want a particular model, pick it in the model selector at the top of the chat. The Knowledge Base article on Eco, Balanced and Premium explains how Lexika chooses. In Council and Agentic Research, write your preferences into the question itself.

When several people write in one conversation

Your questions carry your instructions and a teammate's questions carry theirs, so the tone, the length and the name the assistant uses can change from one sender to the next.

The assistant also sees everyone's earlier messages without names, as if one person wrote them all. The only sentence about a person that it gets is the sender's. When something should hold for the whole group, say it in the conversation.

Filling in the fields

To add a trait, type it in the box that reads Add a trait and press Enter, Tab or Add, or click a suggestion such as friendly or concise under the box. Click a trait to remove it. Set instructions also saves a trait still sitting in the box.

Field	Limit
Nickname	50 characters
Occupation	100 characters
Assistant traits	10 traits, each up to 100 characters
Anything more to add	1,000 characters

The form has no counter and does not stop you at a limit. A save with a field over its limit fails with an error and stores nothing. The error does not say which field is too long, so check each one against the table, shorten it and press Set instructions again.

Getting a rule followed every time

• Put the rule in your message when it matters. The assistant reads your own message word for word. Lexika can condense older messages in a long conversation, so repeat the rule further on. The Knowledge Base article on context compaction explains when that happens.
• Check what came through. In a new chat with no Brain attached, ask the assistant what it knows about you. Its answer shows roughly what your sentence says, with nothing from earlier messages or a Brain mixed in.
• Keep project facts in a Brain. Chat instructions describe you and go with you everywhere. Facts that one line of work needs belong in a Brain. Any member can attach a Brain the workspace already has. Only the workspace Owner and people with the Admin role can save facts to a Brain, or create one on a plan that includes Brains.