Personas

A persona is a saved instruction attached to a chat. Instead of opening with three paragraphs explaining that you want code review focused on real bugs rather than style opinions, you pick the Code Reviewer persona and start with your code.

Using one

Open a chat, click the persona badge in the header, and choose one. Everything you send from then on carries that instruction. Switching persona mid-chat is allowed and does not erase the conversation.

The five built-in personas are aimed at work people actually bring here — content and SEO writing, code review, explaining a trading setup, study and research, and summarising data. You can read exactly what each one does, including its full instruction text, on its own page: they are linked from the personas index.

The instruction is not hidden

Every built-in persona shows you the exact system prompt it carries. If you disagree with a line, duplicate it and change that line. A persona whose instruction you cannot read is a persona you cannot trust to be doing what its name says.

Making your own

Duplicate a built-in one and edit it, or start from nothing. A useful persona is mostly about what to avoid: "do not rewrite my whole file", "do not invent statistics", "ask before assuming which framework I mean". Positive instructions get followed loosely; prohibitions get followed closely.

What a persona is not

It is not memory. A persona does not remember previous conversations, and it does not know facts about you that you have not put in it. If you want the assistant to know your product's name, put the product's name in the persona.

It is also not a filter on the answer. A persona shapes how the assistant approaches a task; it cannot guarantee the assistant never says something outside its remit.