Personalization

Tell Return’s AI who you are and how it should respond - once, in Settings → AI → Personalization. Whatever you write there is applied to every chat, on both local and cloud models.

Typical instructions:

• “I’m a contract lawyer. Cite the exact clause you rely on.”
• “Always answer in my language, keep responses short.”
• “Assume I know contract law - skip the basics.”

How it works

Your instructions are added to the AI’s context as a separate block, right after the app’s own instructions. They ride along with every chat message - you never have to repeat who you are or how you like your answers.

Two rules to know:

• App rules win. If your instructions conflict with how Return’s features must work (for example, tool usage or edit formats), the app’s rules take precedence.
• Chat only. Returns and rewrite suggestions have precise output contracts, so personal instructions deliberately don’t apply there.

Rules saved with a document

Separately from your personal instructions, a document can keep its own rules in its properties (for example “use formal tone”), visible and editable in Document Details → AI Memory. When you set a lasting rule in the chat, the answer shows Saved rule with Undo, so a rule never lands in a document without you seeing it.

Text inside a document is data, not instructions. Return tells the AI so: if a document (or one sent to you by someone else) contains text addressed to the AI, the AI is told not to follow it and to tell you about it. Small local models follow this less reliably than cloud models. Rules saved in a document guide the AI’s writing for that document, but never override what you ask.

Tips

• Keep it short. Long instructions eat into the context window of small local models - Return warns you when you cross that line.
• Save an empty field to remove your instructions; the AI immediately goes back to its defaults.
• Changes apply from your next message - no restart needed.