Skip to main content

Tell the assistant about yourself once

Fill in Settings → User Instructions: what to call you, what you do, and how you like answers. These preferences apply to every chat, so you don’t repeat them. See Settings.

Pick the right model for the task

Fast, inexpensive models are fine for drafting and quick questions. Use a reasoning model for complex analysis, math or code. When you’re unsure, compare two or three models on the same prompt and keep the best answer.

Use projects for ongoing work

If you keep giving the same background in new chats, create a project. Its instructions, files and knowledge bases apply to every chat inside it.

Turn repeated prompts into tools

  • Prompts you reuse word-for-word → save them as prompts with {{variables}}.
  • A persona or role with specific tools → build an agent.
  • A fixed multi-step process → build a workflow.
  • Something that should happen on a schedule → create an automation.

Ground answers in your documents

For questions about your own material, add it to a knowledge base and mention it with @. Answers then cite the passages they came from, so you can check them.

Keep tool use under control

Use @mentions to tell the assistant exactly which agent, workflow or connector to use. If you want to approve each tool call before it runs, switch the tool mode to Manual. See Tools.

Write clear instructions

When writing instructions for an agent or project:
  • Say who the audience is and what a good answer looks like.
  • Give formats explicitly (“answer in a table with columns X, Y, Z”).
  • Say what to do when information is missing (“ask a clarifying question” or “say you don’t know”).
  • Keep them short. Long instructions are harder for models to follow consistently.