Meet your agents

An agent is an AI worker with a name, a face, instructions, skills, and access to certain files and apps. You already have one: “your name’s Assistant”. Teams often add agents for particular jobs, such as a research agent or a reporting agent.

Find an agent

Click Agents on the rail. The views are All agents, Made by you, Shared with you, In your workspaces and Starting points.

Click an agent to open its page. The tabs are:

• Setup: its model, instructions, skills, files and connectors;
• Platforms: other places you can reach it, such as email or Slack;
• Activity: what it has done recently;
• Memory: the notes it keeps for itself;
• Access: where it works and who it is shared with;
• Permissions: what it may do on its own and what needs your approval;
• Evaluations: the tests that grade it, if you can open Evals.

Read an agent’s Setup before you give it a big job. Its instructions show how it will go about the work.

Pick one for a task

In the composer, click the agent chip and pick the agent. Each agent has different access, so check the chip before you press Enter. Use your own assistant until you have a reason to switch. A team agent is worth setting up once the same kind of job comes up every week.

Start from a ready-made agent

Starting points has about 100 ready-made agents for common jobs. Use Filter to narrow them by industry, function or focus, and click Use to start a new agent from one. It’s quicker to edit good instructions than to write them from scratch.

To start from a blank agent, click New agent in the Agents panel, or press ⌘K and type “Create agent”. Build an agent for your team covers the whole setup. On most organizations Members can’t create agents by default. If you can’t see New agent, ask an admin to allow it or to create the agent for you.

On newer organizations. With the newer permissions system, Members can create agents by default. An agent also works with its own access, so the person who asks it doesn’t change what it can reach.