Skip to main content

Chatting with an Agent

Chatting with an Agent is just like a normal chat, except the Agent's instructions shape the whole conversation and its Skills are the ones available to run.

Selecting an Agent​

There are two ways to start chatting with an Agent:

  • From a new chat: Agents marked as Featured appear as cards beneath the chat input on a new, empty chat. Selecting one shows its recommended prompts first. Pick a prompt, or choose Ask anything to start chatting with that Agent in your own words. A Featured Agent with no recommended prompts goes straight to Ask anything.
  • From the Agents page: click Tools in the left navigation to browse all of your organization's Agents, featured or not, and select the one you want to chat with.

An Agent's featured status is set on the Agent itself. See Creating & Managing Agents.

Once selected, a Chatting with [agent name] tag appears below the chat box. You choose an Agent at the start of a conversation: before your first message you can remove the tag with one click to return to a standard chat, but once the conversation is underway it stays with that Agent. Start a new chat to talk to a different Agent (or none).

Agents require the View Agents permission.

If the Agent has Recommended Prompts, they appear as buttons after you select it. Click a prompt's tagline to pre-fill the chat input with its full prompt text. You can edit the pre-filled text before sending, or send it as-is to kick off the conversation.

How an Agent's configuration is applied​

Once an Agent is active, every message you send uses its configuration:

  • Instructions: the Agent's instructions are injected into the system prompt for the conversation. The Agent is told to follow them unless they conflict with higher-level platform rules.
  • Skills: the Agent's Skills are the only Skills available in that conversation. The in-chat Skill MCP tools (skill_list, skill_get, load_skill, skill_execute) are scoped to the Agent's assigned Skills, so the assistant can list, look up, and run those Skills and no others. It can run a Skill inline in your conversation or in an isolated session, at its discretion. For more information about designing Agents around their Skills, see Composing Agents & Skills.

This applies on top of the rest of your chat context; your project, Knowledge access, and any organization- or instance-level system prompts still apply. The Agent's instructions layer in as additional, Agent-specific guidance.

Switching or leaving an Agent​

Before you send your first message, you can cancel the selected Agent and the chat returns to a standard chat with your normal Skills and tools available. After the conversation has started, the Agent applies for the rest of that chat. Start a new chat to use a different Agent or none at all.