Copilot is the conversational chat experience where an end user talks to an AI agent — the front-end people actually use day to day, while the agent builder is where makers configure what sits behind it.

Overview

Copilot opens with a friendly prompt — How can I help you today? — and an input box that reads Ask anything.... You can pick which agent you're talking to, start new chats, revisit past ones, attach files, and use voice.

Which Version of the Agent Copilot Runs

Copilot always talks to the agent's latest published version — never your unpublished draft. When you publish with the "deploy on Copilot" option (on by default), the Copilot deployment is created or moved to the new version in the same step, so publishing is exactly the moment your changes become what Copilot users get.

Note: If Copilot seems to ignore an edit, check the published version and time shown under the agent's name in the editor header. If you've edited since then, Publish again with Deploy the new version on Copilot checked.

Starting and Managing Conversations

Use New Chat to begin a fresh conversation; your previous conversations are listed under Chats so you can return to them. Within a chat, type in Ask anything... and Send message; follow-ups use Ask a follow up question. You can Clear Conversation to start over, and Regenerate Response to ask the agent to answer again. A header lets you switch between the agents available to you, so one Copilot can front several agents.

Each New Chat is its own session: the whole chat — however long it runs — is recorded as one conversation, with your first message as its title, and shows up as one row in the agent's Sessions view. Returning to a chat under Chats continues that same session. Responses can stream in as they're generated, and you can stop a response mid-generation. Because the session keeps the full history, follow-up questions have the context of everything said earlier in that chat.

Attachments and Voice

Use Add attachments / Attach file to include context. Which file types you can attach, how large they can be, and how many you can add per message are set in each Copilot's configuration. A warning also reminds you that "Uploaded files are stored for up to 15 days. Avoid sharing sensitive documents."

For voice, Speak your message lets you talk instead of type, and Read Aloud reads a response back; Stop speaking ends it. These make Copilot usable hands-free and let the agent work from documents you bring.

Seeing Where an Answer Came From

Copilot can show its work. Responses surface Sources and Citations for where the information came from, and a Chain of Thought view — See thought journey — that reveals the agent's reasoning steps (you'll see Thinking... while it works). Citations may appear either as named source references or as numbered markers ([1], [2], …) inline in the answer, matching numbered cards in the citations panel — which style you see depends on how the Copilot was set up.

Note: Whether reasoning is shown depends on the agent's response settings.

Giving Feedback on a Response

Each response has feedback controls: Like message / Dislike message (or Good Response / Bad Response), and Provide Feedback — "Your feedback helps us improve the system." You can also Copy message to reuse a response, and act on any Suggested actions the agent offers. Feedback you leave feeds the agent's evaluation and quality monitoring, so it's worth flagging answers that were especially good or wrong.