--- title: How do I chat with an agent instead of reading the terminal? question: chat with the agent in plain English area: quadrants surface: desktop audience: beginner keywords: [chat, chat panel, chat with agent, plain english, plain language, too much output, wall of text, verbose, terminal is confusing, message my agent, talk to agent, tap an option, options, buttons, answer without terminal, chat button, speech bubble, dictate to chat, send image to agent, paste image, opens in chat, default view] related: [whats-inside-a-quadrant, sleep-a-quadrant, start-the-agent] updated: 2026-09-09 --- # How do I chat with an agent instead of reading the terminal? Use the **Agent / Chat** switch in the bottom-left of a quadrant and choose **Chat**. The conversation appears over the Agent terminal, but the terminal and agent keep running normally behind it. Your agent posts one short, plain sentence at a time instead of pages of output, and you reply by typing, tapping one of the answers it offers, or speaking. ## Before you start - This is a desktop feature in Codus on Mac and Windows. - The quadrant needs a project folder set, and an agent to talk to. If no agent is running, sending a message starts one for you. - The chat and the terminal are two views of one conversation. Anything you say in the chat reaches the same agent, and switching between them changes nothing about the work. - If you write to the agent in the Terminal, the agent adds a very short version of your request to Chat in your voice. It does not copy long pasted code, HTML, logs, or documents into the chat history. ## Open the chat 1. Find the **Agent / Chat** switch in the bar along the bottom of the quadrant. 2. Click **Chat**. The switch stays in exactly the same place, and a note above the conversation confirms that the terminal and agent are still running behind it. 3. Type in the **Message** box at the bottom and press Enter. Shift+Enter gives you a new line instead of sending. 4. Click **Terminal** on the same bottom switch when you want the full terminal back. Your place is remembered. If you leave the chat open and switch to another tab, it is still open when you come back. ## Tapping an answer instead of typing When your agent needs a decision it asks one short question and offers up to four answers underneath it. Click the one you want and it is sent as your reply, so you do not have to type or phrase anything. If none of them fit, ignore them and type your own answer instead. Once you have answered, those buttons stay on screen but stop working, so the conversation still reads back later without letting you answer the same question twice. ## Speaking instead of typing The **microphone** at the end of the message box records what you say and turns it into text. While it is recording you get three controls: - The **microphone** stops recording and puts your words in the box, so you can read and edit them before sending. - The **paper plane** stops recording and sends the message straight away. - The **✕** throws the recording away without transcribing anything. ## Sending a picture Paste an image into the message box, or drag one onto the chat. It is attached to your next message and your agent can see it. This is the quickest way to show a screenshot of something that looks wrong. ## Starting in the chat instead of the terminal If you would rather not see a terminal at all, a project can open straight into the chat: 1. Click the **tools** icon in the quadrant header. A large modal opens, clearly labelled with the quadrant and project folder it applies to. 2. Under **Opens in**, click **chat**. That is saved for the project, not the position, so it follows the project around. You can still switch to the terminal at any time from the bottom bar. ## What your agent will and will not put here Your agent posts a short line when it starts, saying what it is about to do, and another when it finishes with the result. It does not narrate everything in between, because that would recreate the wall of text the chat exists to avoid. When a turn began in the Terminal, its short user message appears on the right like any other message. A tiny terminal symbol beside the time tells you where it came from. Messages sent directly in Chat have no symbol and are never duplicated. While it works you see three dots. Past a minute, the chat tells you how long it has been going and that long runs are normal, so a quiet stretch never looks like something has broken. ## Where else you can chat The same chat is available beyond coding quadrants. The Brain has the same permanent **Agent / Chat** switch at the bottom of its drawer, and your SEO and video teams have chat on the **team** row, where it follows whichever team member you have selected. ## How do I know it worked? - After clicking **Chat**: the conversation covers the terminal, the bottom switch moves to **Chat**, and the note at the top says the terminal is still running normally behind it. - After sending: your message appears on the right straight away, and three dots appear underneath while the agent works. - After sending from the Terminal: a short first-person version appears on the right when the agent starts, with a tiny terminal symbol beside its time. - If your agent asked something: a red dot sits on the **Chat** side of the switch, so you can see there is a question waiting without opening the chat. ## Common problems **The chat is empty.** Nothing has been posted to this project yet. Send a message and your agent will reply there. **I sent a message and it says it could not be delivered.** The agent could not be reached, and your text is put back in the box so nothing is lost. Start the agent from the quadrant's **start** button and send again. **My agent replied in the terminal but not in the chat.** Its answer is not lost, it just went to the other view. Newer versions of codus tell agents to always answer in the chat, so update codus if this keeps happening. **The message box will not clear after I send.** It clears immediately now. If the words stay put, you are on an older version of codus, so update it.