--- title: How does the Brain hand work to a quadrant's agent? question: How the Brain delegates work to a quadrant area: brain order: 3 surface: desktop audience: beginner keywords: [delegate, hand off work, assign task, tell quadrant, ask quadrant agent, orchestrate, multi-quadrant, message brain, unread badge, report back, mesh, brain to quadrant, quadrant to brain, start quadrant, boot agent, Q1, Q2, Q3, Q4, plan for a quadrant, which brain gets messages, brain button, title bar badge] related: [what-is-the-brain, manage-the-brain-drawer, preview-covers-the-drawer] updated: 2026-07-21 --- # How does the Brain hand work to a quadrant's agent? Ask the Brain in plain English to have a quadrant do something, for example "Tell Q4 to rebuild the dashboard". The Brain types that instruction straight into that quadrant's own agent and runs it, so the work starts there in front of you, exactly as if you had typed it into that quadrant yourself. If that quadrant's agent is not running yet, the Brain starts it first. When the quadrant's agent finishes or gets stuck, it can message the Brain back, and a small count appears on the brain button in the codus title bar so you know a reply is waiting. ## Before you start - This is a desktop feature. codus runs on macOS today. - The quadrant you want to hand work to needs a project folder set, so the delegated work has a place to run. If a quadrant has no folder yet, set one first. - You talk to the Brain by typing, the same way you talk to a quadrant's agent. Open it by clicking the **brain button** (the 🧠 near the top of the codus window). ## Give the Brain a job for a quadrant 1. Click the **brain button** (the 🧠 in the title bar) to open the Brain. 2. Type your request in plain English and name the quadrant, for example "Tell Q4 to rebuild the dashboard". Press Enter. 3. Watch that quadrant. The Brain's instruction appears as typed input in that quadrant's terminal, and the quadrant's agent starts working on it as if you had typed it there. There is no separate "delegation" inbox or approval step to find. The request lands in the quadrant and the agent responds. That is the whole handover. The Brain does not do the quadrant's work in its own window and copy it across. It drives the quadrant's own agent, so everything the agent does stays visible in that quadrant's terminal. ## The Brain starts a stopped quadrant for you You do not have to start a quadrant before you delegate to it. If the target quadrant's agent is not running, the Brain starts it first, then types in the instruction. That start is the same as launching the agent yourself from the quadrant's **start** button. It boots with the same access and behaviour, so a Brain-started agent can do everything a hand-started one can. There is nothing to configure for this. If you are looking for a setting that controls how the Brain launches your quadrant, there is not one, and you do not need one. ## Naming the right quadrant Refer to a quadrant by its on-screen position on the tab you are looking at: - **Q1** is top-left. - **Q2** is top-right. - **Q3** is bottom-left. - **Q4** is bottom-right. These positions are relative to the visible tab. "Q1" means whatever sits top-left on the tab that is open right now, so the same "Q1" points at a different quadrant on a different tab. If you have dragged quadrants around to swap them, the position still wins: Q1 is always whatever is top-left. The Brain also tracks each quadrant by a fixed identity of its own. That matters when a quadrant messages the Brain and the Brain replies: it answers the exact quadrant that wrote to it, even if you have since switched to another tab. ## What the Brain does itself, and what it hands over The Brain handles cross-project and universal jobs directly, for example listing your projects or setting up a new worktree. Anything that depends on one project's setup goes to that project's quadrant instead, because only the agent sitting in that folder knows the project. That includes build commands, environment variables, and a framework's particular quirks. Even drafting a plan for a quadrant's project is handed over. When you ask the Brain to plan work for a quadrant, it asks that quadrant's own agent to draft the plan, so the plan anchors to the right project folder rather than to the Brain. While a quadrant is working, you can also ask the Brain to look in on it. The Brain can screenshot a quadrant's live preview, read its preview console, tell you what URL it has loaded, and open or close that preview while you talk. For example, "Take a screenshot of Q3's preview" or "What URL is loaded in Q2?". ## Hearing back when the work is done A quadrant's agent can message the Brain when it finishes a delegated job or when it gets blocked. You do not have to sit and watch the quadrant to catch this. When a quadrant has replied, a small count appears on the **brain button** (the 🧠 in the title bar). It shows the number of unread messages, or **9+** once there are more than nine. The count updates on its own, with no refresh, and it clears once the Brain has read those messages. So an unread count is your cue to open the Brain and see what a quadrant reported. ## Which Brain gets the reply By default, a quadrant's message goes to the main Brain, which is named **Brain 1**. If you run only one Brain, that is the one, and everything above applies to it. You can run more than one Brain, each in its own tab in the Brain drawer. As of the current release, a quadrant's agent can send its message to a specific Brain by that Brain's tab name, for example **Brain 2**. Leave the Brain unnamed in the request and the main Brain answers, exactly as before. Brains can also message each other, and any Brain can push work down to any quadrant. So work and replies can flow in every direction across your Brains and quadrants. ## How do I know it worked? - Right after you delegate: the instruction shows up as typed text in the target quadrant's terminal and that quadrant's agent begins responding. - If the quadrant was idle: it launches, then the instruction appears, without you touching the quadrant. - When the quadrant is done or stuck: a count appears on the 🧠 brain button in the title bar. Open the Brain to read the reply, and the count clears. ## Common problems **I told the Brain to use a quadrant and nothing happened.** Check that the quadrant has a project folder set. A quadrant with no folder has nowhere to run the work. Set a folder for it, then ask again. **I named "Q1" but the wrong quadrant started.** Q-numbers follow on-screen position on the tab you are viewing, and the same number points at a different quadrant on a different tab. Make sure the tab you mean is the one on screen, then name the position you see, or tell the Brain which project you mean. **Clicking an option inside the quadrant did nothing.** The quadrant is a terminal. Use the arrow keys and Enter to answer it, not the mouse. This catches everyone the first time, so if clicking felt broken, it was not you. **I want to change how the Brain launches my quadrant.** There is no such setting, and you do not need one. A Brain-started agent launches the same way as one you start by hand. **The count on the brain button will not clear.** It clears once the Brain reads the waiting messages. Open the Brain so it picks them up.