Can I run more than one coding agent in a project?
You can give any small terminal or added agent a clearer project-specific name. Double-click its title, type the name, then press Return (or click away) to save it. Press Escape to cancel. Renaming changes only the label: the terminal, agent identity, role, and conversation stay exactly the same.
Yes. Keep your main Agent as the Lead, then add Claude Code or Codex agents in the Terminals drawer. They all see the same project, but each one has its own terminal, name, status and inbox.
Add an agent
- Open Terminals at the bottom-right of the quadrant.
- Press + in the right-hand rail.
- Choose Claude Code or Codex — GPT-5.6 Sol.
- Choose Builder or Reviewer.
- Press Start.
The new agent appears like any other terminal. The drawer keeps every terminal at the same comfortable width, shows two at a time, and scrolls sideways for the rest. It stays on the leftmost built-in terminals until you scroll right yourself, even after you add another terminal. You can minimise an agent into the scrolling rail or maximise it into the quadrant's large view. The Lead Agent keeps running when another terminal is large.
Which role should I choose?
- Builder: can change the project. Codus tells builders to agree which files or part of the problem each one owns before editing, so they do not overwrite each other's work.
- Reviewer: checks the work, challenges the approach and runs tests. It has the same full access as every Codus coding agent, so it can also fix an issue when you or the Lead Agent asks. The role describes its focus; it is not a permissions restriction.
A useful team for a difficult bug is one Lead Agent, one Builder investigating a separate theory, and one Reviewer trying to break both proposed fixes.
Can they talk to one another?
Yes. Each added agent receives a stable address such as agent-10. Agents can list everyone in the quadrant and send a message to lead or to one exact agent. Codus saves the message before trying to deliver it, so it is not lost if that agent is busy or stopped. A busy agent is nudged when it reaches its prompt; a stopped agent reads the waiting message when you start it again.
What stays separate?
Each agent has its own conversation and terminal history. They share the project files, Git changes and running services. File references printed by any of their terminals are clickable and open in the same project DevTools drawer, at the referenced line when one is included. Codus launches Claude Code, Codex and added coding agents with full access and without permission pop-ups by default. This is part of the Codus experience: an agent can keep working without waiting for you to approve routine commands.
Because the files are shared, two agents should not edit the same file at the same time. Use a Reviewer for an independent first pass, or tell agents to divide the files or parts of the problem before either starts changing things.