What comes back when I reopen codus?
When you quit codus and open it again, your workspace comes back the way you left it: your tabs, the quadrants open in each, each pinned to its folder, and your multi-window layout if you had one. Any Claude agent that was running when you quit resumes on its own, with its conversation, its files, and its codus tools all back; this auto-resume is always on and has no toggle. The one thing that does not return is anything that was actively running, because your dev server, the preview browser, and any workers or background jobs all stop when you quit and codus will not restart them behind your back. So the resumed agent's first message tells you it is back, names what stopped (for example your dev server on port 5173 and the preview browser), and asks whether to start those again. It starts nothing until you answer, so read the list and, if it looks right, type your reply and press Enter (the panel is a terminal and ignores mouse clicks).
Agents on your other tabs wake when you open the tab
codus resumes the tab you were last looking at straight away; agents on your other tabs wait until the first time you open each one. So a background tab sitting quiet after you reopen is normal, not broken, and its conversation is all still there. Click the tab to open it (the tab strip takes normal mouse clicks) and its agents come back then. This happens once per launch, so if you had already opened that tab and its agent then stopped (you typed exit, or it finished on its own), switching back will not restart it; quit and reopen codus to wake it again.
If a quadrant comes back as a plain terminal
Auto-resume only brings back Claude agents that were genuinely running the moment you quit. So a quadrant comes back as a plain command prompt, instead of your conversation, in three normal cases: you had stopped the agent, you never started one in that folder, or it was running OpenAI Codex (which codus does not resume for you). Nothing is lost; your conversation is saved on your computer. Open the quadrant's start menu (the start button with a small caret, a normal button you click) and choose start unbooted under Claude to load your earlier conversation straight back in. Starting an agent yourself this way continues the saved session quietly and does not ask about restarting anything. For a Codex quadrant, start it again from the same menu under OpenAI Codex. Do not use boot project to recover a conversation; that deliberately starts a fresh one and runs the project's full startup routine.