What Glass Agents Window can do that the CLI can’t:
Run and monitor multiple agents in parallel in one window, what you call multitask. The CLI is one interactive agent per terminal process.
Rich GUI like a collapsible diff view, inline review with Keep and Undo, tabs, and integration with the editor and extensions.
What the CLI can do that Glass can’t, or does differently:
Headless and scriptable mode: cursor-agent -p --output-format text|json|stream-json "<prompt>" for automation and pipelines.
Works anywhere you have a terminal, like SSH, remote servers, tmux and WSL2, no GUI needed.
Isolation via git worktree with --worktree.
Send a task to a background or cloud agent with the & prefix, then track it on cursor.com/agents or from mobile.
What both have in common: session persistence and resume using cursor-agent ls, cursor-agent resume, --resume <id>, --continue, /resume, plus MCP servers and background and cloud agents.
About multitask from the CLI: there is no single dashboard for multiple agents inside the CLI. To run in parallel, start multiple cursor-agent instances, for example in separate tmux panes or with isolated --worktree checkouts, or send requests to cloud and background agents.
About infinite scroll in tmux: this is a known bug. The dialog gets re-rendered from the start when you switch panes or zoom a pane, which causes the infinite scroll. This is not an unavoidable tmux or CLI limitation, it’s a bug on our side, but there is no ETA for a fix yet. Workarounds for now:
Don’t use tmux zoom toggle
Keep the CLI in a dedicated pane
Enable mouse mode in tmux for scrollback
To reduce output, use non-interactive mode: cursor-agent -p --output-format text "<prompt>" prints only the final answer