Agents ask before they do anything that matters. Nucleic turns those asks into fast, real decisions — from the Mac or your phone — without babysitting a terminal.
When an agent wants to run a command or change something risky, it stops and asks — and Nucleic tells you what kind of ask it is. A harmless read never looks like rm -rf. Allow it, deny it, edit it first, or always-allow that kind of action for the session.
Most tools force a bad choice: fully autonomous and hope, or a wall of yes/no you rubber-stamp. Risk-aware approvals put your attention where it counts, and “always allow” quiets the routine without giving up the gate.
The risky asks are flagged as risky — so your attention lands where it matters.
Always-allow the routine, keep the gate on the rest.
Prompts reach your iPhone too — unblock an agent from the couch.
Whoever responds first wins; the prompt clears everywhere else, stamped with who decided.
Nucleic is in open beta. All it asks is macOS 27 on Apple Silicon.