Agent Approve

You're racing to ship. If your agents are idle, you're not gonna make it. Every pending approval is wasted time. So you're chained to your laptop, all day and night. This isn't the AI future we were promised – go out and enjoy life. Agent Approve brings observability & control to your wrist: track AI agent activity, protect yourself with policy, and approve with a tap, wherever life takes you.

HOW IT WORKS
Agent Approve uses the hooks already built into your agent and coding tools, so you don't have to change how you work. When an agent goes to run a command or call a tool, it's encrypted on your machine and sent through our cloud relay, where it's evaluated against policy. Compound commands are separated to catch hidden destructive commands. Matches to your deny list or allow list are handled immediately. Those that need review are pushed to your iPhone or Apple Watch. When it reaches you, you see the full command, tap approve or deny, and can add it to your allow or deny list so you're never asked again. Everything is end-to-end encrypted: you own your data and keys.

WHY IT MATTERS
Some commands have no undo. Agents have wiped production databases, deleted home directories, and force-pushed away weeks of work – usually in seconds, while no one was watching. And the agents you can't see are the ones that bite: a scheduled task that drifts, a retry loop that burns thousands in tokens overnight, or a helpful agent sharing secrets with a bad actor. Agent Approve ships with a Restrictive policy that blocks 250+ destructive commands out of the box, a Permissive policy that auto-approves common safe commands, and you can always build your own.

EVERYTHING YOU NEED TO STAY IN CONTROL
• Approve from your wrist or phone – one tap, wherever you are.
• Centralized policy – allow, deny, or ask, with 250+ destructive patterns blocked out of the box.
• Compound-command parsing – catches the dangerous command hidden inside a scripted command sequence.
• Full observability – every command, tool call, and action in one live feed, including subagents and agent profiles (each Hermes profile shows up as its own agent).
• Readable at a glance – commands, plans, and tool payloads render as formatted markdown and JSON, not a wall of raw text.
• Interactive Q&A – review the agent's plan, and answer questions from your phone (limited support).
• Completion alerts + follow-ups – know the moment an agent finishes, then send its next move by voice or a quick reply.
• Privacy tiers & retention – choose how much of each command is stored (tool name, short summary, or full content) and how long to keep it.
• End-to-end encryption – you own your data and your keys, with optional key rotation for forward secrecy.

WORKS WITH EVERY AGENT YOU USE
Agent Approve runs on the hooks already native to your coding tools. Where there isn't one, we ship a plugin. 14 agents and counting, including: Antigravity, Claude Code, Codex, Cursor, Hermes, OpenClaw, OpenCode, OpenHands, Pi, VS Cod