Human-in-the-loop for agents

Youragentwantstorunacommand.You'reaway.

Pulse sends every Claude Code tool call to your iPhone. Tap to allow or block. Full audit log. Your agent waits.

9:41
Pulse
now

Claude wants to:

Edit /src/db/migrations/001.sql

Allow
Block

Ran: git push origin main

Read: /etc/environment

Background agents are powerful. They're also opaque.

You kick off a task, walk away, and come back to find Claude edited files you didn't expect, pushed a commit you weren't ready for, or deleted something it shouldn't have. You had no say.

How it works

01

Install the hook

One shell command wires a PreToolUse hook into Claude Code. Lightweight, non-invasive.

02

Relay to your phone

Every tool call goes through the Pulse relay. You get a push notification with exactly what Claude wants to do.

03

Tap to decide

Tap Allow or Block. Claude continues or stops. Every decision is logged.

What you approve.

Configure which tools require approval and which can run freely. By tool name or path pattern.

File edits

Review every line changed before it hits disk.

Bash commands

Block destructive scripts before they run.

Web fetches

See exactly which URLs your agent reaches out to.

Git operations

Nothing gets pushed to production without a glance.

Stack

Open source. Self-hostable. iOS app (Swift + Supabase) + Next.js relay + Claude Code hook. No third-party servers see your code.

iOS beta coming soon.

Join the waitlist — first access when we ship the private beta.