Know the moment your agent needs you.
AgentBell pings your phone when an AI coding agent waits for an answer, asks permission, finishes or fails. See every agent on every computer on one live board, with cost and changed files. Approve simple prompts from your phone.
npx -y https://agentbell.surgesoftware.co.in/agentbell.tgz setupOne line on Windows, macOS or Linux (Node 18+). Sends status only, never your code.
Working 4
Fix rounding in partial refunds and add tests
Turn the pricing table into a component
Backfill March orders into the warehouse
Polish onboarding screens
Simulated agents. Tap Allow on a request to try it.
You come back to a stuck agent
It asked “May I run the tests?” 40 minutes ago. Nothing happened since.
You babysit the terminal
You keep checking a tab instead of doing other work, just in case it finished.
You lose track across machines
Laptop, desktop, a build box, a teammate’s run. Which one needs you now?
Set up in one minute
Run one command
It connects this computer (you approve a code in your browser) and adds Claude Code hooks. Nothing to relaunch or wrap.
Allow notifications
Open the app on your phone, add it to your home screen and tap “Turn on alerts”. Telegram works too.
Get back to work
Your phone buzzes when an agent waits, finishes, fails or looks stuck. Tap to see what it did.
Everything you need to leave your agents alone
Four clear states
Working, needs you, finished, failed. Grouped so the ones that need you are always on top.
Approve from your phone
Claude asks to run a command? Tap Allow or Deny on the notification. No answer in 90 s means the normal terminal prompt, never a silent yes.
Cost and tokens
Per session, read on your own computer from Claude Code’s local log. Only the totals are sent.
Changed files
The names of files each session changed, from git. Never their contents.
Stuck alerts
An agent “working” with no activity for 20 minutes usually hung. You hear about it.
Daily digest
Each morning: sessions, agent hours, cost, files changed and anything still waiting.
Team board
See every teammate’s agents and what is waiting on whom. Invite by link.
Timeline per session
Started, prompts, tools, waits, approvals, finish: with times. Share a read-only link.
Every OS, every agent
Windows, macOS and Linux. Claude Code, Codex, Cursor, or any command with agentbell run.
Status only. Never your code.
The hook runs on your computer and sends a small status message. Your source code, diffs, file contents, full prompts and the agent’s answers never leave your machine. Turn on private mode and even the task line and file names stay home.
Read exactly what is sent →Sent
- Status (working, waiting…)
- Project folder name, branch
- One-line task, secrets removed
- Token and cost totals
- Changed file names
Never sent
- Source code or diffs
- File contents
- Full prompts and answers
- Environment variables
- API keys, tokens, passwords
Simple pricing
Free for one computer, forever. Pro is half the price of the closest paid tool.
Free
$0For trying it on one computer.
- ✓ 1 computer
- ✓ Live dashboard on phone and desktop
- ✓ Push when an agent waits, finishes or fails
- ✓ Cost, tokens and changed files
Pro
$5 /moFor developers running agents all day.
- ✓ Every computer you own
- ✓ Approve or deny from your phone
- ✓ Stuck-agent alerts
- ✓ Daily digest of what your agents did
Team
$6 /user/moFor teams who share agents and reviews.
- ✓ Everything in Pro, per person
- ✓ Team board: every teammate’s agents
- ✓ Team digest for leads
- ✓ Invite by link
Questions
Which agents does AgentBell work with?+
Claude Code through its built-in hooks, OpenAI Codex CLI through its notify setting, Cursor through its hooks file, and any other command or agent through `agentbell run -- <command>`.
Does AgentBell see my code?+
No. The hook sends only status: working, waiting, finished or failed. It also sends the project folder name, branch, a one-line task with secrets removed, token and cost totals, and the names of changed files. Code, diffs, file contents, full prompts and answers stay on your computer. Private mode sends even less.
Does it work on Windows?+
Yes. The CLI is a single Node.js file that runs on Windows, macOS and Linux. The dashboard is a web app you can install on any phone or desktop.
How do phone alerts work?+
Open AgentBell on your phone, add it to your home screen, and tap “Turn on alerts”. It uses standard web push: no app store needed. Telegram alerts are also available on Pro.
Can I approve permission prompts from my phone?+
Yes, on Pro and Team, if you turn it on. When Claude Code asks to use a tool, your phone shows Allow and Deny. If you do not answer within 90 seconds, the normal prompt appears in your terminal. Nothing is ever approved by silence.
Will the hook slow down my agent?+
Barely. Each event is a small request with a 2-second limit. If the network is down, events wait in a local queue. The hook always exits cleanly, so a problem on our side never stops your agent.
What happens after the 14-day trial?+
You move to the Free plan automatically: one computer, live board, phone alerts and 7 days of history. Nothing is charged unless you choose to pay for Pro or Team.
Can I get a refund?+
Yes. Full refund within 7 days of payment, no questions asked. Write to [email protected].
Stop babysitting your agents.
14 days of Pro free. No card. Free plan after that, forever.