Your agent froze, waiting for your yes.

Pushary sends the question to your phone. Tap Approve and it keeps working.

Pushary sends the step that needs you, not your codebase.

$ claude refactor auth/ --apply
rewrote session + token refresh
12 files changed, tests passing
Tap Approve to try it
Works with
Claude CodeCodexGemini CLIHermesCursor
Hacker NewsFeatured on Hacker News
6+
agents, one setup
1 tap
to approve or deny
3
question types
< 2 min
one-command setup

What is Pushary?

Pushary is the control panel for AI agents. When your agent needs a yes, Pushary sends the question to your phone, your Mac, Slack or the web app. You approve or deny, and every answer is saved. Works with Claude Code, Codex, Cursor, Hermes, Windsurf, and any MCP agent.

Your AI agents send permission requests to Pushary, which delivers them to your phone, your Mac's notch, Slack, the web app, and an audit trail.

Your agents

Claude Code
Codex
Cursor
Gemini CLI
Hermes
Your phone
Your Mac
Slack
Web app
Audit trail

Where it reaches you

+ any MCP agent, wired the same way

You already know this feeling

You hand off a task, finally stop babysitting it, and come back to find it stuck on a single question.

Without Pushary

  1. Hand off a long task. Walk away.

  2. The agent reaches a step it should not take alone. It stops and waits.

  3. You come back. It has not moved in 37 minutes.

With Pushary

  1. Hand off the same task. Walk away for real.

  2. Your phone buzzes. Approve the change before it goes live? You tap yes.

  3. You come back. The task finished half an hour ago.

Your agent should not wait 37 minutes for a single yes.

Get started

How it works

See the whole flow

One approval, start to finish, in under 90 seconds.

  1. 01Your agent asks. It pauses at a step that needs your call. Run a command, change a file, or pick between options.
  2. 02Your phone buzzes. It buzzes when a decision is needed, not for every step. Approve, deny, choose an option, or type a reply from your lock screen.
  3. 03It keeps going. Your answer flows straight back to the agent. When the task finishes, you get a ping for that too.

Get Pushary on your phone and Mac

Approve, deny, and watch your agents from anywhere. Same login as the web dashboard.

Download on the App StoreGet it on Google Play
Download for Mac
bash
brew install --cask pushary/tap/pushary
See the Mac notch app

Decisions in Slack and n8n too

The same question can reach Slack or an n8n workflow. Answer in one place and every other place updates.

Slack

Your agent's question lands in Slack with Approve and Deny buttons. One click answers it, and the message updates with the result.

n8n

One node adds a human checkpoint to any workflow. The flow pauses, asks you, and continues with your answer.

Stay in control

Set the rules once. Approve only what matters.

YOLO mode approves everything, including the rm and the deploy. Pushary auto-runs only the safe steps and pings you for the ones you cannot undo. You set that line once.

Auto-approve the safe stuff

Low-risk, reversible steps. Set the allowlist once and stop seeing them.

Escalate the risky ones

Deleting, deploying, spending money, anything you cannot undo. These go to your phone.

You set the wait time

Set how long each step waits. If you do not answer, your rule answers for you, deny or hand back to the terminal.

This is not blanket auto-approve. The actions that can hurt you still need a tap.

Your agent control panel

Every agent in one place. See which one is waiting on you, plus the full audit trail.

One command. Done.

Pick your agent and run it. Setup takes under two minutes.

terminal
npx @pushary/agent-hooks@latest setup

Configures MCP server, permission hooks, event tracking, and API key.

Start your trial

$9.99/mo after a 3-day free trial. Cancel anytime.

Questions, answered

Can I approve an AI agent from my phone?
Yes. Setup shows a QR code, you scan it once with your phone camera, and questions start arriving on your lock screen. Works on iOS and Android. App or browser, your choice. From the notification you approve, deny, choose an option, or type a reply, and the agent keeps going.
What is human-in-the-loop for AI agents?
Human-in-the-loop means an AI agent pauses at important or risky steps and waits for a person to decide before it continues. Pushary delivers that decision to your phone as a push notification, so you can keep an autonomous agent moving without watching it and still catch the actions that matter.
How do I get notified when Claude Code needs permission?
Install Pushary's hook with one command (npx @pushary/agent-hooks@latest setup). When Claude Code hits a permission prompt, you get a push notification to approve or deny from your phone. Safe tools can auto-approve so only the decisions that matter reach you.
What AI agents work with Pushary?
Pushary works with Claude Code, Codex, Hermes Agent, Cursor, Windsurf, and any MCP-compatible AI agent. Claude Code gets permission hooks and a setup wizard. Hermes gets a native plugin with automatic error notifications.
How long does it take to set up?
Under 2 minutes. Run npx @pushary/agent-hooks@latest setup for Claude Code or Codex, or pip install hermes-plugin-pushary for Hermes. The wizard auto-configures everything.
What are permission hooks?
Permission hooks intercept Claude Code's tool approval prompts and route them through push notifications. When the agent wants to run a command or edit a file, you get a push to approve or deny from your phone. If you do not answer, your rule answers for you, deny or hand back to the terminal.
How is this different from YOLO mode or auto-approve?
YOLO mode blindly approves everything, including destructive commands. Pushary lets you set per-tool rules: auto-approve safe commands (read, lint, test) and only escalate dangerous ones (rm, git push, deploy) to your phone. You stay in control without the fatigue.
What question types does Pushary support?
Three types: confirm (yes/no), select (multiple choice with 2 to 6 options), and input (free text). Your agent picks the right type for the situation.
Is my code sent to Pushary's servers?
Not on routine product paths. The decision path transmits notification metadata (title, body text, tool name). Pushary-launched transcripts are redacted and encrypted on your machine; your phone can decrypt them, and authorized Pushary compliance admins can recover them for documented legal obligations.
How much does it cost?
The Agent plan is $9.99 a month after a 3-day free trial, with 5,000 notifications a month, about 160 a day. It covers one agent. Running several agents at once is Agent Pro at $19.99 a month with unlimited notifications.

Why I built this

I kept coming back to an agent that had been waiting the whole time

I would kick off a task, switch to something else, and come back to find the agent had paused for a yes 40 minutes earlier. So I built Pushary to move that decision to my phone. The agent asks, I tap, and the work keeps going while I am away. This is the 2 minute version.

Aadil Ghani · Founder, Pushary

What people running agents tell us

Your tool really helped me improve my Cursor workflow and notifies me when the agent needs me or when it has finished a task. I don't know why this problem wasn't solved by someone already.
Andreas Kuoppa · Portality
I instantly related with the pain of coming back to my Claude Code and seeing "waiting for your permission" instead of "your task is completed."
Reagan McCullough · AgentPays
I love your vision of speeding AI agents up by unblocking the stuck permission screen and eventually learning from your permission behaviour based on the context. This is gonna go big very soon.
Takanari · AI Plaza

Send the next yes to your phone.

The next long task can finish while you are away. One tap and the agent keeps working.

Start your trial

$9.99/mo after a 3-day free trial. Cancel anytime.

The readable decision path sends only the title, body, and tool name. Pushary-launched transcripts are redacted and encrypted on your machine; your phone can decrypt them, and authorized Pushary compliance admins can recover them for documented legal obligations. If you miss a question, your rule answers for you, and the question expires after 10 minutes.

Approve your agents from your phone.