Control Codex
from your phone.
Approve an escalated command, answer a question, or stop a runaway session from your lock screen. Codex keeps running on your machine.
npx pushary@latest setupPick Codex when prompted. Installs the hooks and connects over MCP.
Featured on Hacker NewsReview a denied command from your phone.
Left alone, Codex can hand an escalated command to its own reviewer, which decides without you and tells the model the action was rejected. In auto-review mode, Pushary lets that reviewer decide first. After a supported shell command is denied, the Stop hook can ask you whether to approve a retry.
An approval gate
When Pushary asks, the command waits. Silence does not approve a denied command’s retry.
Review, then retry
The auto-reviewer decides first. Approve a supported denial from your phone to permit that exact command’s retry.
One setup
The same command covers Claude Code, Cursor, Gemini CLI and Hermes on that machine.
Let Codex run. Keep the last word.
One command to connect. Approve, answer, or stop from your phone.
Start your trial$9.99/mo after a 3-day free trial. Cancel anytime.
Frequently asked questions
Can I run Codex from my phone?
Yes. The Pushary app can start Codex on a connected computer, send it an instruction, approve or deny an escalated command, answer a question, and stop a running session. The computer has to be awake, logged in, online, and running the Pushary background service. Codex itself never leaves your machine.
Does this work with the Codex CLI or the ChatGPT Codex app?
The Codex CLI, on your own computer. Pushary installs Codex's native hooks and connects over MCP, so it reads the same PermissionRequest and Stop events the CLI already emits. Cloud tasks you start inside ChatGPT run on OpenAI's infrastructure and do not emit those hooks, so they are out of scope.
How is this different from Codex Remote in the ChatGPT app?
Codex Remote links the ChatGPT app on your phone to Codex on a Mac or Windows PC. You can follow a task, send instructions and approve its requests there. OpenAI's setup starts from the ChatGPT desktop app, not from the Codex CLI or the IDE extension. Pushary is built for the CLI and needs no desktop app. It also lets a policy clear safe commands on its own, keeps an audit trail of every answer, and puts Claude Code, Cursor and Gemini CLI in the same app. Use Codex Remote for the desktop app and Pushary for the CLI if you run both.
Can the Codex CLI pair with the ChatGPT app on its own?
Partly. The Codex CLI reference lists a codex remote-control command: codex remote-control start runs a local app-server daemon with remote control on, and codex remote-control pair prints a short-lived pairing code for the ChatGPT app. OpenAI's Remote connections page still says mobile setup starts from the desktop app, so treat the CLI path as early, and it only reaches the session while that daemon keeps running. Pushary works through the hooks the CLI already runs, with no extra daemon, and a policy can clear safe commands without asking you at all.
What happens to Codex's auto-reviewer when Pushary is connected?
In auto-review mode, Pushary lets Codex's reviewer decide first. If the reviewer denies a supported shell command, Pushary can ask you from the Stop hook. Your approval permits a retry of that exact command; it does not turn off review for future commands.
Does an escalated command really wait for my answer?
When Pushary asks for approval, the command waits for the hook's answer. In auto-review mode, the reviewer runs first; a supported denied command can then be offered for a phone-approved retry through the Stop hook. Silence does not approve that retry.
Does it work with codex exec in a script or CI?
Notifications and the audit trail do. Approvals do not gate it: codex exec reports bypassPermissions whatever approval policy is set, which is the correct headless answer. Use phone approvals for interactive sessions, and keep CI runs governed by the sandbox and the permissions you grant the job.
Do I need a separate setup if I also use Claude Code?
No. One run of the setup command lets you pick every agent on the machine, and one subscription covers them. Approvals, questions, and the audit trail work across Codex, Claude Code, Cursor, Gemini CLI, Hermes, VS Code, and MCP clients.