Concierge gives your Codex agent a phone number. Talk through the work, let your agent keep going, and tell it when a callback would help. The task stays in Codex.
Requires eligible Codex access and an awake, connected computer. About the beta.
Live voice
You + Concierge
Talk it through. Put it in motion.
Leave the work with your agent.
Set when a callback would help. Step away.
You have a question about the work, but you are out walking. Call your Concierge number and talk it through with Codex.
Concierge gives Codex a number you can call and a way to reach you. Use the app when you want the screen. Use the phone when you want to talk and keep moving.
The Codex agent on the line can delegate work and discuss the results with you. Ask it to compare approaches, change direction, or keep working after the call.
Choose a monthly plan. Phone minutes refresh every seven days from activation, separately from monthly billing.
For picking up a task, talking through an idea, and keeping things moving.
$15/ month
120 minutes every 7 days
For thinking out loud, directing your agents, and working through the details.
$100/ month
1,000 minutes every 7 days
Both plans keep you in control of who to contact and when.
Billed monthly. Phone time refreshes every 7 days from activation. No rollover or automatic paid overage. Requires your own eligible Codex access and a connected computer; OpenAI usage limits are separate. How phone time works
These example requests show how a phone conversation can fit into a task. Check beta availability for the calling options you can use.
“Have one agent review the code and another check the flow. Work through the fixes. Call me if you need a decision.”
Give the work a direction before you leave. Codex keeps the task and your instructions together after the call, including when you want to hear back.
“Find a dinner spot for four near the theater. Check the options and get the reservation call ready. Ask me before paying a deposit.”
Bring the research, the details, and your limits into one task. Personal phone requests stay connected to the reason you asked in the first place.
“Pick up the launch review. Talk me through the tradeoffs, then have the agents work on the option we choose.”
Resume the task in Codex with the earlier discussion and results available. Use the call to work through your next decision.
Choose who may call, when you want to be interrupted, and which decisions need your approval. Keep those preferences in one place.
Tell Codex what you need done and which decisions need you. Include when you want it to call.
Explain the outcome, the priorities, and anything you want to decide yourself.
Your agent delegates as you talk. Ask questions, react to findings, and refine the plan.
Hang up when you are ready. The task stays in Codex, where your agent can continue on your connected computer.
Set when a call would be useful, or schedule a recheck. Return to the same task when it needs your attention.
“The review is back. Both agents agree on the simpler approach. There is one tradeoff I want to walk you through before we change it.”
Hang up and your agent can keep working on the connected computer. The conversation, decisions, and results stay in Codex. When you return, recent context helps your agent pick up the discussion.
Turn on Work Awareness to let your agent share short updates from its work on the connected computer. A new owner call can use fresh updates to orient the conversation before checking the task itself.
You choose whether to enable it. Updates expire, and work on other or unavailable computers is not treated as current.
Hot Context helps reconnect a recent owner conversation or missed callback to its Codex task. You can review or dismiss that context. It is a shortcut back to recent work, not a record of everything you have ever discussed.
Tell your agent, "Keep working on the review. Call if you need my decision; save routine updates until I ask." Your agent weighs whether your answer would unblock the task, alongside your quiet hours and calling preferences.
A progress update alone does not grant permission to call. When a call is not appropriate, the task can continue or wait for a later recheck.
These owner-only features keep private work away from external callers. Read about continuity in the beta.
Find the supported calling options, read the full FAQ, or get in touch.
Yes. The Codex agent on the call can delegate work to other agents and discuss their results with you. You can ask it to change direction while you talk.
The call ends. The task stays in Codex. Your agent can continue on your connected computer, keep track of what needs your decision, and use scheduled rechecks. Return to the conversation in Codex to pick up where you left off.
Your own eligible Codex access, the Concierge plugin installed and active, a computer to connect, and your phone. Start setup to connect your account and computer, then set up your subscription and phone number. This website completes setup; it does not install the plugin.
Yes. Your paired computer needs to be awake, online, and running the required Codex connection for live phone calls. Your task remains in Codex if the computer disconnects; execution and calls depend on that connection being available again.
You choose who is allowed through. Keep your number private, admit a trusted contact, or allow an expected caller for a limited time. Other callers do not get your private task context or owner notes.
Set who your agent may contact, your quiet hours, and when it should ask you first. You can also review the context for your next call, dismiss an old task, and edit the notes you want it to use.
Your calls to Concierge and its calls to you use your phone allowance. Talking to Codex directly in the app does not. Phone minutes refresh every seven days from activation, separately from monthly billing, with no rollover or automatic paid overage.
No. Concierge provides the phone connection and phone allowance. You use your own eligible OpenAI account for Codex, and its usage limits still apply. More Concierge minutes give you more phone time, not additional OpenAI usage.
