Codex Remote Control
Resolved 💬 57 comments Opened Jan 14, 2026 by sluongng Closed May 14, 2026
💡 Likely answer: A maintainer (sluongng, contributor)
responded on this thread — see the highlighted reply below.
What feature would you like to see?
The ability to remotely control codex cli running on my desktop PC from my phone with ChatGPT app -> Codex tab.
Additional information
More context available here [https://github.com/openai/codex/discussions/9200](<https://github.com/openai/codex/discussions/9200>)
I'm basically doing this already today with a combination of Tailscale (VPN) + Terminus (SSH client).<br>Being able to use a proper mobile app UI would be much better for me than having to access the sessions through ssh + tmux.
57 Comments
It would be great if ChatGPT web and the mobile app could send tasks to Codex CLI. In Claude Code, it implemented via MCP servers, but in ChatGPT, unfortunately, enabling Developer Mode disables memory.
It would be great if the developers did one of two things: either allow memory to work in Developer Mode, so this kind of functionality could be achieved via MCP, or directly add a native connector in ChatGPT to connect to a local remote terminal. That would be a killer feature.
This is what I've been hoping and looking for. I typically use codex via vs code. However now with Codex clienbt out for Mac I'm giving that a shot. Either way, ideally if I start a codex session up on my mac and i'm happily coding away and leveraging helpful agents in side conversations. All works great, agents have the access to the cli and all code and can knock out all sorts of tasks.
The problem, is when I'm not at my desk. Leave to go to a meeting, lunch or any other number of reasons and my only option is to note an idea down or wait till I get back to my desk. I can't be in the moment while away.
Ideally, I should be able to open up the ChatGPT app on iPhone, go to Codex and see the conversation that I started on my Mac and be able to continue the conversation while the actual work continues to be done on the Mac. I just need to be able to monitor things and continue the conversation regardless of being at the Mac. Right now I don't see a way to do this?
That scenario is actually pretty easy to cover.
Tmux on your work machine and termux on your phone.
Not sure if it exists on iPhone, but it does on Android.
Then via SSH I can check what’s going on in my terminal where Codex CLI is running and even able to give it new tasks.
I mean something else though - I want ChatGPT Web, which uses a slightly different model and thinks differently about generalized problems, to have access to my disk, or at least some kind of sandbox on my work computer where my repo files live. That would be way more convenient than manually updating the project files in GPT Projects every time there are serious changes in the repo.
Built something along these lines as a side project. A web client that connects to a local Codex app-server instance and relays events to any device. Execution stays local and connection is outbound only.
Repo is here if anyone's curious: https://github.com/z-siddiqi/zane
The app-server JSON-RPC interface made this pretty straightforward to build on top of.
Thank you for sharing this and your work on it. I kinda forked it (probably not the right way) and made a bunch of tweaks and changes the more I played with it. I feel I simplified it in a way and yet still secure'ish I think. You can check it out here: https://github.com/ddevalco/codex-pocket
It's already become a big help for me!
Cool! Glad it helped.
As of today, Claude Code has just shipped this feature: https://code.claude.com/docs/fr/remote-control
It would be really nice and convenient to add this to Codex as well
Yes, it would be perfect to have a lightweight app made specifically by OpenAI that allows you to connect to your local machine's terminal from an Android or iOS mobile device. That would be just ideal.
It would be great to have this feature - since I started using AI, I often leave my laptop at home and work from my table or the phone on the go. I can start a task and go to the gym and monitor the task remotely. I usually use a Remote Desktop, but being able to do it directly from a plugin or app would be so much easier.
want it remote
maybe just integrated this remote control within chatgpt just like what claude app does.
A major reason advanced users value Clawdbot and OpenClaw is the flexibility to work on our own terms, rather than being constrained by vendor-locked workflows.
Because of that, adding a secure remote control capability would be a high-impact improvement. It would let users continue managing development while away from their desk—during lunch, between meetings, or whenever they are not directly at their machine. For users running longer agentic workflows, that kind of continuity would be extremely valuable.
If this is explored, I believe the strongest implementation would be through the CLI, not only through the Codex Mac app. The CLI is where many of the most technical and workflow-heavy users already operate, especially those coordinating sub-agents, automation, and GitHub-based development. That audience is especially well positioned to benefit from tighter remote orchestration.
I also think there would be meaningful trust advantages in having this capability provided through an OpenAI-owned and primarily OpenAI-maintained solution, rather than relying mainly on third-party implementations. For many of us, the work being handled involves client responsibilities, sensitive code, and operational risk, so trust, accountability, and long-term reliability matter as much as the feature itself.
Connecting Codex App to my remote Dev Codex CLIs would be amazing..
We solved a version of this for Claude Code using IDE Agent Kit. It polls chat rooms (Ant Farm, Discord) and relays messages to a running Claude Code session via tmux. So you can message from any device with a chat client and Claude Code responds.
Not as polished as a native mobile UI, but the architecture (external message source -> relay -> IDE agent session) works for the core use case of "I am away from my desk and need to steer the agent." The main limitation is latency: polling-based relay adds ~2 min round-trip vs. real-time with websockets.
We need this ASAP)
I built an open-source tool called pikiclaw that solves this for Codex CLI (and Claude Code / Gemini CLI as well).
It runs on your local machine and bridges Codex CLI to Telegram (or Feishu). You chat with your agent through Telegram — prompts go to Codex running locally, and output streams back to the chat in real-time. No remote server, execution stays local.
Key things relevant to this thread:
Compared to the web-based solutions shared above (zane, codex-pocket), the main trade-off is that it uses Telegram instead of a custom web UI — which means no extra port to expose, and you get native mobile push notifications for free.
(Disclosure: I'm the author. MIT licensed.)
Claude lets you do exactly that even from iPad: https://code.claude.com/docs/en/remote-control
That's what I need - an official way to use my iPad as a remote controller for my Mac at home.
This is an urgently needed feature. Upvoted
Please!
RC it!
YAP, WE NEED IT!
First party would be really welcomed.
+1 to having this!
let's get this soon~
Feels a bit ironic that in the era of “vibe coding” and agentic AI doing weeks of work in minutes… we still don’t have remote control in the Codex CLI 😄
You’d think this would be the perfect showcase: “feature requested → agents spin up → PR ready by morning.”
But somehow the tool that’s supposed to enable this revolution isn’t exactly leading by example.
Where are the vibe coders when we actually need them?
Haven't read all the comments but as Claude's
/rcwas mentioned I want to add that personally I'd prefer to be able to control Codex Desktop to be able to create new sessions, enable worktrees etc from ChatGPT mobile app, without having to create a session on desktop first, that would be the real game changer 🥰Yeah, especially considering the fact that this has already been implemented by a major competitor that could potentially pull users away from OpenAI.
As the author of this GH issue, I can say that I was pretty early into this and have a few things to say.
---
It worth calling out that Codex has the ability to run abitrary bash/powershell commands on the target machine. Serving this capability remotely essentially means creating a giant backdoor into your computer. Yes sandbox will help here, but a lot of folks I know do not enable sandboxing thus, this feature has a huge security implication. I for once appreciate that OpenAI / Codex team is taking their time implementing this.
Second, it's obviously not possible to serve the traffic directly from your phone to your computer of choice. In some cases, the network topology between the 2 devices can be way too complex to allow access through open port. That means to do this reliably, OA _SHOULD_ have a set of "Bastion" servers (think SSH Jump hosts) to serve the traffic in between: cli register itself to these servers, these server in turn will be available to the Mobile app client.
Due to the security implication, it would be ideal if these traffic is e2e encrypted. That would reduce the legal burden from OpenAI having to operate these servers with user data in the middle, subjected to local government raids and targeted attacks to inspect and steal users data. Operational Reliability is very costly without e2e encryption here.
---
Personal Speculation
I think OA should just try buying one of the VPN vendor (i.e. Tailscale) to make this happen.
I am very tempted to code my own Codex app-server wrapper based on https://tailscale.com/docs/features/tailscale-serve or https://tailscale.com/docs/features/tsnet#make-calls-with-tsnetserver. It can be a simple web-based interface that is accessible via mobile's browser. (may be I _did_ created such a thing privately)
I hope whoever is in the current private beta invite would do some pentest to ensure the setup is reliable and secured. Cheers.
I wanted this too, so I built a local-first workaround for it: RemCodex.
It is not an official ChatGPT app integration, but it gives me a browser-based control surface for Codex sessions running on my own machine, including from my phone.
Current workflow it covers:
GitHub: https://github.com/lupishan/remcodex
Try it:
npx remcodexIt does not replace a native Codex tab inside the ChatGPT app, but it solves a very similar remote-control workflow today.
The reason I built it was pretty close to what this issue describes.
I was already using a remote workflow with SSH / tmux, but the UX was rough for real session control:
That pushed me toward building a browser-based control layer instead of just another terminal access setup.
hi @lupishan will it work for android phone and linux pc ?
@Kunal1522 yeah should work 👍
it’s basically browser-based on the client side, so anything with a browser (android included) works
and you can run it locally on linux just fine
still a bit rough on mobile though 😅
and now, as even the code of claude code is leaked it should be even easier for codex to figure out how 🤣
Wtf how is this still not implemented in the age of agentic engineering. Come on, OpenAI...
Yes, you and everyone else built their own. Github comments aren't really your marketing platform. Ask your agent to do better.
@etraut-openai was this closed by mistake by linear bot? If it’s duplicated, could we get a link to the tracking issue to subscribe for updates?
+1 I am doing the Codex CLI/Tailscale/tmux/Termius workaround like everyone else but the convenience of Claude Code + /remote-control + Claude iOS integration is so convenient, I find myself defaulting to that setup. Would love to see feature parity in Codex CLI!
OpenAI, please, make this functionality opt-in and make it possible to disable it via managed defaults
This is the only feature that I cannot leave Claude for Codex yet, please.
come on, this is super low lift. everybody's done their own but we need an officially supported version due to very reasonable security concerns of using third-party solutions.
Adding another concrete user workflow here, because this is now the main feature gap that keeps pulling me back toward Claude Code Remote Control.
My day-to-day Codex usage is long-running local work: multiple repos, local worktrees, local env/DB context, browser/tool use, PR cleanup, tests, and follow-up while an agent is already mid-task. When I step away from my Mac, I do not want to move execution to the cloud or expose my filesystem broadly; I just want to keep supervising the same local Codex Desktop/CLI session from my phone.
The important capabilities for me are:
I am aware there are SSH/tmux/Tailscale and third-party app-server wrappers, but those are not a great replacement for an official flow. The UX is rough on mobile, approval flows are awkward, and for sensitive work I would strongly prefer an OpenAI-supported implementation with a clear security model instead of depending on random remote-control bridges.
The MVP I would personally find useful:
This does not need to be a cloud execution feature. Local-first is the point: the phone/web client should be a controller/viewer, while the local Codex session remains the source of truth.
One clarification on the ideal UX from my side: I am not mainly asking for a separate generic browser UI.
What I would really like is native ChatGPT mobile integration:
So the mental model is: my Mac keeps running the real local Codex session, but ChatGPT mobile becomes a remote controller/viewer for that same session. Similar to how Claude Code Remote Control lets the mobile app reconnect to a desktop/terminal session, but for Codex and ideally integrated directly into ChatGPT's Codex surface.
A browser link or QR pairing would still be useful as the pairing mechanism, but the destination I care about most is the ChatGPT mobile app's Codex tab, not a separate third-party web console.
One more point on business/user impact: this is not a nice-to-have feature for me. Remote control is heavily used in my daily workflow today, and it is currently the main blocker preventing me from fully switching my day-to-day agentic coding workflow from Claude Code to Codex.
I already want to use Codex more heavily. The model/app direction is strong, and for local engineering work Codex is very close to being my primary environment. But I currently rely on Claude Code Remote Control throughout the day: I start longer-running local tasks, step away from my computer, then continue monitoring or steering the same session from my phone. That workflow is now part of how I work, not an edge case.
So the decision is not just "would remote control be convenient?" For users like me, the lack of this feature changes the default product choice. Even if Codex is strong in other areas, I cannot fully move my daily workflow over while I lose the ability to reopen and control my active computer session from the ChatGPT mobile app.
From a business/product perspective, I think this is a high-leverage retention and migration feature:
For my workflow, this is currently the one missing feature that prevents a full switch.
+1 on this. Claude Code's remote control is the one feature keeping me from fully committing to Codex as my daily driver. A native iOS implementation with a clean mobile UI would close that gap entirely. What I want is simple: open the ChatGPT app, see my active local Codex session, and be able to monitor progress, send follow-ups, and approve or deny actions without ever touching a terminal on my phone.
A large number of duplicates (or earlier versions) of this issue has been closed as "not planned". But this is now open, so is it this now something that is planned soon? It would be great to be able to work on mobile without using an ssh, vpn and a tiny terminal.
You can try my little IPC script, which I've been using recently. At least the prompt executions then also synchronize with the Codex GUI. It partially reverse-engineers the communication between the VSCode Plugin and the Codex GUI and works well on my Mac. It allows reading the agent chat messages (read-full), sending messages (sent), and listing the chats (list-all). It may also work on other platforms, but I haven't tried it yet.
https://gist.github.com/InfinityMod/ecc1f441f7447824ff114b8a41debec2
+1 on this!
This has become an invaluable part of my Claude workflow and I really think this would be an incredibly helpful feature to use Codex more, rather than having to ssh into a tmux session from my Android tablet.
(Please ensure that we can still upload files from the remote app if this implemented, as that makes providing screenshots much less painful than transferring them via scp/sftp and instructing the tool to look at the uploaded file.)
I'm excited to say that this feature is now available in preview! For details, refer to this announcement.
Please give it a try! File bug reports if you run into problems and feature requests for follow-on improvements.
Uff amazing, just updated Codex desktop, noticed the new mobile menu entry and set it up successfully, can manage my codex sessions live from my phones ChatGPT app 🥳
Only pinned chats would be awesome to have the same on mobile, I use them heavily for more complex tasks that I want to keep track of
@etraut-openai Do you plan to provide the same service to the command line app? I prefer to run codex, and other agents in VMs and on other machines. Both for security and convenience reasons.
@solomatov For CLI control specifically (not Desktop), the community has shipped a handful of solutions:
The OpenAI Desktop announcement targets the Desktop GUI app. The CLI side — relevant for VMs, SSH-only boxes, and the "agents in another machine" workflow you described — isn't covered, so first-party CLI mobile control would still be a real gap. The community tools above cover different points on the security/convenience curve.
It is already available in the alpha releases
I didn't tested it yet but it should work
https://github.com/openai/codex/blob/main/codex-rs%2Fapp-server-daemon%2FREADME.md
I can confirm that it's working
The only issue I see is this one: https://github.com/openai/codex/issues/22762
Using it right now. Works like a dream, great job !
codex remote-control start --jsonThe above just works for me. I just ran it on the remote machine where I use the codex CLI to do all my development. Codex has two separate paths:
codex remote-control start --jsonenrolls this server with ChatGPT remote-control relay, so same-account mobile app can discover the server through backend relay. When I open my mobile app, in the codex tab, I am able to see all the codex sessions from the server.To work with the same server using the Codex Desktop app running locally, you will have to do:
Settings > Connections > Add/enable SSH host. That triggers the Codex Desktop app to do 2 things:codex remote-control start --jsonon the server, that would have already started the "codex app-server" on the server, so, this step gets skipped.codex app-server proxy, which just bridges desktop stdio to remote Unix socket. Only this step gets executed and the proxy gets set up.Then, all the codex sessions from the server start showing up on the Desktop app as well.
After connecting remote codex to my Mac, I got Codex to actually generate me a pairing code that worked on my Android phone. Now my Chatgpt app shows both my mac and my remote dev box. I asked Codex to generate reproducible instructions for others, see file attached:
instructions.md