NativeAIHub

How to Control AI Coding Agents from Your Phone

Three ways to run, steer, and approve AI coding work from your phone in 2026: Claude Remote Control, ChatGPT Mobile to Codex, and the open source Remodex.

9 sections·8 min read

01Why This Matters

AI coding agents like Claude Code and OpenAI Codex run on your computer (or a cloud server) and do real work: reading files, writing code, running tests, making commits. They are powerful, but they take time. A complex task can run for five or ten minutes, and the agent may need your approval before doing something risky, like deleting a file or pushing to a remote repository.

In 2026, all major AI coding platforms added a way to monitor and steer these agents from your phone. This means you can kick off a coding task at your desk, walk away, and approve or redirect the agent from your pocket. You do not need to sit at your laptop watching a terminal cursor blink.

This guide compares the three main approaches available today, explains what each one can and cannot do, and helps you pick the right one for your situation.

02The Three Approaches at a Glance

Claude (Remote Control + Dispatch)ChatGPT Mobile to CodexRemodex (third-party, open source)
Who makes itAnthropic (first-party)OpenAI (first-party)Indie developer Emanuele Di Pietro (open source, Apache 2.0)
Which agent it controlsClaude Code (terminal agent) and Claude Desktop (Cowork)Codex App (desktop agent with full feature set)Codex CLI (terminal agent only, not the full Codex App)
Phone appClaude mobile app (iOS and Android)ChatGPT mobile app (iOS and Android)Remodex iOS app ($3.99/mo or $29.99/yr for full features)
Host machineAny machine running Claude Code (macOS, Linux, Windows)macOS only (Windows host coming soon)macOS primary; Node.js bridge works on other platforms with reduced features
Setup difficultyLow. Run one command, open mobile app.Low. Scan QR code in Codex App sidebar.Medium. Install Node.js bridge, run a daemon, scan QR code.
CostIncluded with Claude Pro ($20/mo), Max ($100 or $200/mo), Team, EnterpriseIncluded with all ChatGPT plans, even Free ($0) and Go ($8/mo)Remodex subscription + your separate Codex/ChatGPT plan
MaturityResearch preview (early 2026)Preview, rolling out (announced 14. 5. 2026)Early stage, version 2.0 (3.1k GitHub stars, single primary maintainer)

The big picture

If you use Claude Code, your only option is Anthropic's own Remote Control and Dispatch. If you use OpenAI Codex, OpenAI's native ChatGPT Mobile integration is the default choice, with Remodex as a niche alternative for users who want open source, self-hosted infrastructure, or platform coverage that OpenAI does not yet provide. All three require your host machine (the computer running the agent) to stay powered on.

03Claude: Remote Control + Dispatch

🪞
Remote ControlMirror an existing Claude Code session to your phone. See everything, type commands, approve actions. Session must already be running on your computer.
📤
DispatchStart a brand new task from your phone. Claude spawns a fresh session on your machine. No pre-existing session needed. Push notifications on completion.

Strengths

Zero third-party dependencies. The two mechanisms together cover the full workflow: monitoring existing sessions (Remote Control) and delegating new work (Dispatch). Dispatch's automatic routing between Cowork and Claude Code is unique. Full access to your local filesystem, custom MCP servers (plugin connections), and project configuration is preserved.

Limitations

Both features are in research preview, which means behavior may change. Remote Control requires the local process to keep running. Dispatch requires Claude Desktop to be running on your Mac. There is no free tier for Claude Code.

04ChatGPT Mobile to Codex

Strengths

Deepest feature integration of the three options. You get the full Codex App surface: Computer Use, Appshots, browser automation, multi-host switching, and Remote SSH chaining (phone to Mac to cloud development server). Available on all plans, including Free. First-party OpenAI infrastructure with no third-party relay.

Limitations

Announced 14 May 2026 and still rolling out in preview; not every user has access yet. Currently requires macOS as the host (Windows support announced, not yet available). Computer Use (the feature that lets the agent control desktop apps) is not available in the EEA, UK, or Switzerland.

05Remodex: The Open Source Alternative

Remodex (third-party)

What it controls

Codex CLI only (terminal agent)

Open source

Yes (Apache 2.0, fully inspectable)

Self-hostable relay

Yes (eliminates third-party metadata exposure)

Platforms

iOS primary; Android beta (pairing issues reported); community forks for Windows and web

Cost

$3.99/mo or $29.99/yr (bridge is free, self-hosting is free)

OpenAI native mobile (first-party)

What it controls

Full Codex App (Computer Use, Appshots, browser, plugins)

Open source

No (proprietary relay and app)

Self-hostable relay

No (uses OpenAI's managed relay)

Platforms

iOS and Android; macOS host (Windows host coming soon)

Cost

Included in all ChatGPT plans

When Remodex still makes sense

If you need to audit every line of code between your phone and your agent, if you want to run the relay on your own infrastructure, or if you need a platform combination OpenAI does not yet support (such as a Windows host or a web-based client), Remodex fills those gaps. For everyone else, OpenAI's native integration is simpler and more capable.

06How to Choose

The choice depends mostly on which AI coding agent you already use.

  • Claude Code users: Use Remote Control + Dispatch. It is your only option, and it works well. Remote Control mirrors existing sessions; Dispatch lets you start new ones from your phone.
  • Codex users (most people): Use OpenAI's native ChatGPT Mobile integration. It is included free with every plan, provides the richest feature set (Computer Use, Appshots, browser automation), and requires zero third-party software.
  • Codex users who need open source or self-hosting: Remodex fills a real niche if you need to audit or self-host the relay infrastructure, or if you need platform coverage (Windows host, web client) that OpenAI has not yet shipped.

All three approaches share one important requirement: the host machine (the computer running the coding agent) must stay powered on with the agent process or app running. None of these work with a shut-down laptop. If you need the agent to run independently of any local machine, look into cloud execution options like Claude Code Routines or Codex Cloud, which run on the provider's servers instead.

07A Note on Claude Voice Mode

A question that comes up often: can you use Claude's voice mode on your phone to talk to Claude Code? The short answer is no. Voice mode and Claude Code are separate features that serve different purposes.

What voice mode is: Claude's voice mode (launched in beta on 27 May 2025) provides full two-way spoken conversations. You talk, Claude listens, thinks, and speaks back using generated voice responses powered by ElevenLabs. It supports five selectable voices, hands-free and push-to-talk modes, and multilingual input in beta. It is available on all Claude plans, including Free, on iOS, Android, and the web.

Which model powers it: Anthropic does not officially disclose this. Their support documentation deliberately omits any model name. TechCrunch reported at launch that it was "powered by Claude Sonnet 4 by default." Multiple Reddit users across several independent threads have reported that it later switched to a lighter model (Haiku 4.5) for faster responses, but there is no official Anthropic confirmation of this. The exact model powering voice mode remains uncertain.

What voice mode can do: Web search works during voice conversations (confirmed by Anthropic's official support page). Google Workspace integration (Gmail and Calendar for paid plans, Google Docs for Enterprise) was reported by TechCrunch at launch, but as of May 2026, an Anthropic collaborator confirmed on GitHub that "connector tools are not currently supported" in voice conversations. Web search is the only confirmed integration that works in voice mode today.

What voice mode cannot do: Voice mode does not connect to MCP servers (the plugin system that powers Claude Code's tool access), does not run terminal commands, cannot read or write files, and cannot control Claude Code sessions. It is a consumer conversation feature, not an agentic interface. If you want to control Claude Code from your phone, use Remote Control or Dispatch as described above.

Separately, Claude Code's terminal interface has a /voice dictation command that converts speech to text for typing prompts. This is not conversational voice mode; it just transcribes what you say into the text input. It requires Claude Code v2.1.69+ and a Claude.ai account.

08Security Comparison

Claude Remote ControlChatGPT Mobile to CodexRemodex
Relay operatorAnthropic (first-party)OpenAI (first-party)Managed relay by indie developer (default) or your own self-hosted server
Third-party dependenciesNoneNoneNode.js bridge + relay (both open source)
End-to-end encryptionAnthropic infrastructureOpenAI secure relayYes: X25519 key exchange, AES-256-GCM, Ed25519 signatures, replay protection
Self-hostableYes (relay can run on your own VPS or Tailscale network)
Metadata visible to relayConnection metadata visible to AnthropicConnection metadata visible to OpenAIIPs, session IDs, device IDs visible to relay operator; payloads encrypted

Practical takeaway

Claude and Codex users trust their respective providers (Anthropic and OpenAI) with connection metadata, which is standard since you already trust them with your prompts and code context. Remodex adds a third-party relay by default, but you can eliminate that by self-hosting. The end-to-end encryption in Remodex is well designed: the relay cannot read your messages, only route them.

09Practical Tips

A few things to keep in mind when using phone-based agent control, regardless of which platform you choose.

  • Keep your host machine awake. All three approaches require the computer running the agent to stay on. Disable sleep timers or use a tool like caffeinate on macOS to prevent your machine from sleeping while a task runs.
  • Set up push notifications. Both Claude and Codex send push notifications when a task completes or needs your approval. Make sure notifications are enabled for the respective mobile app, so you get alerted without having to check manually.
  • Start with small tasks. Phone screens are not ideal for reviewing large diffs or complex terminal output. Use mobile remote control for monitoring, quick approvals, and short steering instructions. Save deep code review for when you are back at your desk.
  • Consider cloud execution for truly hands-off work. If you want tasks to run without any machine of yours being on, look at Claude Code Routines (Anthropic's cloud agent) or Codex Cloud (OpenAI's cloud sandbox). These run on the provider's servers, produce pull requests when finished, and do not require your laptop to be open.