How to work with coding agents from your phone
Plan a remote coding workflow with a desktop host, mobile approvals, readable diffs, and a reliable way to reconnect.
A phone is useful for giving a coding agent a bounded task, answering a question, and reviewing a small change while you’re away from your desk. The computer still does the work. Your repository, development tools, and running processes stay on the host you’ve chosen.
A useful remote workflow makes that division clear. Before choosing an app, check where commands run, how approvals reach you, and what happens when either device loses its connection.
Choose the connection that fits your work
If you already use Claude Code, start by checking its native Remote Control, which connects local sessions to the Claude app or a browser. Read the current account and version requirements before adding another service. A companion app makes sense when you need a different mix of agents, editing, or project management.
Some clients connect to agents through the Agent Client Protocol, or ACP. Protocol support isn’t a guarantee that every agent version works with every client. Check the exact combination: phone operating system, desktop operating system, agent, and adapter. Include the agent’s own account or API charges when comparing costs.
Prepare the desktop before leaving
Open the project on the host and run the agent there first. Resolve sign-in, missing tools, dependency installation, and repository access while you have a keyboard. A remote app can’t finish a task if the host lacks the tools the task needs.
Use a branch or isolated working copy for a new change. Record the starting Git status so existing edits don’t get confused with the agent’s work. Choose a project folder with the necessary files, rather than granting access to your entire home directory.
Check the computer’s power and network settings. A session that survives a phone disconnect still needs its computer to be awake and reachable. Don’t assume a notification can wake a computer that’s shut down or get through every workplace network.
Give the agent a task you can review
Small, explicit requests fit a phone better than a broad instruction to improve the whole application. For example:
Find why submitting an empty search shows a loading indicator forever. Explain the cause first. Then fix that behavior, add a focused regression check, and report the files changed. Don’t commit or deploy.
This request says what goes wrong, what to investigate, what the agent may change, and when to stop. Add the steps to reproduce the problem or the error message if you have them. Leave unrelated cleanup out of the task so the final diff stays easy to read.
Ask the agent to tell you when it’s unsure before it takes on more work. If the fix needs a database migration or an account permission you didn’t expect, stop and decide. Don’t approve a chain of actions you don’t understand.
Review commands and changes separately
A permission request asks whether an action can run. A diff shows what changed. You need both. Before approving a command, check which folder it runs in, which files it touches, and whether it sends anything outside your computer. Before accepting the result, compare the actual edits with the original task.
Use this review sequence:
- Confirm that you’re looking at the intended project and session.
- Read the explanation and inspect the changed files.
- Check the relevant test result, including failures and skipped checks.
- Leave committing, merging, and deployment as separate decisions.
A green status is a reason to look at the evidence, not to skip it. For a large diff, or a change that touches credentials, permissions, or production data, finish the review on a larger screen.
Test reconnection with a harmless task
Before relying on the workflow away from home, ask for a small read-only investigation. Disconnect the phone, reconnect, and confirm you can identify the same session and read its latest state. Check what an outstanding approval looks like after reconnection.
If output stops, work out which one it is: the phone disconnected, the computer went to sleep, the agent stopped, or the agent is waiting for your input. Check the computer directly when you can. Don’t send the same instruction again and again: a silent connection doesn’t prove that the previous action never started.
Keep a short recovery note with the host name, project location, agent used, and where to inspect its session. It saves time when several projects or computers are involved.
Where Mono fits
Mono Code lets you open terminals, edit files, and run coding agents such as Claude Code, Gemini CLI, or OpenCode on your computer from your phone. Sessions keep running on the desktop when your phone disconnects. Check that your agent and devices are supported for the job you have in mind. The computer still does the work, and an encrypted connection between your devices doesn’t change how an AI provider handles code sent to its model. Our guide to AI privacy and encrypted sync explains that difference.
Mono is currently on a waitlist. Join the Mono waitlist for updates. In the meantime, use this checklist with your current coding setup, including a built-in remote-control option if it already does what you need.
