The run does not stop because you did.
You started something substantial and left the building. Twenty minutes later it asked for permission to do one thing, and it has been waiting ever since. This is the job Virex Remote was built for.
// The agent stays on your machine. Only the decision travels.
What actually goes wrong.
Nothing dramatic. The run just quietly stops being productive at the exact moment nobody is watching.
You cannot tell stalled from busy
A session waiting for permission produces no output. Neither does a session thinking hard. From a phone with no view into the run, both look like silence.
The answer would take two seconds
Most permission requests are decisions you could make instantly if you could see them. What costs the hour is access, not judgement.
Getting back is the expensive part
The fix requires being physically at a specific machine. That is a scheduling problem dressed up as a technical one.
A live link, and a decision surface.
A daemon on the machine reports the session; you answer from whatever you have with you.
See the run live
Session activity streams as it happens, so you can tell at a glance whether the agent is working, blocked, or done.
Answer what blocks it
Approve or deny permission requests and plan decisions from a phone browser, and the session resumes on your machine.
Redirect without returning
Send an instruction into the running session, or interrupt the current activity when the answer is neither approve nor deny.
A Tuesday afternoon.
The same sequence, with and without a way to answer remotely.
- 01
You start a long refactor and leave
Claude Code is working through a substantial change on your workstation. You have somewhere to be.
- 02
It stops to ask about one action
The session raises a permission request and waits. On your machine, nothing further happens.
- 03
The request reaches your phone
Rather than sitting in a terminal nobody is reading, the pending decision shows up on the surface you actually have open.
- 04
You answer in a few seconds
Approve or deny. If it is heading somewhere wrong, interrupt instead and send a correcting instruction.
- 05
The run continues locally
Your answer travels back down the live link. Claude Code picks up where it stopped, on your machine, under your own model subscription.
When this will not help you.
It solves one specific failure mode, and it is worth knowing whether that is yours.
- Your machine is off or offline while you are away, in which case there is no session to observe.
- You want an agent that runs without any machine of yours involved, which is a hosted-agent product rather than this.
- Your primary coding agent is not Claude Code.
- You are always at the desk anyway, in which case the terminal is already the shortest path.
Common questions
Can I really drive a coding session from a phone?
You can do the parts that block progress: follow live activity, approve or deny a permission request, approve or deny a plan decision, send an instruction, and interrupt. Writing code is still the agent job on your machine.
Does my laptop need to stay awake?
Yes. The session runs on your machine, so the machine has to be running and connected for the daemon to maintain the live link.
Is this a mobile app?
Virex Remote is designed to be used from any browser or device, so a phone browser works. Check the product pages for current surfaces.
What if I answer wrongly from a small screen?
Interrupt is available alongside approve and deny, so stopping the current activity is always one action away.
Which coding agent does this work with?
Claude Code is the primary integration for Virex Remote.
Answer it from wherever you are.
Subscribe to Virex Remote, assign a seat, and install the daemon on the machine your sessions run on.