Run repeat work in a local container

A developer used always-on local agents on their own Mac to turn a three-prompt task into one that runs without approval steps.

Done with Custom agent· Advanced· Oct 2026

What they did
The poster put each repeating worker into a separate Linux container on the Mac, gave it its own workspace and sign-in, and then let Codex handle it locally on a Plus account. At startup, they checked again that the container could not reach either the Mac itself or the home network, which meant they no longer had to approve every command. For research runs, they had the agent finish by creating a file that captured the run size, how many sources were used, a hash, and a note saying the sources were only enumerated rather than checked.
How it went
The biggest difference was the drop in supervision: a job that once took three approval prompts now takes none. The downsides did not change: the laptop must stay awake, and plan limits are still the main constraint.
Tips
This was only tried with Colima on an Apple Silicon Mac, and the container wall is useful but not perfect. If you try it, keep the job isolated and re-test network access every time it launches.

Try it yourself with OpenClaw

Set up a persistent local agent for this recurring job: every [schedule], do [task]. Save the instructions so it can keep running without being re-prompted, and write the output to [file or folder]. Run one full test now and show me the result. Pause and ask for approval before any irreversible action, including sending, deleting, or submitting anything.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.