The Runner and the overnight queue

Agent tasks that run on your machine, on your repository, and one morning report.

  • A task has an owner (an agent or a person), a model, effort, priority, a budget estimate and "Run tonight".
  • The Runner, in the Mac app or the command line tool, takes queued tasks, runs your coding agent (for example Claude Code or Codex) on your repository, takes before and after screenshots and opens the pull request.
  • It reports cost and result on the task; every step also shows in the linked threads.
  • In the morning you get one report: what was done, what failed, the cost, and which flows changed.

Agent tasks and tap-to-edit jobs run on a Runner of yours today: the Mac app or the command line tool on your own machine. Each Runner takes only the work it can check: iOS edits need a Mac with Xcode, Android edits the Android SDK, web edits any machine with Node. There is no hosted Runner yet, so keep one online for edits to leave the queue.

The command line

The same Runner runs from a terminal on macOS or Linux. Sign in once; the token is kept in the Keychain (or the system's secret store). Code edits always run on your workspace's own AI key.

CommandWhat it does
redly loginSigns in through the browser and keeps a runner token for this machine.
redly login --token <token>Keeps a runner token made in Settings, API tokens (for CI or a machine without a browser).
redly runner startTakes queued tasks and tap-to-edit jobs until stopped. --slots <n> runs several at once, --agents claude_code,codex,api picks the agents, --once takes what is queued now and exits, --shots off takes no screenshots.
redly runner statusThis machine's Runner: online or not, what it runs, what it can do (agents, web, iOS, Android).
redly runner stopGives running tasks back and stops.
redly reportThe morning report: what the night's runs did and what they cost.
redly usage import <file>Adds model calls made outside a run to the workspace's AI spending.