Skip to main content
Workers in Flowfield are AI agents that implement tasks in isolated Git checkouts. The queue controls which tasks run and when — it starts paused, giving you full control over when execution begins. You move tasks into position, enable the queue, and Flowfield takes care of scheduling and running each eligible task in order.

​
Start the Queue

To start executing tasks, move the task you want to run to Up Next in the board, then enable the queue. You can do this from the browser by clicking Run Queue, or from the CLI:
Once the queue is enabled, it stays enabled. New tasks that become ready — because their dependencies are satisfied and they’re in Up Next — start automatically without you needing to re-enable the queue each time.

​
Task Readiness

A task can only start if all of the following conditions are met:
  • The task is in Up Next
  • All prerequisite tasks are Done and their code has been delivered
  • The worker environment has been validated
  • There is a free worker slot available
Tasks that don’t meet every condition remain in Up Next but don’t start until they do. Check flowfield task show APP-1 to see a task’s current readiness state.

​
Monitor Progress

The task feed shows worker activity in real time as it happens — each step, question, and status change appears there as the worker runs. The browser’s Connected indicator shows whether the live update stream is active. If questions or interventions are waiting for your attention, they appear in the Needs You panel on the board.

​
Answer Questions

Workers may ask questions during execution when they need clarification or have reached a decision point. Questions appear in the task feed and are also collected in the Needs You panel. Answer them directly in the task feed — your saved answer lets the service continue managed work when the queue permits it. To review and answer questions from the CLI:
Answering a question does not automatically enable a paused queue. If you paused the queue before the question appeared, you need to re-enable it separately after answering.

​
Pause and Stop

You have two levels of control over running work: Pause the queue to prevent new tasks from starting while leaving any active workers running to completion:
Or click Pause Queue in the browser. Active workers continue and finish normally — pausing only prevents the queue from starting new attempts. Stop a specific worker to halt a running attempt immediately:
Or click Stop Worker in the browser next to the running task. To inspect attempts before stopping them:

​
Worker Capacity

Each project starts with a single worker slot, meaning only one task can run at a time. To allow two independent tasks to run simultaneously, increase the limit in the browser:
  1. Open your project in the browser at localhost:8765
  2. Click the Workers settings for your project
  3. Set Maximum Parallel Workers to 2
Each worker uses its own isolated Git checkout, so they don’t interfere with each other. Dependent tasks always wait for their prerequisites to finish regardless of available slots — parallel capacity only benefits tasks that are genuinely independent.

​
Choose a Worker Model

The model and reasoning effort used by workers are configured per project. You can set them by asking your coordinator in a Codex conversation, or by adjusting the Workers settings in the browser. To see which models are available for your project:
Model capability is determined by the harness adapter, so available options depend on which models Codex supports in your environment.
Closing your Codex coordinator conversation does not stop the service’s enabled queue. Workers continue running as long as the service is running. Use Pause Queue or flowfield project workers pause if you want to halt new starts.