Teach a journey, then start watching
Every new journey gets an attended first run. You watch what Journeys does, correct it when necessary, and approve a successful path before any scheduled run is allowed.
Run the Draft
Open a Draft journey and choose Run. If the journey requires variables, Journeys asks for values before the run. You can override a default for this attempt without changing the saved default.
The run page shows the browser and Activity as the agent works. A run you started yourself has controls to:
- Pause before the next browser action.
- Resume a paused run.
- Send a short piece of guidance when the agent needs help.
- Stop a queued or running attempt.
Guidance is recorded in Activity and marks the attempt as assisted. Scheduled, CI, dependency, and named-chain runs cannot be guided; their result must be unassisted to be useful monitoring evidence.
Do not leave a run paused indefinitely. Journeys eventually stops a run that uses its pause allowance, releases its browser slot, and keeps the evidence already collected.
Review the result
If the run fails, use its Activity, screenshot or recording, and diagnostics to understand where it went wrong. Choose Edit journey when the definition needs work, or Retry when the instructions are sound and the path needs another attempt. A failed learning run cannot be watched.
After a pass, Journeys shows the memory learned from the run and asks whether it followed the path your product intends. Choose:
- Retry when the result happened to pass by the wrong route.
- Yes, start watching when the route and outcome are appropriate.
Approval is for the run and its path, not for each sentence in memory. Choosing Yes, start watching opens the schedule dialog; saving that dialog approves the journey and arms it together.
How memory works
Memory is practical, fallible advice attached to one journey. It is grouped into what worked, what to avoid next time, orientation around the product, and What you told it. The next run receives all current lessons as advice, not as guaranteed truth.
Later passing scheduled runs can add or refine lessons without asking for approval again. A failing run does not rewrite model-generated memory. Editing the journey also does not silently discard it.
Guidance you send during a manual run is also offered to later runs. Journeys keeps a correction only when it is durable advice; guidance that names a page element, selector, or one-off value still helps the live run but is not saved. The Activity feed tells you when guidance was not kept. Guidance can be kept even when that run fails or is stopped, because it records what you said rather than what the failed trace inferred.
If a later run follows saved guidance and shows that it no longer works, the Memory section marks the guidance Stopped working and links to that run. It does not remove your words or hide them from the next run. A later pass clears the warning; remove the guidance yourself when it is no longer useful.
Open the journey's Memory section to remove one obsolete lesson or piece of guidance, or choose Clear all. Clearing all preserves run history, but the next run is cold and the journey must learn successfully before it can continue unattended.
Retry instead of approving a lucky pass
A pass is not enough if the agent used a route that is unreliable, unsafe, or different from the customer path you mean to monitor. The attended run is your opportunity to make that distinction.