A Play keeps looping and never writes anything (the Flywheel, the lead engine, Library Radar)

You started something and it will not end. One performer put it this way: "My lead engine keeps looping with the same notifications, nothing gets made, and I'm using a lot of tokens". Another sent a screenshot of the Flywheel where every single line read "Reading your files", then wrote back the next day: "processing flywheel still uses too many tokens". A third tried the bounded version instead and got nowhere: "I ran the Library Radar and it did the same thing", then "Kept looping, asking me what county I was in", and after an update that was supposed to fix it, "Nope. Still looping." If any of that is your week, this article is the whole picture: there are two different loops with two different causes, and the escape is different for each.

First, whatever else you do: stop the run yourself

Whenever Otto is working, a bar at the top shows what it is doing and a button to end it. The button reads Turn off when Autopilot is running and Stop when it is a single job. Press it. The current run ends right away and no more of your Claude usage is spent on it.

Be clear about why this is on you. Nothing in Booked Solid will stop a looping run because of what it has spent. There is no spend limit, no budget, no automatic pause. A run that is circling will keep circling until it finishes, until your own Claude plan runs out of usage, or until you press the button. If you have watched it read the same files three times, you already have all the information you are going to get. Press Stop.

What you must not do is sit there tapping Keep going. It feels like progress and it is the opposite. Every resume drops the run straight back into the same loop it was already in, and each pass costs you again. One performer tapped it roughly fifty times and got nothing out of it. If a run has stopped twice without producing anything, it is not going to produce anything on the third try either. Stop it and come back down this page.

What the Keep going button means today

The Keep going button on the Activity screen now has exactly one meaning: your own Claude plan hit its usage limit. That is the only situation in which it appears. The status line next to it says so: "paused at your Claude plan's usage limit - run it again after the limit resets and it picks up from its log".

So if you see it, the stop did not come from Booked Solid. It came from Anthropic, on your own plan, because your usage window ran out. Waiting for the window to reset and running the job again is the right move, and the run picks up from its log rather than starting over. That is different from a loop, and it is not a fault.

If your run is looping and you have no Keep going button, nothing has paused it and nothing is going to. Press Stop.

There used to be a spend brake. It is gone

If you have read older support replies or an older version of this page, you may be waiting for a safety net that no longer exists. Between 24 July and 4 August 2026, Booked Solid watched what a run spent and interrupted it: a heads-up around 100k weighted tokens and a hard stop around 200k on a focused job, with much larger numbers on a full Autopilot sweep. It was removed outright in version 1.11.42. None of it is in the app any more.

It was removed because the line could not be drawn anywhere honest. 200k was chosen as the allowance for one narrow typed request, on the reasoning that a one-liner costing that much has gone wrong. The same line then applied to a Play, and three complete runs of one working Play on the same desk measured 100k, 205k and 204k. An ordinary successful run sat right on the line and got stopped about half the time. A brake that fires at the normal cost of a working feature is not protection, it is a fault you have to clear.

Two consequences you should plan around:

• The Settings control When to stop and ask, with its Recommended, No limit and Custom choices, was deleted along with the brake. If you go looking for it, you will not find it. There is nothing to configure.
• Your spend is still measured to the token and still shown to you. Open Advanced in the top bar, then press Token history to see where your usage actually went, run by run. Measuring did not go away. Interrupting did.

For what actually spends your Claude usage, see the article on your Claude usage limit in this help center. And to put one worry to rest: emailing support costs you zero Claude tokens. The help desk and Otto are separate systems.

Which of the two loops do you have?

• Every line on the Activity screen reads Reading your files, the token number climbs, and no draft is ever written. That is loop 1.
• The run pauses to ask you something, you answer, it resumes, and "it keeps asking me the same question". That is loop 2.

Loop 1: the Flywheel says "Reading your files" and writes nothing

The symptom is unmistakable once you look at the Activity screen: under The Flywheel, every line reads Reading your files. The token number climbs and climbs, and there is still no thank-you, no review ask, no referral note and no draft anywhere. This is a confirmed defect on our side. It is not something you did, and it is not your Claude plan.

Here is the honest cause. On a busy workspace, a Flywheel launched broadly has to work out which finished gig it is supposed to act on, so it starts reading a large chunk of your workspace to orient itself. On a desk with real history in it, that reading alone can run for a long time and spend heavily before it ever reaches the thank-you or the review. The job should cost a fraction of that. And since nothing interrupts it on spend any more, it will keep reading until you press Stop.

Escape 1: point the Flywheel at one gig, and tell it not to read anything else

1. Press Stop on the broad run if it is still going.
2. In chat, give Otto one boxed-in instruction in this exact shape, with your own gig name: For the [gig name] gig only, write the thank-you draft into that gig's folder. Do not read anything else first.
3. When that one lands, ask for the review the same way. Then the referral. Then the rebooking. One at a time, one message each.

That last sentence is the part that matters. Do not read anything else first is what short-circuits the reading loop. Each of these narrow asks should finish quickly instead of circling.

One old instruction here no longer applies

Earlier versions of this page told you to quit and reopen the app before starting the Flywheel, so it would not inherit a bigger token allowance from an Autopilot sweep that had just run, and to watch the pause card for an offer to carry on "for another 1000k". There are no allowances any more and no pause card, so there is nothing to inherit and nothing to spot. Restarting the app does not shorten the reading loop. Stop the run and use the one-gig instruction above instead.

Before you re-run anything, check what is already banked

Drafts are saved as Otto works, and the Flywheel does the thank-you first, then the review, the referral and the rebook. Look in leads/<gig-slug>/ inside your BookingHQ folder, and at the Board view on the Activity screen. Quite often part of what you were waiting for is already written and you do not need to re-run the whole thing.

Loop 2: the lead engine or Library Radar re-asks the same question forever

This one has a different shape. The run pauses to ask you something, most often which US county you are in. You answer. It resumes. It asks the same question again. The same notification comes back, nothing is ever produced, and it never finishes. Both Set up my lead engine and Run Library Radar do it.

This was a genuine bug on our side, not anything you did. When a run paused to ask a question and then resumed, it could lose track of what it was already doing and start the task over from the top. The Library Radar instruction carries a line telling Otto to ask which US county to scan if you have not already said it, and that line only works while the run can still see the conversation where you answered. A resume that could not see that history read the instruction fresh, had no idea you had already told it Oakland County, and asked again. Answering more clearly would never have helped.

What Library Radar actually is

It is one of the one-tap jobs in the Get set up group on the Plays tab, launched with Do it for me. It reads a US public libraries dataset that is bundled inside the app: offline, free, nothing to connect, so that part cannot hang or time out. It asks for your state and county, then hands back a ranked list of the public libraries in that county worth approaching, ordered by how much children's programming each one runs, with a market-rate fee band. For the strongest prospects it then searches the live web for the youth-services or programming contact and prepares a short outreach note for each. Every note stays a draft you approve. Nobody is ever contacted automatically.

Missing data was ruled out early, so do not go looking there. Michigan alone carries 666 library branches in the bundled dataset, and Oakland County on its own returns 40 of them. If Library Radar loops on you, it is the pause-and-resume bug, not your location.

The fix, and the version number that will trip you up

The fix is real and it shipped. The trap is the version number. It was first announced as 1.11.23, and that build never published: it failed on the way out. So did 1.11.24. Anyone who went to update after that announcement got nothing and stayed on the same broken build, tested it in good faith, and quite reasonably reported that it was still broken.

The version that carries the fix is 1.11.25. If you are on 1.11.20, 1.11.21 or 1.11.22, you still have the bug no matter what you were told.

1. Quit Booked Solid completely, not just closing the window, then reopen it. It checks for a newer version on every launch and downloads it quietly in the background.
2. If nothing lands, click Check for updates in the top bar. When it tells you the new version is ready, press Restart & install.
3. If the update still will not land, reinstall over the top from booked.kivimedia.co/download/win, or /mac on a Mac. Your leads, settings and license carry over.
4. On Windows, the blue "Windows protected your PC" screen means our installer is not signed yet, not a real warning. Click More info, then Run anyway.
5. Confirm the version before you conclude anything is still broken. Open Settings > General and read the Updates section: it tells you which version you are running. It must say 1.11.25 or higher.

What 1.11.25 changed, in plain words

• Your county is remembered the moment you say it, and the instruction is rewritten to state it outright, so there is no question left in the run for it to re-ask.
• A resumed run is now told plainly that anything you have already answered stays answered.
• Keep going always picks the same run back up, instead of sometimes starting a fresh one that had forgotten everything.
• Every setup run can now finish even when a step will not work. Instead of circling on the thing it cannot do, it writes down what blocked it, moves on, and ends with a short Needs you list. That circling is what was eating the tokens.

If you cannot update yet: the narrow single-pass escape

In a fresh chat, not by tapping Keep going on the stuck run, paste one message like this, with your own county and state:

Run Library Radar for Oakland County, Michigan. Just give me the ranked list of libraries from the built-in data. Skip the web search for contacts and skip the outreach drafts for now.

The ranked list comes from the offline data, so that part should finish in a single pass and never pause, which means it cannot loop. Ask for contacts and outreach drafts afterwards, in separate small messages.

An honest warning about this one: on the buggy builds it still looped once it reached the contacts step. The performer who tried exactly this wrote back "It looped asking for the contacts." Treat it as a partial escape that gets you the list, not a cure. Updating to 1.11.25 or later is the actual fix.

Two things that look like the answer and are not

Support offered both of these and both failed on the customer. Skip them.

• Swapping the open-ended setup for the bounded Play does not help. Running Run Library Radar instead of Set up my lead engine looped in exactly the same way, on the same question. The loop lives in how a paused run resumes, not in how broad the job is.
• Updating to 1.11.20 does not help. The county re-ask survived that build. If someone told you 1.11.20 fixed this, it did not.

When one card keeps coming back

If a specific item returns on every pass, take it out of circulation.

1. Open the Activity screen and stay on the Board view. The cards that need you sit at the top.
2. Find the card that keeps repeating.
3. Hover that card and press the X on it, the one labelled "Clear this off the board". A cleared item is kept in your record for the audit trail, but it disappears from every board view and is never picked up again, so the loop cannot restart from it.

One thing that catches people: a job that is running right now cannot be cleared. Press Stop first, then clear the card.

One more cause worth ten seconds, and no more

A play that needs the web and cannot reach it will also come back on every pass without finishing, because it never reaches its done mark. Open the Connections screen and look for anything that needs reconnecting, and check the browser connection there if the play needs the web. It is worth a quick look. It is not worth your afternoon: neither of the two loops above was caused by a connection, and this used to be the only cause this article named. If nothing on Connections is complaining, come back up the page.

Getting back to normal afterwards

Turn Autopilot off while you sort this out, then start one play by hand and watch it on the Activity screen. If that single play finishes cleanly and writes something real, turn Autopilot back on. Running one thing at a time is also the cheapest way to work: a narrow ask finishes fast, while one big "do everything" request is exactly the shape that wanders. Since nothing will interrupt a wandering run for you, keeping each ask small is now your main protection, and watching the Activity screen for the first minute of a new play is worth the minute.

Which versions this applies to

This page describes Booked Solid 1.11.42 and newer. The current shipped version is 1.11.43. On these builds no run is ever interrupted by what it has spent, there is no spend setting to configure, and Keep going only ever means your own Claude plan reached its limit.

To see which version you are on, open Settings > General and read the Updates section: it says "You're running Booked Solid" and then the number. If it is older than 1.11.42, update before you use this page - quit the app completely and reopen it, and it updates itself. If this page does not match what you see, tell us in a ticket so we can update it.

If it still loops

If you are on 1.11.25 or later and a run still re-asks a question you have already answered, that is a bug that survived the fix and we want to see it rather than guess at it. Turn on Support logging in Settings > General, reproduce the loop once, then generate a support bundle from the same section (the support bundle article in this help center has the full steps). Attach the bundle to your ticket along with the version number from Settings > General, and, if it is Library Radar, your state and county.

Still stuck? Submit a ticket from this help center and we'll take it from there.