If you work with Grok Bot for a while, you'll hit it sooner or later: your weekly limit runs out earlier than you expected. And suddenly the question isn't what your bot can do anymore. It's how to be careful with what you've got.

I work with a team of about twenty bots myself. Below is everything I know about making your weekly limit last as long as possible, with the source next to it. When something is in Cursor's official documentation, I say so. When it only comes from Cursor staff on the forum, or when it's simply how I work, I say that too.

How Grok Bot usage works, and why your week can run out so fast

The meter first, because that's where it often goes wrong. "I don't send many messages, so I'm fine" isn't true. Usage isn't a message counter. According to the official explanation for the free trial, credit comes off based on the work your agents do, measured in agent steps and tokens, not in the number of messages. Bigger or longer jobs use more. (Plans and billing)

A Cursor staff member explains it a bit further on the forum. According to him, the weekly percentage is based on cost, not on a raw token count. Every model call is priced on input, output and cached tokens. Steps, tool calls and run time don't count on their own, only through the model calls they cause. (forum, @deanrie, Sep 10, 2026) Note that this isn't in the official help, so treat it as background, not as a hard product rule.

There are two kinds of plans, and what happens when your week runs out depends on yours.

The X Subscription (Premium, Super or Ultra; Plus and Free don't include Grok Bot). Here Grok Bot shares one weekly pool with Cursor and the Grok apps. So on this plan Grok Bot doesn't have a separate bucket. Super and Ultra have a bigger pool than Premium. Once the pool is empty, you continue on prepaid credit, with a manual top-up or Auto Top-up. You can optionally set a monthly cap under Additional usage credits at grok.com/dashboard/billing. You see your usage under Usage at grok.com/dashboard. (Plans and billing, Usage and limits, Usage-based charges)

Legacy Cursor plans (Pro, Pro+, Ultra, Teams), or a SuperGrok or X Premium+ link. Included usage is called the Weekly usage limit in Grok Bot and resets every week. If it runs out and on-demand is on, you keep going on On-demand usage. Cursor bills that, under your On-demand monthly limit (called Spending or Monthly Limit in the web dashboard). If on-demand is off, Grok Bot stops until the weekly reset and you get a message saying you've hit your Grok Bot usage limit. (Plans and billing, FAQs)

A few more things people often mix up:

  • Grok Bot usage is measured on your Cursor account, not in the Grok app. A limit you see in the Grok app, like the "Weekly SuperGrok Limit", isn't about Grok Bot. (FAQs)
  • A Cursor plan and a SuperGrok or X Premium+ link don't stack. Grok Bot uses whichever grant gives more weekly usage. Pro equals SuperGrok and X Premium+, Pro+ equals SuperGrok Plus, and Ultra equals SuperGrok Heavy. (Plans and billing)
  • macOS and iOS share one usage bucket, tied to your Cursor account. (Plans and billing)
  • The weekly limit doesn't roll over. Whatever you don't use this week is gone. (Usage and limits)
  • You see your usage in Grok Bot under Settings > Usage & Billing. On The X Subscription also at grok.com/dashboard, on a legacy Cursor plan also at cursor.com/dashboard under Spending.
  • A cap isn't a hard stop in the middle of a run. A bot that's already working may go a little over the monthly on-demand limit. After that, extra usage stops until you raise the limit or the cycle resets. (Plans and billing, Spend limits)

How many euros or how many tokens a weekly limit holds isn't published anywhere. Only the order is known: Pro has the least, then Pro+, then Ultra. So I'm not going to fill that in.

What counts? According to the same forum thread, everything Grok Bot does itself comes out of the weekly Grok Bot pool: chats, routines, browser and computer use, shell commands on the bot's computer, subagents, connectors and repo work on that machine. One exception: if your bot starts a Cursor cloud agent (the same as on cursor.com/agents), that run goes to your Cursor plan. That also comes from the forum, not from the help.

And then the warning Cursor itself gives officially about routines: "An hourly schedule, a short interval, or a Slack trigger on a busy channel can use a week of usage in a day." Every run costs usage. A routine that never runs costs nothing. A test run does cost usage, because it simply does the work. (Routines)

Routines and triggers: the biggest lever

If your usage drops too fast, look here first, before you move to a pricier plan or turn on on-demand.

1. Check how often your routines run, and pause what you don't need. That's Cursor's own first piece of advice too. Start with frequency, pause routines you don't need, and ask your bot to space out schedules and narrow Slack triggers. Only then look at a higher plan or a higher on-demand limit. (Routines)

2. No short intervals "just to be safe". A routine every hour, every fifteen minutes or on a busy Slack channel can burn a week in a day. For checks, daily, or a few times a day, is usually plenty.

3. Pick the right kind of trigger. Officially there are three: a schedule, an event (including email, Slack, GitHub, Linear, Sentry and PagerDuty) and a webhook. (Routines)

4. Replace "check every hour whether there's anything" with an event or a webhook. @Michael_Fenech_ put it well on November 10: a routine that checks every hour whether there's work uses usage, even when nothing happened. An event or webhook only starts when it actually happens. He's also honest that not every routine should switch. A morning brief often belongs on a schedule. He shares an audit prompt that only reads and asks the bot not to change anything without his approval. @gambrill said something similar the same day: with a webhook you keep usage in check, because the routine does nothing until it's triggered.

5. Let a real event be the start, not a fixed clock. Also from @Michael_Fenech_, on November 5. When he turns off his morning alarm, an iPhone Shortcut sends a webhook and starts his daily brief. No fixed schedule, and he doesn't even have to open the app. Other moments he mentions: leaving work, getting in the car with CarPlay, getting home or an NFC tag. How to build that Shortcut is in his post. The principle is what matters: start at a moment when there's actually something to do.

6. Narrow your Slack triggers. A trigger on every message in a busy channel runs often, and uses usage on messages your bot doesn't need to do anything with. Limit it to a mention, a fixed phrase or one channel. Only new activity counts. (Routines)

7. Put work routines on weekdays and office hours, if nobody uses the output at night or on weekends. This is my own practice, and it follows logically from "every run costs usage". The help doesn't say it literally. And it doesn't apply to everything. A news digest, a birthday check or a security scan can quite rightly keep running every day.

8. Don't let a failed run keep hammering. Routines officially have an hourly limit of 300 runs in total and 200 per routine. Go over that and the run fails with "Automation run rate limit exceeded" and isn't retried automatically. (Routines)

One front door, not a choir of bots

Usage doesn't only go to heavy tasks. It also goes to extra replies.

9. Every bot reply counts, also in a group and between bots. Cursor says it literally: "Every Bot reply counts toward your usage, including replies in a group and messages between Bots. Fewer Bots in a group means fewer replies." There's no switch to turn off messages between bots. You can instruct your bots, and have them remember it, not to repeat each other's answers or to ask you first. (Group chats and Bot-to-Bot messages)

10. Work with one front door. For coordination I talk to one bot, my Chief of Staff. The specialists report to it and I only get the decisions and the blockers. The specialists work one-on-one, not as a choir in a group. That's my setup, not an architecture Cursor prescribes. But with point 9 in mind, it makes sense.

11. One bot per workstream. Don't let three bots redo the same research. My Chief of Staff gives a job to one specialist. This isn't in the documentation either, it follows from point 9.

12. Tell specialists in a group chat to stay quiet when the question isn't for them, and have them remember it. You can officially set a lasting rule, with lines like "Remember: in group chats, reply in the group" or "Don't message other Bots. Ask me first." Memory is per bot and stays when you reset a conversation. (Group chats, What your Bot remembers)

13. Let bots and routines stay silent when there's nothing new. For me that's a fixed rule: "Stay silent when nothing new." It isn't a Cursor feature, but it's a habit that fits "every reply counts" and "every run costs usage".

Effort, scope and browser: what one task costs

Besides how often something runs, how heavy one run is counts too.

14. Keep tasks small and tightly scoped, especially before a long job or a job in the browser. Official advice for the trial period, but it holds just as well afterwards: start small, scope tasks tightly and check your usage before you start a big or long job. One big run can use up a credit in one go. (Plans and billing)

15. You don't pick the model, but you do pick the work. Grok Bot has no model picker, it chooses per task. If a task gets routed to a pricier-sounding model (Cursor itself uses Opus 5.5 as the example), the price per token doesn't change for you, according to Cursor. How fast your usage runs out depends on the kind of work and how much effort the task takes. (Which AI models power Grok Bot?, FAQs)

16. Low effort as the default, high only when the task is worth it. This is my own habit. Cursor does mention effort as a factor, but there's no documented setting for it in Grok Bot. So whether "do this on low effort" in the chat really changes the meter, I don't know for sure. I do it, but don't see it as an official setting.

17. Subagents and extra bots are extra runs. According to Cursor staff on the forum, Grok Bot's subagents come out of the weekly pool. And every extra bot that joins in produces extra replies (see point 9). Use them deliberately, not "throw five at it" by default.

18. Cloud agents are a separate bill. If Grok Bot starts a Cursor cloud agent, that run goes to your Cursor plan, on top of Grok Bot's own turn. That isn't in the help, but Cursor staff repeat it on the forum. It isn't a hidden penalty. It's mainly useful to recognize in your Cursor dashboard, by the row with the cloud agent icon. Mostly relevant if you have bots writing code.

19. Prefer a connector or a focused question over a long browser session, if the app can be connected. Browser and computer use comes out of the weekly pool, and as long as such a session runs, extra model calls pile up. (In the dashboard this used to show as grok-bot-cua.) Apps like Gmail, Slack and Notion you connect with an authorized account, so your bot works with them directly. (Connect apps, Grok Bot computer, FAQs) There's no public number saying browsing costs more per minute. A long, open browser session just usually has more steps, and so more model calls.

Context and memory: no extra baggage in every run

20. Keep a routine's instruction short and focused. According to Cursor staff on the forum, big spikes in a single run often come from routines that carry a long conversation history along. That's something different from "start a new chat to save usage", and that's advice I don't give. The practical lesson: don't let a routine reread a whole archive every time when a short status or one file is enough.

21. Use memory for lasting rules. Memory is kept when you reset a conversation. It's per bot, and a few facts like your name and time zone are shared. Explicitly ask it to "remember this" for rules like staying quiet, working through the front door or asking first. (Memory)

On X, @Marko_Poly wrote on November 5 about a user where a big part of the usage supposedly went to bots rereading their old chat. He also mentioned a Usage & Billing screen that splits the week per bot. That's an observation from the community, not an official measurement. I mention it because people are talking about it, not as a fact.

Measure where your usage goes, and look at it every week

22. Check in Settings > Usage & Billing where your week went, not just how full the bar is. And check the dashboard that goes with your plan. Your included weekly usage is something different from your monthly on-demand limit. (Plans and billing)

23. Have one bot go over the rest every week, advice only, and don't let it change anything without your okay. @CarolinaDill described on November 9 how one bot checks the other bots every Monday and sends her a short list: a routine that runs too often, duplicate rules, a routine that was accidentally left on, a rule that's missing. Nothing changes until she says okay. I do the same, more on that below.

24. Don't set up a routine that checks your usage every hour. Every check costs usage itself. One audit a week, or a reminder around your reset, is enough. That follows from Cursor's warning about routines, it isn't a literal quote.

25. Know your reset moment and your on-demand switch. Your weekly usage resets every week, and the date is shown under Usage or Spending. Don't want to keep paying when your week runs out? On The X Subscription you turn off Auto Top-up or set a monthly limit. On a legacy Cursor plan you turn off on-demand or put a cap on it. (Plans, FAQs, Usage-based charges)

26. A small extra: invite friends. Inviting friends to Grok Bot can earn you a usage reset. You'll then see it under Settings > Usage & Billing, if resets are enabled for your account. How much it gets you, I can't promise. (FAQs, Plans and billing)

How I do it at Full Spectrum Creations

This isn't theory. This is how my own bot team works, in broad strokes.

I work with one Chief of Staff as a fixed front door. Behind it sit specialists, for example for copy, design, development, ads and research. That's part of a team of about twenty bots. The specialists report to the Chief of Staff, and I get the decisions and the blockers. I don't have groups anymore where specialists repeat each other. They work one-on-one, and collaboration goes through the Chief of Staff. It also has the rule "Stay silent when nothing new".

One audit bot looks every week at where the usage went. It works from the report Grok Bot itself provides, doesn't make up numbers, and says so honestly when the report is missing. For each point it says whether it works, whether it's fixed, or whether I need to do something. It doesn't switch anything off itself.

My routines don't all run on weekdays either, and they don't have to. Work routines like a daily brief run on weekdays. Some quite rightly keep running daily or weekly.

What does an audit like that give you? For me it showed that one check running on a schedule cost far more than it was worth. It was a check whether a site is online. I switched it to an email trigger: it now only starts when an alert actually comes in that the site is down. That's exactly the approach I'd recommend. Look for the routine that keeps checking whether there's anything, and let it start only when something happens.

How usage splits across your bots also differs per week, by the way. If I have a week with a lot of design work, the design bot suddenly takes a big share. If I'm building a site, development costs a lot. So look at patterns over several weeks, not at one snapshot.

Where to start

Do you have a Grok Bot team, or are you thinking about one, and do you want to know where your usage is going right now? Get in touch. Then I'll look at your routines and triggers with you. With me, a pricier plan is never the first step. I test all of this myself, with my own bot team.