Grok Bot Ultra users: how do you make the weekly allowance last? Mine reached 99% in three days

Discussions
1

Hi everyone,

I’ve used Cursor for years, and I’m really excited about Grok Bot. It’s the kind of tool I want to build my daily workflow around. But my first experience with Ultra’s usage limits has left me frustrated and wondering what I’m doing wrong.

I upgraded to Ultra at $200/month, and most of my Grok Bot weekly allowance disappeared on the first day. I noticed how quickly it was dropping and reduced my usage, but it still reached 99% by the end of the third day, leaving me waiting several days for the reset.

Support checked my account and gave me this breakdown:

  • Day 1: 65% of the weekly allowance used
  • Day 2: 84% used
  • Day 3: 99% used

They confirmed that the upgrade applied correctly and said the usage came from activity after the upgrade. Most came from Grok Bot Default, with a smaller share from automations.

For context, my export for days 2–3 alone showed around 284 million Grok Bot tokens, approximately 84% cache reads. That’s only part of the week, and I understand raw token counts don’t directly translate into the allowance.

From my perspective, I wasn’t running a huge bot operation or doing unusually intensive work. But clearly, my idea of “moderate usage” and what the allowance supports are quite different.

How are other Ultra users making this work?

  • Does your Grok Bot allowance usually last the whole week? What kind of work are you doing, and how many bots are active?
  • Have you also seen most of a week’s allowance disappear in a day?
  • What changes made the biggest difference: shorter conversations, fewer routines, tighter tasks, or something else?
  • Do you mainly use Grok Bot as a manager and delegate implementation to Cursor Agents?
  • How do you avoid spending heavily on background checks, status updates, and bots coordinating with each other?

Support explained that Grok Bot has its own weekly pool. When it delegates to a Cursor Agent, the Bot’s coordination consumes Grok Bot usage, while the Agent’s model calls consume the separate Cursor allowances. I’m considering making that the basis of my workflow, but I’d love to hear concrete examples from people who have made it work.

For the Cursor/Grok Bot team: could you share a practical guide to using Ultra sustainably?

For example, a recommended setup for someone working daily, how to configure routines economically, when to start fresh conversations, and how to spot an expensive session before it consumes most of the week. “Keep sessions shorter and review routines” is a starting point, but I’m struggling to turn that into a predictable daily workflow.

I’m a bootstrapped indie developer without outside funding. I already spend around $600/month across $200 Max/Ultra subscriptions to Cursor, Claude, and Codex. That’s a substantial investment for me, and additional on-demand spending isn’t something I can comfortably absorb.

The frustrating part is that I want to use Grok Bot more. I like it a lot. I upgraded because I hoped I could rely on it throughout the week, and now I’m hesitant to give it work because I don’t know how quickly the remaining allowance will disappear.

If you’ve found a sustainable way to use it, please share your setup. And if you’re experiencing the same thing, I’d appreciate hearing that too.

What am I missing, and what should I change?

5

Hey @olegdater,

Grok Bot bills for the work the bot does (tools, browser use, subagents, and re-reading the conversation each turn), not the number of messages you send. Your ~84% cache reads is the tell: long-running conversations get more expensive every turn because the whole history is re-read each time.

Biggest wins, in order:

  1. Start a fresh conversation per task. A long session is far pricier per turn than a fresh one doing the same work.
  2. Space out routines. A routine every few minutes fires ~100x/day. Prefer hourly or a few times a day, and run recurring jobs on their own dedicated bot (not your main chat).
  3. Delegate implementation to Cursor cloud agents. The manager pattern you described works: the bot’s coordination is light and the heavy execution runs on your separate Cursor plan allowance, so it doesn’t eat the weekly pool.
  4. Scope tasks tightly and check the meter before big jobs (you can even ask the bot to audit its own usage).

More on how the allowance works: Grok Bot plans & billing.

6
  1. Start a fresh conversation per task. A long session is far pricier per turn than a fresh one doing the same work.

how do you do it? As I understand 1 Grok Bot is one big long conversation, how do you start a fresh conversation with the bot? makes no sense to me?

7

Click +, bottom left, then click your Bot of choice.

9

fresh conversation per task on grok bot? How is that even possible and doesn’t that defeat the purpose? In Cursor this makes sense but not in Bot

10

OMG, where did you learn it? and where I can learn all the tricks like this

I did not expect you could have multi chats with same Bot, it does not fit human pattern so I did not get it, thanks for pointing out

11

Oleg, that Day 1 65% spike then creeping to 99% is brutal. One pattern that helps: treat long-lived bots like leaky sessions. Cap routine frequency, force a fresh chat when context is mostly cache reads, and move heavy implementation to short-lived Cursor Agents so the bot only coordinates. Happy to compare what your day-2/3 export looked like if useful.

12

I also didn’t know that you could start a new chat with a specific bot, I’ve just found the (+) button you’re referring to. But then how do we clear the original long chat with each bot? Mine are enormous.

Thanks a lot!