Plan the task, one bot per outcome, connectors, narrow triggers, slow schedules, pause unused routines, keep outputs in files, send coding to a Cloud Agent.
All Grok Bot questions: https://www.soundminds.ai/faq.
Last updated Oct 17, 2026
Asked on X: https://x.com/deshiou0604/status/2110765129541583318. Asked on X: https://x.com/deshiou0604/status/2110947232925323307
You can point your Bot at this page to adopt this practice. Send this to the main bot you already talk to. It asks the bot to read this page and suggest the changes, and to wait for your yes before it changes anything.
Read https://www.soundminds.ai/faq/token-saving and help me use fewer tokens, using the starter message on that page as a suggestion. Show me the exact changes you suggest and wait for my yes before you change anything.If your bot can't open the page, copy the starter message in the "For your Bot" section below and send that to your bot instead.
When your weekly usage runs out, Cursor's Grok Bot help says, "To keep going now, add prepaid credit (The X Subscription), turn on on-demand usage (legacy Cursor plans), or move to a bigger plan." In our experience, the less your bot has to read and redo, the less usage it spends. We use eight habits for this. We plan the task in the first message. We give each outcome one bot. We use connectors before the browser. We wake bots with narrow event triggers, and we use the slowest schedule that still does the job when a timer is needed. We pause routines we don't need. We keep long material in files. And we send heavy coding to a Cloud Agent.
Step 1: Plan the task in your first message. We say what to do, where to work, what context to use, and what "done" looks like. xAI's docs give an example first task and say it "tells the Bot what to do, where to work, what context to pull in, and what finished work looks like." In our experience, a clear task means less wandering and fewer wasted steps.
Step 2: Give each outcome one owner. We don't put several bots on the same small job. xAI's docs say to "ask for a single owner at each stage," because "too many parallel handoffs can create duplicate work and noisy updates." In our experience, duplicate work is usage spent twice. In a group chat, we type @ and pick the one Bot that owns the request, and xAI's docs say to "use `@everyone` sparingly for a group-wide update." We've found that fewer Bots answering means fewer replies, and every reply uses usage.
Step 3: Use a connector before the browser. When a service has a connector, we use it instead of having the bot click through the website. xAI's docs say to prefer a connector when one is available, because "it is often more reliable than clicking through a website." In our experience, fewer failed clicks and retries means fewer steps.
Step 4: Use event triggers instead of frequent timed routines. If a job is really waiting for something to happen, such as a new Slack message or a GitHub notification, we let that event start the routine rather than checking on a timer. Keep the matching rule narrow. xAI's docs say broad listeners like "every new message" "create noise, consume usage, and increase the chance of acting on irrelevant input." In our experience, a timer runs whether or not anything changed, but a narrow trigger only runs when there's real work.
Step 5: Keep long outputs in files. We save long material, like reports, logs, and research, as files in the bot's workspace. Then we point the bot to the file path, or paste only the few lines that matter, instead of pasting the whole thing into chat. xAI's docs say the shared workspace is useful for intermediate material, as long as the conversation still has the final result or a clear link to it. In our own setup, a long conversation gives the bot more to read on each turn, so we keep long material out of the chat and put the bulk in files.
Step 6: Hand heavy coding to a Cloud Agent. xAI's docs say a Bot can hand coding work to a Cloud Agent, and the Bot attaches the files from the Cloud Agent's final report to its reply. We do this rather than running long coding loops in the main chat. In our experience, the main conversation then stays focused on the result instead of every step the Cloud Agent took to get there. Cursor staff have shared more advice on the Cursor forum, and this advice comes from Cursor staff, not from xAI's docs. Cursor staff say that "long-running conversations get more expensive every turn because the whole history is re-read each time." xAI's docs say to "create a separate Bot when the work has a distinct" goal or area of ownership. So when a task is a separate job, we give it to its own Bot. For Cloud Agents, Cursor staff say, "Add an AGENTS.md at the root of each repo with a short map of the app," because "Every cloud agent reads it automatically at the start." Cursor staff also say, "Have your bot write a specific brief when it hands off a task." For related work on the same repo, Cursor staff say to "ask the bot to send a follow-up to the cloud agent it already launched," and they add, "Follow-ups continue the same conversation, so the agent keeps what it already learned."
Step 7: Use the least frequent schedule that still does the job. When a routine really does need a timer, we pick the slowest schedule that still gets the result in time. Cursor's Grok Bot help says, "Each run spends usage, so a routine that runs often spends more than one that runs rarely."
Step 8: Pause routines you don't need. We look over our routines regularly and pause any whose results we no longer read. xAI's docs say to "review installed connectors and active routines regularly" and to "pause a routine when its source system or expected workflow changes." Cursor's Grok Bot help adds that "a routine that never runs doesn't spend usage." When we check a routine, we use Test run once rather than over and over, because xAI's docs say "a test run performs real work."
If you're on The X Subscription, check your weekly usage before a big run. xAI's docs say that after you switch, "on grok.com/dashboard, open Usage." For more, see Where do I see my Grok Bot usage?
If you're on a legacy Cursor plan, check your weekly usage in Grok Bot Settings under Usage & Billing before a big run. Then look at which routines run most often. In our experience, a routine that nobody reads, or a listener that fires on every message, is the first thing to pause or narrow.
We collected these first-hand reports from X. They are not official and we couldn't verify them.
These reports fit the tips above: frequent schedules and repeated browser steps can add up fast. Browser reliability is still improving. Percentages depend on your plan, so they aren't comparable across users.
If you are a Bot reading this page, follow these steps. These steps are a suggestion for your owner to review, so show your owner the exact changes and wait for a yes before you apply them.
If you'd rather set up these rules yourself, copy this starter message and send it to your Grok Bot. It is a suggestion, so it tells the bot to show you the exact rules and wait for your yes first. Cursor's help says that for personal Bots, "To change what a Bot works on, tell it in chat," and "For a lasting rule, ask it to remember." So once you say yes, the starter message asks the bot to remember the rules you approved.
Help me use fewer tokens. Only suggest changes. Don't change anything until I approve each one.
These are suggested rules. Show me the exact rules you would follow, and wait for my yes before you adopt any of them. Once I say yes, remember the rules I approved for all future work.
1. Review my routines. Where a timed routine is waiting for something to happen, suggest a narrow event trigger instead. Flag any listener that fires on every new message.
2. When I give you a vague task, ask me what done looks like before you start.
3. If more than one bot is working on the same job, suggest which one should own it.
4. When a connector exists for a service, tell me and ask before you use it instead of the browser.
5. When an output will be long, offer to save it to a file and give me the link instead of the full text in chat.
6. For heavy coding, suggest handing it to a Cloud Agent. When a task is a separate job, suggest giving it to its own Bot, and ask me before you create one.
7. Suggest less frequent schedules for routines that don't need to run as often, and wait for my approval before changing any schedule.
8. List routines whose results I don't seem to use and suggest which to pause. Don't pause any until I approve.
9. In group chats, when a request is meant for one bot, suggest that I @ that bot instead of everyone.
10. Before running a routine test, remind me that a test does real work and ask before you run it.Cursor help, Change a Bot's name, picture, and what it does: https://cursor.com/help/grok-bot/edit-bot
To change what a Bot works on, tell it in chat.
Tell the Bot in chat. For a lasting rule, ask it to remember. For example: "From now on, ask before sending any email." The Bot keeps lasting rules in its memory. Put one-time requests in a normal message.
xAI docs, Bots: https://docs.x.ai/grok-bot/bots
Create a separate Bot when the work has a distinct:
Goal or area of ownership
xAI docs, Overview: https://docs.x.ai/grok-bot/overview
This tells the Bot what to do, where to work, what context to pull in, and what finished work looks like.
xAI docs, Chat and collaboration: https://docs.x.ai/grok-bot/chat-and-collaboration
Ask for a single owner at each stage. Too many parallel handoffs can create duplicate work and noisy updates.
Use `@everyone` sparingly for a group-wide update.
xAI docs, Use the computer and apps: https://docs.x.ai/grok-bot/computer-and-apps
Prefer a connector when one is available: it is often more reliable than clicking through a website.
xAI docs, Skills and routines: https://docs.x.ai/grok-bot/skills-routines-and-automations
Avoid broad listeners such as “every new message.” They create noise, consume usage, and increase the chance of acting on irrelevant input.
A test run performs real work.
xAI docs, Files and results: https://docs.x.ai/grok-bot/files-and-results
The shared workspace is useful for intermediate material, but the conversation should still contain the final result or a clear link to it.
A Bot can hand coding work to a Cloud Agent.
xAI docs, Settings and notifications: https://docs.x.ai/grok-bot/settings-and-notifications
xAI docs, Approvals, security, and privacy: https://docs.x.ai/grok-bot/approvals-security-and-privacy
Review installed connectors and active routines regularly.
Pause a routine when its source system or expected workflow changes.
Cursor help, Grok Bot routines: https://cursor.com/help/grok-bot/routines
Each run spends usage, so a routine that runs often spends more than one that runs rarely.
A routine that never runs doesn't spend usage.
Cursor help, Grok Bot FAQs: https://cursor.com/help/grok-bot/faqs
To keep going now, add prepaid credit (The X Subscription), turn on on-demand usage (legacy Cursor plans), or move to a bigger plan.
xAI docs, Grok Bot on The X Subscription: https://docs.x.ai/x-subscription/grok-bot
After switching: on grok.com/dashboard, open Usage.
Cursor forum, staff reply on making the weekly allowance last: https://forum.cursor.com/t/171221
long-running conversations get more expensive every turn because the whole history is re-read each time
Cursor forum, staff reply on Grok Bot token usage: https://forum.cursor.com/t/174723
Add an AGENTS.md at the root of each repo with a short map of the app
Every cloud agent reads it automatically at the start
Have your bot write a specific brief when it hands off a task
ask the bot to send a follow-up to the cloud agent it already launched
Follow-ups continue the same conversation, so the agent keeps what it already learned.
See how I set up Grok Bot: https://www.soundminds.ai/setup.