Grok Bot vs Muse vs ChatGPT Dots: Which Personal Agent to Try First
Compare Grok Bot, Muse and ChatGPT dots on accounts, price, country rules and tasks, then pick the first personal agent to try.
Grok Bot, Muse and ChatGPT dots are personal agents: apps that keep working in the cloud after you close them. For a reader in China who uses Codex, Grok Bot is the one to try first.
Grok Bot starts at $20 a month and lists no country limit.
None of the three lists China, so check sign-up before paying.
One reader asked which of the three to start with. They live in China, use Codex for most daily work, and have tried none of them. All three reached the public in seven weeks: Grok Bot launched in beta on August 11, Muse on September 8, and OpenAI's dots on September 29.
Grok Bot does not need an X account, and other naming traps
Grok Bot is a separate app from SpaceXAI, the company formerly called xAI. It is not the @grok account on X and not Grok on grok.com. You sign in with a Cursor account, or link a SuperGrok or X Premium+ plan to one, as Cursor's plans page explains.
Muse is Meta's agent. Meta describes it in its launch post as "a personal AI agent. It doesn't just answer questions, it actually does the work." Press and search results often write "Meta Muse". The agent's own name is Muse.
OpenAI calls its product "dots". Each user has "a dot". People search for "ChatGPT Dot". OpenAI's launch post uses the plural.
One more trap: xAI also has "Automations in Grok", a different feature for scheduled tasks in the Grok app. It is not Grok Bot.
Words used below
- Bot: one named agent inside Grok Bot, with its own job and its own chat.
- MCP server: a standard way to give an agent access to a tool or service.
- Cursor: a company that makes an AI code editor. Its account is the login for Grok Bot.
- SuperGrok and X Premium+: paid plans from xAI and X. You can link one to a Cursor account to use Grok Bot.
- Accounts Center: Meta's setting that joins your Facebook, Instagram and Threads logins.
- Plugin, connector: a link from the agent to another app, such as Gmail or Slack. Grok Bot keeps these in a Marketplace list inside the app.
- Routine: a saved job with a schedule or a trigger, such as a new Slack message.
- Skill: saved instructions for a job the agent repeats.
- EEA: the European Economic Area.
- Prompt injection: hidden instructions in a web page or email that trick an agent.
What the three products compare on
Table: Grok Bot, Muse and dots on one page
VibeCoding.Builders read every figure on 2026-10-12. All three can keep working in the cloud while you are away. Tasks that use your own computer need it switched on.
Grok Bot has no plan of its own: paid access comes with the Cursor or SuperGrok plan, and the weekly allowance has no published number. When it runs out, Cursor can bill extra usage on demand if you turn that on. Muse's free limit is also unpublished. For dots, OpenAI has not yet published usage terms for after the first month, and Codex tasks a dot starts count toward your Codex limits.
No company compares itself with the other two, so the "better at" row comes from independent write-ups. Read it as opinion.
Readers of VibeCoding.Builders' earlier setup guides, dots on October 5 and Grok Bot on October 6, should know three things those guides did not cover. VibeCoding.Builders found them when it re-checked both on 2026-10-12. Texting a dot is now a limited beta for Pro users in the US. Grok Bot's plans now also include X Premium+, and its app now covers Windows, Linux and Android.
Run the same three tasks on all three
The three tasks below are daily news, a reminder, and a read-only inbox check. Each one shows the setup steps for every product and what the docs say you should see.
No source shows a finished result from any of these exact prompts. The "you should see" lines come from each product's documentation. VibeCoding.Builders has not run them.
Task 1: A daily news summary
Use this prompt, written for this article, in all three:
Every weekday at 8:00 AM in my time zone, find the five most important news stories about AI agents from the last 24 hours. For each one, give a headline, one sentence on why it matters, and a source link. Do not send anything to anyone else. If a source is unavailable, say so instead of using old results.
Table: The daily news task, product by product
| Product | Steps | You should see |
|---|---|---|
| Grok Bot | 0. Install the desktop app and sign in with Cursor. 1. Create a Bot with a name and one job. 2. Run the prompt once and fix the format. 3. Ask it to save the method as a skill, then create the routine. 4. Press Test run, which does real work. | A brief in the Bot's thread each morning. The routine shows its next run and recent runs under View conversation details. |
| Muse | 1. Open Settings, then Permissions, and choose "Always ask". 2. Send the prompt. 3. Confirm the schedule. Meta's help pages name a morning news digest as a skill. | A confirmation, then the digest as a chat message. One tutorial says scheduled items are listed under Upcoming, but one reviewer could not see a scheduled task in the desktop app, though it showed in the web version. |
| dots | 1. Create your dot in the ChatGPT desktop app. 2. Send the prompt and add "Confirm the schedule." 3. Open the Scheduled tab in the dot's profile. | A schedule confirmation and an entry under Scheduled that you can pause or move. |
The three products set up the same job in three different places. Grok Bot adds a skill step before the schedule.
Grok Bot's docs give a rule for this order: start with a one-time task, make it reliable, save the method as a skill, and only then automate it. The rule matters because routines can fail.
On October 8, Cursor staff reported that "scheduled routines in Grok Bot were delayed and then marked as Failed during a period of very high load" and that missed runs would not replay automatically.
A routine may also pause: Grok Bot may ask whether to keep routines running after a long absence and pause them if you do not answer.
A tester of dots reached a different limit. One reviewer saw a scheduled job run only once in a two-day test, so a week of runs is still unproven.
Task 2: A reminder
Use this prompt in all three:
Remind me to call the dentist tomorrow at 10am.
Table: The reminder task, product by product
| Product | Steps | You should see |
|---|---|---|
| Muse | 1. Open Muse. 2. Send the prompt. 3. Check the time Muse confirms. 4. Open Upcoming to see the reminder. | Meta lists this exact prompt. Muse confirms the details and the exact time, then sends a message at that time. |
| dots | 1. Open your dot in ChatGPT. 2. Send the prompt. 3. Open the Scheduled tab. 4. Pause or move the reminder if needed. | OpenAI says you can ask your dot to set a reminder or run a recurring check. The reminder appears in the conversation. |
| Grok Bot | 1. Open a Bot. 2. Send the prompt. 3. If it offers a routine, check its time and owner. 4. Delete the routine after the call. | A routine proposal if Bot accepts the job. If none appears, that is the answer. The routines page describes runs on a schedule or after an event, and VibeCoding.Builders found no one-time reminder feature in the pages it read. |
Muse and dots document reminders. The Grok Bot pages VibeCoding.Builders read show only scheduled routines.
A reader who mainly wants reminders may do better with Muse or dots.
Task 3: A read-only inbox check
Use this prompt, written for this article, in all three:
Read my unread email from the last 24 hours. List the three that need an answer today, with one line on why each. Do not reply to, send, or delete anything.
Table: The inbox task, product by product
| Product | Steps | Permission to check |
|---|---|---|
| Grok Bot | 1. Open the Marketplace in the app and add Gmail. 2. Create a Bot, or paste the read-only "Inbox Manager" Bot from xAI's work guide, which never speaks as you and waits for you to type "send". 3. Send the prompt. 4. Read the result before enabling any routine. | Settings, General, Auto-review: rules marked "Ask first" always stop the action |
| Muse | 1. Open Settings, then Connectors, and connect your email. 2. Set Permissions to "Always ask". 3. Send the prompt. 4. Read the list. Connecting email lets Muse search your email, but it won't download the whole inbox unless you tell it to. | Settings, Permissions: Allow once, for this task, for this site, Always allow or Deny |
| dots | 1. Open the dot's plugin settings and connect Gmail. 2. Allow reading without allowing sending. 3. Send the prompt. 4. Read the list. | Custom rules: Take action without asking, Take action when you say so, Ask before taking action, or Hand off to you |
Only dots document a read-only email permission. Grok Bot's read-only comes from the Bot's prompt, and its connectors are shared by every Bot. Muse relies on approvals. Check the setting and the first run.
You should see a short list of three items.
Expect errors. In one ten-task test of dots, the inbox result scored 7.5 out of 10, and the reviewer wrote that the agent "keeps treating a closed issue as an open urgent task".
Tasks that fit one product best
The next nine tasks each name a product. Each uses a feature the other two document less well or not at all. A prompt marked "adapted" is shortened or reworded from its source.
Grok Bot: a daily scout of your X bookmarks
Source: xAI's Grok Bot 101 guide. The prompt is adapted from the guide's Bot description.
Look at my X bookmarks each weekday and pick one technology worth demoing. Draft a prompt in my style and wait for me to approve or deny it.
- Connect the X plugin from the Marketplace.
- Create a Bot with a name such as "Tech Demos" and paste the prompt as its description.
- Run it once by hand and correct the draft.
- Ask the Bot to turn it into a weekday routine.
You should see a drafted prompt waiting for your approval. In the guide's full version, an approval starts a Cursor cloud agent that builds the demo.
Grok Bot: listening for what people say about your product on X
Source: xAI's X announcement of August 29 says you can connect X, "then ask a Bot to search posts, read your timeline, check mentions, or pull together what's happening on X". On October 7, the @bot account posted that Grok Bot "can now search, read, and monitor X".
Adapted from the announcement's wording:
Search X for posts about <my product> from the last 24 hours. Give me the ten most useful, with author, link and one line each. Read only. Do not post, reply or like.
- Connect your X account in Grok Bot.
- Send the prompt.
- Check the ten items against the links.
- Ask for a daily routine once the output is useful.
You should see a "working" status, then a summary. Whether a Bot can post is unclear. The official text lists searching, reading and bookmarks, and none of those is posting.
Grok Bot: a private personal CRM from the people you follow
Source: Grok Bot 101, with the prompt adapted from the guide. Its author followed about 800 to 900 accounts and built this from a phone in 10 to 15 minutes.
Turn the people I already follow on X into a private personal CRM in Notion, using only public profile information.
- Connect Notion and X.
- Send the prompt.
- Open the Notion page when the Bot finishes.
You should see a Notion board with each person's profile image, description and a link to the public profile.
Muse: patterns in your saved Instagram reels
Source: a creator's video of October 1. This is one creator's account. It works through Muse's Instagram connection, which links automatically when your accounts share an Accounts Center. It needs a Meta account.
Find the pattern in my saves. Group them by the problem they solve for me, then turn the top three formats into post ideas for my niche.
- Save reels into collections on Instagram.
- Open Muse with the same Meta account.
- Send the prompt.
- Ask for a 30-day content plan in your voice.
The video shows hook patterns, groups, three formats and a week-by-week plan.
Muse: a Facebook Marketplace listing and buyer replies
Source: a reviewer's video of October 5, with the prompt adapted from it. Muse needs a Meta account and a connected Facebook account here.
Write a Facebook Marketplace listing for this chair. List it at $30 and post it.
- Connect Facebook to Muse.
- Attach photos of the item.
- Send the prompt and approve the post when Muse asks.
- Answer any buyer question Muse passes back.
The reviewer saw a description with a suggested price, an approval prompt, and a posted listing. In an earlier listing in the same video, Muse relayed a buyer's question and replied.
Muse: a grocery list from saved recipe reels
Meta's launch post says Muse can turn a recipe reel a person saved on Instagram into a grocery list. Meta gives no prompt, so this wording was written for this article.
Turn my saved Instagram recipe reels into a grocery list for the week.
- Save recipe reels on Instagram.
- Open Muse with the same Meta account.
- Send the prompt.
- Check the list against the reels.
You should see a grocery list. No source shows a real result, so treat the first run as a test. This task needs a Meta account.
dots: a bug report that becomes a draft pull request
Source: OpenAI's launch post, which says dots deliver "complete PRs to review with attached videos". The prompt wording was written for this article.
Investigate the checkout failure described in this issue. Reproduce it, then open a draft pull request with test evidence. Ask before changing dependencies, deploying, or merging.
- Connect GitHub as a plugin.
- Set up a Codex cloud environment first, using the steps in OpenAI's dots help page.
- Send the prompt with a link to the issue.
- Review the pull request yourself.
You should see a Codex cloud thread and a draft pull request waiting for review.
dots: a calendar-aware errand
Source: Wired, which describes a demo example shared by OpenAI. A dot noticed that the calendar showed the user working through dinner and messaged two delivery options with prices. You should see a message like that, with no order placed until you pick.
Watch my calendar. When I am booked through a meal time, message me two delivery options and wait for my pick before ordering.
- Connect your calendar and email.
- Save a card on the merchant's own website, because a dot can use a card you have saved there.
- Choose the "Ask before taking action" custom rule.
- Send the prompt.
Purchases need your approval, and the approval can be given in advance only when it covers that purchase.
dots: a Slack channel check that speaks up only when a deadline is at risk
Source: OpenAI's task documentation, which gives this prompt as its example.
Check the connected planning channel each weekday at 9 AM Pacific for the next four weeks. Update the project checklist when a deadline changes. Message me in ChatGPT only if a deadline is at risk or you need a decision. Confirm the schedule.
- Connect Slack as a plugin.
- Add your dot to the planning channel. Adding it alone does not start monitoring.
- Send the prompt.
- Look for the entry in the Scheduled tab.
You should see a schedule confirmation. After that the dot stays silent until something is at risk.
How personal agents differ from Codex and Claude Code
Coding agents work on code in a project you point them at. Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools". Personal agents work on your accounts and your day. They keep a cloud computer for it.
The line has moved on the Codex side. OpenAI's docs for Codex now describe scheduled tasks that run in the background, with triggers for new Gmail messages, Slack messages and GitHub pull requests. Codex also supports plugins, MCP servers and a web search tool.
Codex scheduled tasks on the web run in the cloud. Tasks tied to a local project in the desktop app need your computer on. Triggers work on web and mobile, depend on your plan, and cannot be combined with a time schedule.
So a reader who uses Codex may already have half of what a personal agent offers.
Table: Which personal-agent jobs Codex can already do
| Job | Codex today | Personal agent |
|---|---|---|
| Daily news summary | Likely: a scheduled task plus web search; the docs show no news example | Any of the three |
| Weekday inbox check | Likely: the Gmail plugin on a schedule, or a new-message trigger | Any of the three |
| Recurring report | Can already: scheduled tasks and skills | Any of the three |
| One-time reminder | Partly: Codex can come back to a chat later; reminders are documented for dots | Muse, dots |
| Watching a social feed | Not documented: triggers are Gmail, Slack and GitHub | Grok Bot (X). Muse works with your Instagram and Facebook accounts |
| Chatting in WhatsApp | Not documented | Muse |
| Acting without a schedule | Not documented: it runs what you schedule | dots do read-only research on their own |
| Buying things | Not documented | Muse and dots, with approval |
| Its own identity | Not documented | Grok Bot has its own email; a dot can have its own Slack account |
Sources: Codex scheduled tasks, dots help, Grok Bot's X post, and Meta's Muse FAQ.
Codex already covers jobs that start from a schedule or a message in Gmail, Slack or GitHub. For jobs that start from your social feed, your phone chat or a change the agent notices itself, the Codex docs show nothing, and the personal agents do. Build to Launch has a map of AI scheduled tasks and a guide to four Codex use cases for going further.
Where each one broke in the first weeks
All three are weeks old, so each has a failure on record.
- Grok Bot: besides the October 8 routine failures above, in September Bots stopped responding. Staff said such stalls usually clear within about half an hour, and one user reported 14 hours and 25 minutes.
- Muse: ZDNET's hands-on test found that Amazon blocked Muse and that it logged the tester out every couple of hours.
- dots: Every's Dan Shipper wrote that he had "frequent issues with permissions and dropped messages" in the first days.
Start every first task read-only. Keep sends, purchases and deletes behind approval, as each product's settings allow.
Which one to try first
- Muse is out. Meta lists the US and Canada, and its help center says the app is "not yet available everywhere". A reader in China is outside the listed markets.
- dots are a poor first step. They need ChatGPT Pro from $100 a month. A reader on the $20 Plus plan has no dots. China does not appear on OpenAI's list of supported countries. Codex comes with ChatGPT plans, and OpenAI's pages show no separate Codex country list, so the Codex you already use is most likely covered by the same list. OpenAI says access outside the listed countries "may result in your account being blocked or suspended", and payment methods from outside them "will result in you being blocked from using our services".
- Grok Bot is the one to try first. It needs no X account, because a paid Cursor plan is enough, and Cursor says a trial uses a usage credit with a seven-day window and no charge to cancel.
Start with the read-only news task.
The answer changes with the accounts you hold. A reader in the US or Canada who wants a free start should try Muse. A reader already on ChatGPT Pro, in a supported country, can try dots first, because the connectors and memory they have set up come with it. Dan Shipper of Every says people already using Grok Bots or Muse should probably not switch to dots "unless you're also a big ChatGPT or Codex user", and he suggests waiting "a week or two".
Codex stays where it is: keep it for code, and use the personal agent for the jobs that start outside your editor.
FAQ
Do I need an X account to use Grok Bot?
No. Grok Bot needs a paid Cursor plan or a linked SuperGrok or X Premium+ plan. You need X only for the X connector, or for an X Premium+ link.
Is Muse available in China?
The Meta pages VibeCoding.Builders read never mention China. They list "the US and Canada" and say the app is not yet available everywhere. So China is not an officially listed market.
What do these agents cost?
As of 2026-10-12, Grok Bot comes with Cursor Pro at $20 a month or SuperGrok at $30. Muse is free with a usage limit, with paid plans at $20 and $100 a month. Dots need ChatGPT Pro at $100, $200 or $500 a month.
Can Codex already run a daily news summary?
Likely yes. Codex scheduled tasks run in the background on a minute, daily or weekly schedule, and Codex has a web search tool. OpenAI's examples are code-focused, so the news summary is a use of the parts rather than a documented recipe.
How far should I trust a personal agent with my accounts?
Trust it only as far as its approval settings.
Meta says Muse always asks for approval before a purchase, and OpenAI says dots need approval for purchases. A model checks Grok Bot's approvals, and one reviewer calls them looser than those of other agents.
All Bots in Grok Bot share one cloud computer and every login on it. xAI says: "Do not use separate Bots as a security boundary." TIME reported that Muse writes hourly "dossiers", meaning notes about users and about people they mention.
OpenAI says its protections against prompt injection "do not eliminate it", and PortEden quotes Cursor saying similar of its own controls. Start read-only and widen access slowly.
Sources
- Grok Bot launch, SpaceXAI
- More plans for Grok Bot, SpaceXAI
- Grok Bot and X, SpaceXAI, 2026-08-29
- Grok Bot pricing, x.ai
- Grok Bot 101 guide, SpaceXAI
- Grok Bot for work guide, SpaceXAI
- Grok Bot overview, xAI docs
- Skills, routines and automations, xAI docs
- Approvals, security and privacy, xAI docs
- Security, xAI docs
- Grok Bot plans, Cursor help
- @bot on X
- All Grok Bot routines failed to run, Cursor forum
- Grok Bots not responding, Cursor forum
- Automations in Grok, xAI
- Introducing Muse, Meta, 2026-09-08
- Muse for small business, Meta, 2026-09-29
- Muse, Meta AI
- Muse Terms, Meta
- Muse help: availability
- Muse help: accounts
- Muse help: connectors
- Muse help: reminders
- Muse help: skills
- Muse help: browser
- Muse help: purchases
- Muse help: permissions
- Muse help: subscriptions
- Introducing dots, OpenAI, 2026-09-29
- Getting started with your dot, OpenAI help
- Dots privacy, security and safety FAQs, OpenAI help
- Meet dots, OpenAI docs
- Connect computers and apps to your dot, OpenAI docs
- Dots tasks and memory, OpenAI docs
- About ChatGPT Pro tiers, OpenAI help
- ChatGPT supported countries, OpenAI help
- ChatGPT and API services in unsupported countries, OpenAI help
- Codex scheduled tasks, OpenAI docs
- Codex pricing, OpenAI docs
- ChatGPT pricing, OpenAI
- Dots controls, OpenAI docs
- Codex plugins, OpenAI docs
- Codex MCP, OpenAI docs
- Codex web search, OpenAI docs
- Claude Code overview, Anthropic docs
- Wired on dots
- TIME on Muse privacy, 2026-10-06
- ZDNET hands-on test of Muse, 2026-09-28
- Muse vs Grok Bot vs dots, OneWave, 2026-09-30
- ChatGPT dots vs Grok Bot vs Meta Muse vs Hermes, Wright Mode, 2026-10-03
- Meta Muse vs Grok Bot, CatDoes
- What each agent can see, PortEden, 2026-09-29
- I tested OpenAI dots on 10 business tasks, AI Agents Library
- How to use Meta Muse, AI Agents Library
- Dan Shipper on dots, X, 2026-09-29
- Instagram saves with Muse, Joon Ahn, YouTube, 2026-10-01
- Muse Facebook Marketplace test, Leslie Samuel, YouTube, 2026-10-05
- AI scheduled tasks: the complete map, Build to Launch
- ChatGPT Codex guide: 4 use cases, Build to Launch