Primary Bot is the new centre of SpaceXAI’s Grok Bot. Announced on 1 October 2026, Primary Bot is a main assistant that watches for work it can take off your plate, offers to do it before you ask, and passes tasks to your other Bots. On the same evening, Grok 4.7 started appearing as the model behind every mode in the Grok web and mobile apps: Fast, Expert, Build and Heavy.
The two changes arrived within an hour of each other. Together they mark a shift from an agent you instruct to one that volunteers. xAI, which now trades as SpaceXAI after Elon Musk folded it into SpaceX in February, is doing this in the same week OpenAI launched its Dots agents and Meta pushed Muse.
This article explains what Primary Bot does, what “suggestions don’t count against your usage” really covers, what Grok 4.7 brings to the Grok apps, what SpaceXAI has not yet documented, and what a proactive agent means for businesses weighing up AI teammates.
Table of contents
- What Primary Bot Is
- How the Proactive Primary Bot Works
- What “Suggestions Don’t Count Against Usage” Means
- Grok 4.7 Becomes the Base Model in the Grok Apps
- How Much Better Is Grok 4.7?
- What SpaceXAI Has Not Said Yet
- Proactive Agents and Control
- Primary Bot in a Crowded Agent Race
- Questions to Ask Before Switching On Primary Bot
- What Primary Bot Means for Businesses
- Primary Bot FAQ
- References and Further Reading
What Primary Bot Is
Grok Bot gives each user a set of named AI teammates, called Bots, that work on a persistent cloud computer. Primary Bot is the one that sits above the rest.
The announcement
The official @bot account posted at 17:35 UTC on 1 October: “Grok Bot can now suggest ways to help without you needing to ask.” The post, with a ten-second video, passed two million views within a day. A reply in the same thread added the detail: “Your primary Bot will spot work it can take off your plate and offer to handle it. It’s rolling out over the next few hours, and suggestions don’t count against your usage.”
How the app introduces it
TestingCatalog found Primary Bot in the iOS app about six hours earlier and published screenshots. The onboarding screen reads: “Introducing Primary Bot. Primary Bot does work proactively. It is your go-to for everyday tasks, unblocks your work, and checks in when it needs your help.” Two buttons follow: “Use an existing Bot” and “Create Primary Bot”.
What it says when it starts
In the screenshots, a new Primary Bot opens with: “I’m here to be your primary bot. I can manage your other bots. Task me with anything. I’ll make sure it gets to the right bot.” It then adds: “I’m taking a look around to see if there’s anything I can pick up for you. I’ll follow up in a moment.”
How the Proactive Primary Bot Works
The design is close to a chief of staff: one assistant that knows what all the others are doing and decides who handles what.
Choosing your Primary Bot
According to TestingCatalog, users are asked either to set up a new Primary Bot or to promote a Bot they already have. Someone who has spent weeks training a research or admin Bot can keep that context rather than starting again.
Spotting work
Primary Bot looks at what it can see, which in Grok Bot means the apps, inboxes and files you have already connected, and offers to act. SpaceXAI has not published a list of the signals it watches, so the clearest evidence is a real example.
The Uber example
Matt Palmer, who works on Cursor and Grok Bot at SpaceXAI, posted one about 20 minutes after the announcement. He had switched flights, but his Uber reservation was still tied to the old one. Grok Bot caught it without being asked: “Your Uber reserve home to … is still tied to the old UA 2837, which lands at SFO at 8:28pm PT. You’re now on UA 233, landing at 7:15pm PT.” It pinged him during his connecting flight, and he fixed it in the air. Musk quoted the post: “Grok @Bot is an amazing assistant.”
Routing to other Bots
Because Primary Bot can “manage your other bots”, a request does not need to go to the right specialist. You tell Primary Bot, and it hands the job on. Grok Bot already lets Bots message each other and share group chats, so Primary Bot builds on plumbing that exists.
| Time (UTC, 1 Oct) | Event | Source |
|---|---|---|
| 11:23 | Primary Bot screens spotted in the iOS app | TestingCatalog |
| 17:35 | “Grok Bot can now suggest ways to help” | @bot |
| 17:57 | Uber reservation example | Matt Palmer, SpaceXAI |
| 18:29 | “Grok 4.7 is now in app” | LuminaBench |
| 18:36 | Grok 4.7 confirmed across Fast, Expert, Build and Heavy | TestingCatalog |
| 19:39 | Grok 4.7 on Google’s Gemini Enterprise Agent Platform | @SpaceXAI |
What "Suggestions Don't Count Against Usage" Means
The usage line is the most commercially important sentence in the launch, and it is narrower than it sounds.
Grok Bot has its own allowance
Since 26 August, Grok Bot has been included with SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Pro+, Ultra and the Cursor Teams plans. SpaceXAI says it “comes with its own usage, separate from your Grok and Cursor plans.” The documentation adds that the allowance resets weekly and that eligible accounts can buy extra usage billed by model and token cost.
Free to suggest, not free to do
SpaceXAI says suggestions are free. It does not say the work is. When you accept an offer and Primary Bot books, drafts, files or researches something, that task almost certainly draws on the same weekly allowance as any task you give it yourself. The announcement does not spell this out.
Why the distinction matters
A proactive agent that charged for every idea would punish users for its own curiosity. Making suggestions free removes that worry. But a busy Primary Bot that offers ten jobs a day, each one accepted, could use up a weekly allowance faster than a user who only asks for what they need.
| Step | Counts against usage? | Basis |
|---|---|---|
| Primary Bot makes a suggestion | No | @bot, 1 October |
| You accept and the Bot does the task | Not stated; likely yes | Our reading of the weekly usage model |
| Tasks you assign directly | Yes, from the weekly allowance | Grok Bot documentation |
| Usage beyond the allowance | On-demand, billed by model and token cost | Grok Bot documentation |
Grok 4.7 Becomes the Base Model in the Grok Apps
The second half of the news is quieter but affects far more people: every mode in the consumer Grok apps now runs on Grok 4.7.
Ten days from API to app
SpaceXAI released Grok 4.7 on 21 September in Cursor, Grok Build and the xAI API. The consumer apps followed on 1 October, ten days later. The account LuminaBench put it bluntly: “Grok 4.7 is now in app. This took way too long.” We tracked the model’s long road to launch in our report on the Grok 4.7 quota-page sighting.
Every mode, one model
TestingCatalog’s screenshots of the web app show the mode menu as Auto, Fast, Expert, Build and Heavy. Hovering over Expert shows “Thinks hard · Grok 4.7”. Hovering over Heavy shows “Team of Experts · Grok 4.7”. TestingCatalog reports the same model behind Fast and Build, which makes Grok 4.7 the base model across the app rather than a premium option.
What changed in the model
SpaceXAI’s launch post says Grok 4.7 “uses a new, larger base model compared to Grok 4.6,” trained with “a longer reinforcement learning run on a harder mix of tasks, weighted toward problems that take many hours to complete.” It is “better at verifying its own work and managing longer context.” Crucially for this story, SpaceXAI says it also “trained Grok 4.7 to natively understand the Grok Bot harness, making it better at conversational tasks and general knowledge work.”
| Grok app mode | Tooltip in the screenshots | Model |
|---|---|---|
| Auto | Not captured | Not stated |
| Fast | Not captured | Grok 4.7, per TestingCatalog |
| Expert | “Thinks hard” | Grok 4.7, shown in the tooltip |
| Build | Not captured | Grok 4.7, per TestingCatalog |
| Heavy | “Team of Experts” | Grok 4.7, shown in the tooltip |
How Much Better Is Grok 4.7?
SpaceXAI’s own benchmark table compares Grok 4.7 with Grok 4.6 at the same price. The gains are real but uneven, and they help explain why the model is worth putting behind Primary Bot.
The headline numbers
On SpaceXAI’s figures, Grok 4.7 scores 46.3% on CursorBench 4.0 against 40.4% for Grok 4.6, and 37.6% on Terminal-Bench 4.0 against 20.3%. On the EEBench electrical engineering test it scores 64.0% against 53.0%. Prices are unchanged at $2 per million input tokens and $6 per million output tokens below 200,000 prompt tokens, rising to $4 and $12 above.
Read the footnotes
The Grok 4.7 column is labelled “xHigh” reasoning effort and the Grok 4.6 column “High”, so part of the gap may come from more thinking time. One Grok 4.7 score, 71.0% on DeepSWE v1.1, is marked as a high-effort result. Benchmarks published by a vendor should be checked against independent tests before they drive a buying decision.
Still behind the frontier on some tasks
The same table shows Anthropic’s Fable 5.1 ahead on CursorBench (51.8%) and well ahead on Terminal-Bench (57.9%), at five times the input price ($10 against $2). On 14 September Musk had said Grok 4.7 “should be roughly on par with Opus 5.0, not 5.1.” The pitch is value, not outright leadership. For how the model looked when it reached listings, see our note on Grok 4.7 on AIxploria.
Grok 4.7 gain over Grok 4.6, percentage points (bar length relative to 17.3)
Each gain is the Grok 4.7 score minus the Grok 4.6 score from SpaceXAI’s 21 September table, for example 37.6 – 20.3 = 17.3. Bars are scaled to the largest gain, so 11.0 ÷ 17.3 = 63.6%, 8.2 ÷ 17.3 = 47.4%, 5.9 ÷ 17.3 = 34.1% and 3.8 ÷ 17.3 = 22.0%. The biggest improvement is on multi-hour terminal work, the kind of long task a proactive agent takes on.
What SpaceXAI Has Not Said Yet
The launch came through social posts and an app update rather than a blog post or documentation, so several practical questions are open.
No documentation yet
On 2 October we checked nine pages of the Grok Bot documentation, including the overview, Bots, settings, approvals and FAQ pages. None mentions Primary Bot. SpaceXAI’s newest news post is still the 28 September Team Bots announcement, and the developer release notes do not mention the consumer app change to Grok 4.7.
Which model Grok Bot itself uses
The Grok Bot settings page says “Cursor manages model selection, so there is no model picker.” SpaceXAI says Grok 4.7 was trained for the Grok Bot harness, and Musk praised “major speed improvements” to Grok Bot on 1 October. But no published source states which model each Bot, including Primary Bot, runs on today.
Controls and limits
There is no published detail yet on how often Primary Bot will interrupt, whether users can limit what it watches, or how to switch the proactive behaviour off. Grok Bot’s existing approval controls still apply, and they are the safety net for now.
Proactive Agents and Control
A teammate that acts before it is asked is useful only if it stops before it does the wrong thing. Grok Bot’s existing guardrails matter more with Primary Bot in charge.
Approvals and Auto Review
Grok Bot documentation says sensitive or consequential actions can stop for approval, and recommends narrow “Ask first” rules for “sending, publishing, deleting, purchasing, or changing production systems.” Admins can enforce Auto Review rules for a whole team. With a proactive Primary Bot, those rules decide which suggestions become actions without a human click.
One shared computer
All of a user’s Bots share one cloud computer, with its files, browser sessions and app logins. That makes handoffs easy, and it means anything one Bot can reach, Primary Bot can reach too. Treat that computer as a single trust zone, and connect only the accounts you would hand to an assistant. Our IT security team helps organisations set that boundary.
Read-only where possible
Some recent integrations are deliberately limited. The Finance connector added on 26 September links bank, card and investment accounts through Plaid, and SpaceXAI says access is read-only. That model, where a Bot can see and suggest but not move money, is a sensible default for anything proactive.
Primary Bot in a Crowded Agent Race
Primary Bot arrives in the busiest week yet for always-on assistants.
OpenAI and Meta
OpenAI launched Dots on 30 September, a paid agent product running on GPT-6 Astra. Meta’s Muse offers a free tier. We compared the two in our report on Dots vs Muse. Grok Bot’s answer is to bundle the agent with plans people already pay for, and now to make it proactive.
Proactivity is the new battleground
OpenAI tried proactive help with ChatGPT Pulse, which delivered daily briefings to Pro users from September 2025. Primary Bot goes further by offering to do the work, not just report on it. TestingCatalog framed it as SpaceXAI’s “turn to hop into a private assistant area and compete with Muse.”
A fast release rhythm
Grok Bot has shipped a major update most weeks since its beta launch on 11 August, from voice to Team Bots to coding handoffs. The voice mode launch in September was one step; Primary Bot is the biggest change to how users interact with it since launch.
| Date (2026) | Grok Bot update |
|---|---|
| 11 August | Grok Bot launches in beta |
| 26 August | Included with SuperGrok and Cursor plans, with its own usage |
| 17-18 September | Voice conversations, then voice notes |
| 21 September | Grok 4.7 released, trained for the Grok Bot harness |
| 26 September | Read-only Finance connector via Plaid |
| 28 September | Team Bots in public beta for Teams and Enterprise |
| 30 September | Bots hand coding tasks to Cursor |
| 1 October | Primary Bot and proactive suggestions; Grok 4.7 across the Grok apps |
Questions to Ask Before Switching On Primary Bot
Because Primary Bot is new and largely undocumented, a short checklist helps before letting it loose on work accounts.
What can it see?
List the connectors, inboxes and browser sessions on your Grok Bot computer. Primary Bot can only suggest work from what it can reach, so that list is also the boundary of what it might act on. Remove anything you would not hand to a new assistant on their first day.
What stops for approval?
Check your Auto Review and “Ask first” rules before the first suggestion arrives. Sending, publishing, deleting and purchasing should all require a click. If your organisation uses Team Bots, ask your admin which team rules are enforced and whether they also cover a personal Primary Bot.
Who is accountable?
A suggestion accepted with one tap is still a decision. Decide who owns the outcome when Primary Bot books the wrong car or replies to the wrong client. In a business, that should be the person who accepted it, and they should be able to see exactly what the Bot did.
How will you measure it?
Track suggestions made, accepted and reversed for the first few weeks. A Primary Bot that offers useful work most of the time earns more scope. One that needs frequent correction should watch fewer sources.
What Primary Bot Means for Businesses
For organisations testing AI agents, Primary Bot is a preview of where every assistant is heading: from waiting for instructions to managing a queue of work.
Start with low-risk signals
Let a Primary Bot watch calendars, travel bookings and shared documents before it gets near customer email or payments. The Uber example shows the value of catching small mismatches. The same habit applied to invoices or contracts needs tighter approval rules first.
Write the boundaries down
Put standing limits in each Bot’s description and use “Ask first” rules for any outward-facing action. A proactive agent without written boundaries will eventually do something reasonable that nobody wanted. Our guide to AI employees and autonomous AI agents covers how to scope that kind of digital worker.
Watch the weekly allowance
Because accepted suggestions likely consume usage, track how much of the weekly allowance goes on Primary Bot’s own ideas in the first fortnight. If it is high, tighten what the Bot watches rather than turning the feature off. Our AI strategy team helps businesses set those policies across tools.
Primary Bot FAQ
What is Primary Bot?
Primary Bot is the main assistant in SpaceXAI’s Grok Bot. It suggests work it can do for you without being asked, and it routes your requests to your other Bots.
When did Primary Bot launch?
SpaceXAI announced Primary Bot on 1 October 2026 and said it would roll out over the following hours.
Do Primary Bot suggestions cost anything?
SpaceXAI says suggestions do not count against Grok Bot usage. Carrying out a task you accept is not covered by that statement.
Can I make an existing Bot my Primary Bot?
Yes. The iOS app offers “Use an existing Bot” alongside “Create Primary Bot”.
Is Grok 4.7 now in the Grok app?
Yes. Since 1 October, Grok 4.7 is the base model across the Fast, Expert, Build and Heavy modes in the Grok web and mobile apps, according to TestingCatalog’s screenshots.
Which plans include Grok Bot?
SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Pro+, Ultra and Cursor Teams plans all include Grok Bot, with its own weekly usage.
References and Further Reading
“Grok Bot can now suggest ways to help without you needing to ask” (@bot on X)
“Your primary Bot will spot work it can take off your plate” (@bot on X)
Grok Bot is getting a Primary Bot, with iOS screenshots (TestingCatalog on X)
Grok 4.7 rolling out as the base model across all modes (TestingCatalog on X)
Grok Bot is now more proactive: the Uber example (Matt Palmer on X)
Introducing Grok 4.7 (SpaceXAI)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
