Compare·Managed always-on cloud teammates (SpaceXAI)
Continuum
Multi-agent workbench
VS
Grok Bot
Managed always-on cloud teammates (SpaceXAI)

Continuum vs Grok Bot

One product rents you a shared cloud computer and messages you back. The other runs agents on machines you already own and shows you the branch, the diff, and the bill.

Grok Bot is SpaceXAI's always-on agent product, launched in beta on 11 August 2026. You install a native app on macOS, Windows, or iPhone and then message Bots the way you message a colleague. Named Bots keep memory, files, browser sessions, and preferences between turns, learn a task by watching you do it once, save it as a skill or a routine, and re-run that routine on a schedule. The headline is that a Bot has its own computer. The documentation says something narrower and it matters: all of your Bots use the same persistent cloud Linux computer, each with its own screen, and xAI's own overview tells you to treat a login or a file placed on that computer as available to all of your Bots, "without getting separate security boundaries". Access runs through a Cursor account on every route, and the Bots run on Cursor's infrastructure rather than xAI's: SpaceX announced its $60B all-stock acquisition of Anysphere, Cursor's maker, in June 2026 and closed it on 14 and 15 August, and xAI itself was renamed SpaceXAI in July 2026. Linking the two accounts is permanent, which Cursor's own documentation states and which no article covering the launch mentions. Continuum is a different shape: a free workbench that runs the official coding agents you already pay for, Claude Code over PTY plus Codex, Cursor, and Grok through the ACP harness, alongside Gemini and OpenCode, on hardware you already own, with a git worktree per session, plan and diff gates, live 5h and weekly quota gauges per account, and a local dollar ledger by repo. The teammate metaphor is genuinely good. This page is about what it costs. Background: what Grok Bot is.

Download for Mac All comparisons
Workflow Monday path Matrix Decide Money FAQ

Updated 2026-08-03 · Mac stable · Win/Linux desktop beta

Choose Continuum when

Pick Continuum when the output is code somebody has to review: a worktree per session, per-hunk diff staging, a PR pane, live quota gauges across multiple accounts, spend by repo, and phone or web control, on an app that costs nothing. Pick it too when one lab is not enough, because Grok Bot is xAI only and will not tell you which model ran your task. Grok Bot alternatives places it beside the rest of the field.

Choose Grok Bot when

Pick Grok Bot when the appeal is that nothing runs on your machine: no agent host or server to install or maintain, only the native client, a Bot that keeps working after your laptop shuts, computer use for apps and sites with no clean API, and a messaging surface a non-developer can use on day one. Those are real advantages and this page does not pretend otherwise.

Snapshot direct alternative
Dimension Continuum Grok Bot
Where agents run Your Mac, enrolled Linux or Windows hosts One shared cloud Linux VM per account
Isolation unit Git worktree + branch per session One screen per Bot on the same computer
Labs available Claude · Codex · Cursor · Grok · Gemini · OpenCode… xAI only, model undisclosed, no picker
Works with the laptop shut On an enrolled host or Cloud runner Yes, that is the point of the product
Review surface Plan gate · per-hunk diff · PR pane A chat transcript per Bot
Automation Host daemon HTTP/WS API + CLI client No public API or webhooks
Account lock-in Sign out or delete the app Grok to Cursor link is permanent, by Cursor's docs
Price Free app · hosted inference optional $40 to $300 plan, plus uncapped metered overage
Mac stable · web · iPhone · Watch · Win/Linux desktop beta · free app
01

Message a teammate vs supervise a branch

In Continuum

A day whose output is a pull request

09:10

Two worktree sessions open on the product repo, Claude on the bug and Codex on the tests. The gauges already show how hot the weekly window is.

11:40

A plan lands during standup. Approve it from the iPhone over the paired host. The Mac keeps the session and the write permission starts only after the approval.

15:20

The refactor goes sideways. Restore the pre-prompt checkpoint, re-prompt with a tighter spec, and the worktree is clean without touching main.

17:30

Stage three hunks, drop the clever rewrite nobody asked for, open the PR from the same pane. The ledger closes the day in dollars by repo.

In Grok Bot

A day on the shared cloud computer

09:10

You message a Bot from the phone on the train. It picks up the browser session it already holds and starts the errand with nothing running on your laptop.

11:40

A second Bot joins the thread and takes ownership of a subtask. Both are using the same computer, the same files, and the same logins, on separate screens.

15:20

A site asks for a password and a 2FA code. The Bot pauses and hands you the computer to type it. The session then persists for every other Bot on the account.

21:00

The Bots keep working with your laptop shut, which is exactly the appeal. What the week cost is harder to answer: the weekly allowance is not published and the overage has no cap.

02

Monday path

Monday with an always-on agent in the evaluation

Install, connect, first session - steps you can run the same day.

01

Price the access path before anything else. Grok Bot needs a Cursor account on every route, so work out whether you are buying one plan or a second one: SuperGrok Plus or Heavy, or Cursor Pro+, Ultra, or Teams. The cost calculator does that arithmetic.

02

Install Continuum for Mac and pair the iPhone with the QR flow over Tailscale, so plan approvals can leave the desk without a rented VM in the path.

03

Spawn a worktree session per ticket, Claude on one and Codex on the next. Each gets its own branch, its own diff, and its own history.

04

Keep plan mode on for anything near main. Read the plan, approve it from the Mac or the phone, and let write permission start only then.

05

If you are trialing Grok Bot alongside, give it its own logins. Anything you unlock on that shared computer is available to every Bot on the account.

06

Do not rehearse on Grok Bot. There is no dry-run mode, so a test routine performs the real work, and an approval afterwards cannot undo an email that already left.

07

Back in Continuum, stage the diff by hunk, open the PR from the same pane, and watch checks without leaving the session. See reviewing AI-generated code.

08

Friday: read the 5h and weekly gauges and the per-repo ledger. Grok Bot's weekly allowance is not published and its overage has no spend cap, so that column stays an estimate. Cost by repo explains the ledger.

How Continuum runs agents on your hardware

How Continuum runs agents without renting a computer

  • Sessions run on hardware you already own: the Mac itself, an enrolled Linux or Windows host over Tailscale, or a Continuum Cloud runner if your account carries one.
  • Claude Code spawns over PTY; Codex, Cursor, and Grok run through the ACP harness. Official engines under your own logins, with no Continuum-specific model account in between.
  • Each Code session binds one conversation to one real git worktree and branch, with plan, diff, PR, terminal, files, and artifacts as panes on the same run.
  • A git checkpoint is taken before every prompt, so a turn that goes wrong is a restore instead of an argument with the agent.
  • Multi-account isolation uses a per-instance CLAUDE_CONFIG_DIR and CODEX_HOME, so a second subscription can never bill the first.
  • The daemon accepts loopback and Tailscale peers only and every request carries a bearer token. No shared cloud computer, no public messaging network in the path.
  • Local analytics price session history with a LiteLLM snapshot into dollars by repo, provider, and day, beside live 5h and weekly quota gauges.
03

Always-on agent matrix

Both products promise agents that keep working while you are somewhere else. They disagree about who owns the computer, whether one lab is enough, and how much of what happened you are allowed to see afterwards.

Capability Continuum Grok Bot
Runs on hardware you ownNothing to rent, nothing to migrate off Mac, enrolled Linux/Windows, or a Cloud runner One xAI-hosted cloud computer, no self-host route
Keeps working while your device is offGrok Bot's home ground Yes on an enrolled host or Cloud runner, no on a sleeping Mac Always-on by design
A real boundary between agentsFiles, logins, and browser sessions are shared on Grok Bot Separate git worktree and branch per session Docs: shared computer, not a security boundary
More than one lab Claude · Codex · Cursor · Grok · Gemini · OpenCode xAI only
Choose the modelGrok 4.6 is available in Grok Bot, but routing is automatic Model and effort per session, swap mid-thread No picker; the default model is never disclosed
Per-hunk diff review and PR merge Stage, revert, commit, open PR, watch checks Chat transcript; no repo review layer
Enforced gate before writes Read-only Plan mode, explicit approve and run Prose approval rules and Auto Review, not enforced policy
Rehearse an action without doing itRehearse a send on Grok Bot and you sent it Plan mode is read-only; checkpoints make a turn recoverable Docs: a test run performs real work
Live quota gauges and cost by repo 5h + weekly rails · $ by repo/provider/day Weekly allowance undisclosed, no Grok Bot spend cap
Public API or webhooks Host daemon HTTP/WS API + CLI client None published
Audit trail Local hashed audit JSONL with a Diagnostics viewer Audit view listed as coming; 20-record limit reported by Daily Dose of Data Science and unconfirmed by xAI
Native phone controliPad is not supported at launch on Grok Bot; Android is listed as coming iPhone and Watch on the same live session iPhone app, iOS 18 and later
Computer use on apps with no API Deliberately out of scope Browser and desktop control on the shared computer
Reversible account linkCursor's docs: you cannot unlink it or move it to another Cursor account Sign out, remove an account, or delete the app A SuperGrok to Cursor link is permanent once created
Reported ceilingsAlso 20 retained run records, a 10-minute teach cap, and 2 to 6 Bots per group chat; reported by Daily Dose of Data Science and unconfirmed by xAI No product cap on sessions, worktrees, or hosts 50 Bots and group chats per account, 50 routines per Bot; reported by Daily Dose of Data Science and unconfirmed by xAI
Price of the software itself Free app; bring the plans you already hold Requires a qualifying SuperGrok or Cursor plan
04

Where the products actually diverge

01 · Shape

A rented computer versus hardware you already own.

Grok Bot's whole product is a managed Linux computer in xAI's cloud with a browser, a filesystem, and a terminal on it. Continuum has no computer to sell: it drives official agent CLIs on your Mac, on a Linux or Windows box you enrolled over Tailscale, or on a Continuum Cloud runner. Neither answer is universally right. The question is whether you want the machine, the credentials, and the transcript to sit somewhere you control.

02 · Reach

Where Grok Bot wins outright.

Zero setup, nothing to keep awake, and work that continues after your device is off. Add computer use, which drives apps and sites that have no clean API, learning a task by demonstration, and routines on a schedule, all reachable from a messaging UI that a non-developer can use immediately. Continuum has no equivalent to any of that and is not building one. If those are the job, this comparison ends here.

03 · Boundaries

One computer, many screens, no boundary between Bots.

The marketing says Bots have their own computer. The docs say all of your Bots use the same persistent cloud computer, share files, browser sessions, and app logins, and get their own screen "without getting separate security boundaries", and that you should treat a login placed there as available to all of them. Several launch-week write-ups repeat the marketing line as fact, including one comparison table that ranks well for it. Continuum's boundary is a git worktree per session and per-account config isolation, which is a smaller promise honestly kept. More in agent security.

04 · Model

You cannot see which model ran your task.

Grok 4.6 shipped on 12 August 2026, one day after Grok Bot, and it is live across the xAI API, Grok Build, Cursor, and Grok Bot. What xAI never states is which model Grok Bot runs by default, and there is no picker, so routing is automatic and invisible. That is not only a transparency question: without a picker you cannot send a cheap task to a cheap model, which is the usual way people control an uncapped meter. Continuum pins model and effort per session and lets you swap mid-thread.

05 · Money

A plan floor you can read and a meter you cannot.

The seat price is knowable. The consumption is not: the weekly token allowance is undisclosed, there is no Grok Bot spend cap, overage bills at raw model rates, and the launch-week usage display was visibly wrong. Continuum charges nothing for a session you run under your own login and prices your actual history into dollars by repo, provider, and day. Model both sides before committing a seat.

06 · Governance

No dry run, no queryable organisation-wide audit view yet, and compliance that points at Cursor.

No published SOC 2, ISO 27001, GDPR, or HIPAA claim was found in the Grok Bot doc set, nor a stated retention period or data-residency option; compliance defers to Cursor's terms, which is now literally true of the hardware as well, since Bots run on Cursor's infrastructure. The audit view is listed as coming. The twenty-record routine retention limit was reported by Daily Dose of Data Science and remains unconfirmed by xAI. There is no dry-run mode, so a test run does the real thing. Deleting a Bot does not remove the shared files or sessions it left behind, and Cursor's Legacy Privacy Mode blocks Grok Bot outright. A buyer with a security review should read the full objection set first.

07 · Lock-in

The link you cannot undo.

This is the most under-reported fact about buying Grok Bot, and it comes from Cursor's own help pages rather than a critic: "A SuperGrok link is permanent once created. You can't unlink a Grok account from a Cursor account, and you can't move a link to a different Cursor account." Every launch write-up we read omits it. Pair that with an account-wide shared computer, credentials you unlocked by hand, no export path, and no API, and the exit cost is real even though the monthly price looks like an ordinary subscription. Continuum's exit is signing out or deleting the app: the repos, the transcripts, and the analytics were already on your machine.

05

Coexistence

Stack recipe

How people run both.

Stack recipe: let Grok Bot do browser errands and app operation on its cloud computer, and keep repository work in Continuum where the branch, the diff, and the pull request live. If you run both, hold one line: never unlock a credential on the Grok Bot computer that can merge to main. A session you authenticate there persists and is inherited by every other Bot on the account, and by xAI's own wording those Bots are not separate security boundaries. The security and compliance page works through the rest.

Download Continuum Providers
06

A rented computer or your own

Continuum

Choose Continuum when the work ends in a pull request

  • The work ends in a pull request that somebody has to own and review.
  • You want a git worktree per agent so parallel runs cannot collide.
  • You want more than one lab, and the ability to name the model that ran the task.
  • You need live quota gauges and dollars by repo rather than an undisclosed weekly allowance.
  • The repository and credentials should remain on your host; selected context is sent to the configured provider for inference.
  • You are not willing to make a permanent, unbreakable link between two vendor accounts.
  • You want the software to cost nothing on top of the subscriptions you already pay for.
Grok Bot

Choose Grok Bot when the machine must not be yours

  • Nothing may run on your own machine, and work must continue while your device is off.
  • The tasks are app and browser errands rather than repository changes.
  • You want an agent that learns a task by watching you do it once, then runs it on a schedule.
  • A messaging interface matters more than a diff, because the users are not all developers.
  • You already hold a qualifying SuperGrok or Cursor plan, so the marginal cost looks small.
  • You are comfortable with one lab, an undisclosed model, and a meter without a cap.
07

What you pay

Continuum

App + your labs

$0 for the app on Mac, web, iPhone, and Watch. Bring your own keys or the Claude and ChatGPT plans you already pay for, and Continuum takes no cut. Optional hosted inference: Plus $25/mo, Max 100 $100/mo, Max 200 $200/mo, Ultra $500/mo.

Grok Bot

Their bill

Grok Bot has no standalone price and needs a Cursor account on every route. As of 22 August 2026 the eligible plans are SuperGrok Plus at $100/mo, SuperGrok Heavy at $300/mo, Cursor Pro+ at $60/mo, Cursor Ultra at $200/mo, and Cursor Teams at $40 or $120 per seat per month for Standard and Premium. The base $30 SuperGrok plan does not include Grok Bot, although it does include Grok Build, which is a different product and a common and expensive mix-up. At launch on 11 August only Heavy, Ultra, and Teams Premium qualified; the expansion landed on 21 August along with a free trial with limited usage for everyone else. xAI states no trial duration and the third-party numbers disagree, so this page names none. Enterprise is waitlist only. Note that xAI publishes no Grok Bot price of its own and its pricing page does not serve to automated checks, so the SuperGrok figures here are reconciled across xAI's and Cursor's documentation and independent pricing trackers rather than quoted from one authoritative page.

How to compare

Total cost of work

The seat price is the part you can read. The meter is not: the weekly token allowance is undisclosed, there is no Grok Bot spend cap, overage bills at raw model rates, and with no model picker you cannot route cheap work to a cheap model. Then there is the part that is not a price at all: linking SuperGrok to Cursor is permanent, so the cheapest on-ramp is also an irreversible decision. Continuum meters nothing on a session you run under your own login and prices what you actually ran. Different bills, not different prices for the same thing.

Grok Bot plans checked 3 August 2026 against their pricing page. Vendors reprice often - if you spot a stale figure, tell us and we will fix it. Continuum never marks up a session it runs under your own login.

Continuum ladder: Free app · Plus $25/mo · Max 100 · Max 200 · Ultra - full pricing. Also analytics, multi-account, devices.

08

Questions

Continuum vs Grok Bot.

Deep dives: docs, providers, sessions.

Grok Bot is SpaceXAI's always-on agent product, launched in beta on 11 August 2026. Native apps on macOS, Windows, and iPhone let you message Bots like colleagues. Bots keep memory, files, browser sessions, and preferences between turns, use connectors and MCP where they exist and computer use where they do not, learn a task by demonstration, save it as a skill or routine, and run it on a schedule. They work on one persistent cloud Linux computer that every Bot on your account shares.

There is no standalone price. It rides a qualifying plan and requires a Cursor account either way: SuperGrok Plus at $100/mo, SuperGrok Heavy at $300/mo, Cursor Pro+ at $60/mo, Cursor Ultra at $200/mo, or Cursor Teams at $40 and $120 per seat for Standard and Premium. The base $30 SuperGrok plan does not include it. On top of the plan sits a weekly token allowance of undisclosed size and uncapped overage at raw model rates, so the seat price is a floor rather than a bill.

No, and this is the least reported fact about the purchase. Cursor's own documentation states that a SuperGrok link is permanent once created, that you cannot unlink a Grok account from a Cursor account, and that you cannot move a link to a different Cursor account. Decide which Cursor account should own the link before you create it, because the decision is not revisable, and none of the launch coverage warns buyers about it.

The ceilings we found were reported by Daily Dose of Data Science and remain unconfirmed by xAI: 50 Bots and group chats per account, 50 routines per Bot, 20 most-recent run records retained, a 10-minute cap on teaching a task by demonstration, and 2 to 6 Bots per group chat. The token side is not published at all: the weekly allowance has no stated size and there is no Grok Bot spend cap.

No, and this is the most misreported fact about the product. xAI's own overview says all of your Bots use the same persistent cloud computer, that they share files, browser sessions, and app logins, and that each Bot gets its own screen on that computer "without getting separate security boundaries". It also tells you to treat a login or file placed there as available to all of your Bots. Several ranking articles state the opposite.

xAI does not say. Grok 4.6 released on 12 August 2026 and is live across the xAI API, Grok Build, Cursor, and Grok Bot, so it is the plausible default, but no xAI page names Grok Bot's default model and the product offers no picker. Routing is automatic, which means you cannot know or control which model handled a given task, or what that choice cost you.

No public API and no webhooks as of 22 August 2026. Bots are reachable from the native apps by message, and routines run on a schedule or on demand from inside the product. If you need to drive agents from your own systems, that is a real gap: Continuum's host daemon exposes an HTTP and WebSocket API with a CLI client.

The docs state plainly that a Linux desktop app is not currently available, yet the download page has reportedly served a Debian package, and at least one hands-on report on Hacker News describes being offered a Linux download. Several articles assert Linux support with no caveat. Until xAI reconciles its own pages, treat Linux as unsupported and unreliable. iPad is not supported at launch and Android is listed as coming soon.

Only for the coding half. Continuum does not operate your apps by computer use, does not hold browser logins for errands, and does not run a Bot while your Mac is asleep unless you enrolled another host. What it does is run Claude Code, Codex, Cursor, Grok, Gemini, and OpenCode on hardware you own, with worktrees, plan and diff gates, a PR pane, quota gauges, and cost by repo, for free.

For repository work, not today. It has a terminal and a filesystem on its cloud computer, so it can genuinely change code, but there is no worktree per task, no per-hunk diff staging, no PR pane, no model choice, and no dry run before a real action. It is a teammate you message, not a review loop. The head-to-head is on Grok Bot vs Claude Code.

No published SOC 2, GDPR, or HIPAA claim was found in the Grok Bot doc set, nor a stated retention period, data-residency option, or encryption specification of xAI's own; compliance defers to Cursor's terms. There is also no queryable organisation-wide audit view yet. The twenty-record routine retention limit was reported by Daily Dose of Data Science and remains unconfirmed by xAI, and Cursor's Legacy Privacy Mode blocks Grok Bot entirely. Ask xAI directly and get the answer in writing before a security review depends on it.

No. Grok Bot is the always-on teammate product on a managed cloud computer. Grok Build is xAI's coding CLI, which defaults to grok-4.6 and speaks ACP, which is how Continuum runs Grok as one provider among several. Grok Automations, formerly Tasks, is a third thing again: scheduled prompts with email triggers. The names collide constantly and almost nobody disambiguates them.

Further reading

Guide

What is Grok Bot? xAI’s always-on AI agents, explained

A read of the documentation rather than the launch post, including the one claim the marketing pages and the docs do not agree on.

Read guide → Guide

Grok Bot pricing: $40 to $300 a month, and the meter nobody caps

Six eligible plans, three of them added on 21 August, an unpublished weekly allowance, and no spend cap behind any of it.

Read guide → Guide

Grok Bot alternatives (2026): 8 always-on agent options compared

Every Grok Bot alternatives list currently ranking was written by one of the products on it. This one is not, and it says plainly where each option loses.

Read guide →
Begin

Run your agents
in Continuum.

Free app. Your subscriptions. Optional hosted inference. Mac stable - Windows and Linux desktop are beta.

Download for Mac See pricing

vendor-neutral · local-first · multi-device