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

The short answer

Grok Bot launched on 11 August 2026 as SpaceXAI's always-on agent product: named Bots that live on a persistent cloud computer, remember what they did last time, learn a routine by watching you do it once, and message you back like a colleague. It is a genuinely good product shape. It is also gated behind a Cursor account and a subscription that starts well above what most people expect, and it makes a set of trade-offs that not everybody wants to make.

Before the table, the one question that sorts this whole category: is the work code, or is it office admin? Almost every "best Grok Bot alternative" argument collapses once you answer it, because the products that are good at driving a browser through an expense report are not the products that are good at running four coding agents against four branches of the same repository.

What Grok Bot actually does, so you know what you are replacing

A replacement list is worthless without an accurate description of the thing being replaced, and the descriptions in circulation are not accurate. The most-repeated error is the claim that each Bot gets its own computer. xAI's own documentation says the opposite, and it says it twice.

• "All of your Bots use the same persistent cloud computer." One Linux VM per account, with a browser, a filesystem and a terminal.
• "Each Bot gets its own screen on that computer, so several Bots can use browser and desktop tools in parallel without getting separate security boundaries." Parallelism, yes. Isolation, no.
• "Treat a login or file placed on the computer as available to all of your Bots." This is the sentence to read twice before you sign a Bot into your bank.
• Named Bots keep memory, files, browser sessions and preferences across turns. The docs describe context that "compounds instead of resetting", which is the real product advance here.
• Routines are learned by demonstration. Walk a Bot through a multi-step path once, and it saves it and re-runs it on a schedule or on demand.
• Connectors and MCP where an API exists, computer use where it does not. That second half is what lets a Bot operate a SaaS product that never shipped an API.

So Grok Bot is four things bundled: a place for agents to live, memory that survives, a way to teach procedures without writing code, and the ability to drive a graphical interface. Different alternatives replace different subsets, and none of them replaces all four. Be honest with yourself about which of the four you are actually buying.

There are also hard ceilings that xAI does not put on its marketing pages, and they matter if you were imagining a large fleet of Bots. These come from a single published source, Daily Dose of DS, and are the only figures we could find anywhere, so treat them as indicative rather than contractual and verify before you plan around them.

The retention figure is the one with teeth. The twenty-run-record limit is reported by Daily Dose of Data Science and unconfirmed by xAI. It is not an audit trail, and combined with the documentation describing an audit view as still coming, it means a Bot running a daily routine has forgotten most of what it did inside a month. If you need to reconstruct what happened for a customer, a regulator or an incident review, that is the constraint to check first.

The access path matters too, because it is the first thing that disqualifies people. As of 22 August 2026 the documentation lists the eligible plans as "SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium", and a Cursor account is required on every path. The 21 August expansion announcement puts it as "Grok Bot is now included with all SuperGrok Plus, Cursor Pro+, and Cursor Teams plans". Plain Cursor Pro at $20 a month is still not eligible.

Where each option runs, and what that actually costs

This is the table the vendor lists will not print, because four of its columns are uncomfortable for somebody. Price shapes are given qualitatively where a vendor has not published a number we could verify this week; inventing one would defeat the purpose of the page.

Read the third column first. "Works when your device is off" is the single honest advantage of every cloud product on this list, and it is the reason people put up with the rest. If your answer to it is "I have a Mac mini or a Linux box that never sleeps", the entire cloud tier loses most of its argument and you should be shopping in the own-hardware tier instead.

Read the fifth column second. A shared computer that holds every login your agents have ever used is a convenience feature and a blast radius at the same time. xAI is refreshingly direct about this in the docs, which is more than most vendors manage, but the consequence stands: a Bot that goes wrong has your sessions, and the audit trail says it was you.

The developer shortlist: Continuum, Claude Code, Devin

If the work you want an always-on teammate for is code, this is the tier to shop in, and Grok Bot is a slightly odd fit for it in the first place. A Bot on a shared cloud VM can clone a repository and run tests, but it has no worktree model, no per-branch isolation, no diff review surface, and no way to tell you which subscription paid for what.

Continuum is the closest replacement that does not rent you a machine. It is a free workbench that runs the agent CLIs you already have, under your own logins: Claude Code, Codex, Cursor, Grok, Gemini, and OpenCode side by side in one sidebar, each Code session in its own git worktree so two agents can work the same repository without fighting over files. Every session exposes chat, plan, diff, pull request and terminal panes. Live five-hour and weekly quota gauges sit beside the work, and spend is recorded by repository, provider and model rather than reconstructed later from a vendor's usage bar. It runs on Mac, Windows, Linux, iPhone and the web, and the desktop and mobile clients drive the same session on the same host.

The objection to that model is well known, and it was stated on Hacker News by phoghed on 20 August 2026: they very much did not want to run this locally, they wanted one instance running somewhere else that they could reach from all their devices, and their conclusion was that nobody was going to run this locally. That is a fair objection and it deserves a direct answer rather than a shrug. Continuum's answer is that the execution host and the control surface are separate things: enroll a Linux box or an always-on Mac as the host, and steer it from an iPhone, a browser or a second laptop. You get the "reachable from everywhere" half without handing your logins to a shared VM you do not control.

• Continuum wins when the work is code, you already pay for one or more agent subscriptions, you want more than one lab available, and you have a machine that can stay on.
• Claude Code wins when you want one excellent agent in a terminal and nothing else. It is not always-on and does not pretend to be; it is the engine, not the garage.
• Devin wins when you want to hand over a whole ticket and get a pull request back without watching, and you are willing to pay platform prices for managed autonomy and governance.
• Grok Bot still wins when the coding work is incidental and the real job is a Bot that logs into six SaaS tools and pushes work through them.

The own-hardware shortlist: OpenClaw and Hermes Agent

OpenClaw is in every Grok Bot alternatives list currently ranking, and in the OpenAlternative directory page too. It has earned that slot: it is the open-source, MIT-licensed, run-it-yourself answer, and if what you object to about Grok Bot is that somebody else holds the computer, OpenClaw removes that objection completely.

What the lists do not do is price it honestly. "Free, minus engineering time" is the phrase that keeps appearing, and the second half of that phrase is doing enormous work. A realistic OpenClaw deployment is a $5 to $30 a month VPS, plus model API spend that is metered with no ceiling, plus the ongoing cost of being your own operator. A commenter in the Hacker News launch thread put the maintenance cost in one sentence: OpenClaw broke every update for a month, so they gave up and moved to Hermes. That is one person's experience and not a verdict, but it is the failure mode to plan for, and it is the reason a free framework and a free application are not the same product.

Hermes Agent, from Nous Research, is the other serious desktop-native option and the one the churning OpenClaw user in that thread moved to. It is cross-platform, keeps persistent memory, and sandboxes what it runs. It is a closer shape match to Grok Bot than Continuum is, in the sense that it is a general teammate rather than a coding workbench. If your work is not primarily code, look there before you look at Continuum.

The office-automation shortlist: ChatGPT Work, Manus, Lindy

This is the tier where Grok Bot is genuinely strong and where we would not tell you Continuum is the answer, because it is not. If the job is "log into our expense system, pull last month's approvals, reconcile them against the card statement, and email me the exceptions", you want an agent that can drive a graphical interface, keep a session alive between runs, and be taught the path by demonstration. That is exactly the product Grok Bot is.

ChatGPT Work is the closest competitor by market position rather than by mechanism. Its agent surface is organised around tasks and sessions rather than named persistent teammates, which is a real architectural difference: a per-session sandbox is a narrower blast radius than a shared computer that accumulates every login your agents have used, and a persistent named teammate accumulates context a fresh session throws away. Neither is strictly better. They are opposite bets, and the head-to-head comparison works through where each one lands.

Manus is the purest "it runs in a cloud somewhere and bills you by credit" option, and it is the answer if you want the always-on property without a Cursor account attached. Lindy and the products around it are the business-automation lane: named assistants, trigger-driven, aimed at operators rather than engineers, and generally sold per seat.

• Pick ChatGPT Work if the company already runs on OpenAI and procurement is the hard part.
• Pick Manus if you want a cloud agent with no subscription entanglement and are comfortable with credit metering.
• Pick Lindy or a peer if the buyer is an operations lead rather than a developer and the workflows are trigger-shaped.
• Stay on Grok Bot if the learn-by-demonstration routine and the computer-use fallback are the features you actually use. Nothing else on this list does that pair as cleanly today.

One caveat that applies to this whole tier, Grok Bot included, and that no vendor page mentions: none of these products ships a dry-run mode. eesel's review of Grok Bot documents it directly, and the consequence is worth stating in plain language. A test run performs real work. If you rehearse an email send, you sent an email. An approval gate that fires after the fact does not undo anything the agent already did. Rehearse in a sandbox account, not in production.

Why every alternatives list you find is already stale

Grok Bot is eleven days old at the time of writing and the commercially decisive facts have already moved twice. That matters for a shopping guide, because the pages currently ranking were written before the changes and nobody is maintaining them.

Two practical consequences. First, price any Grok Bot comparison against docs.x.ai directly rather than against an article, including this one, and check the date stamp on anything you read. Second, treat the entry price as genuinely lower than the internet thinks: the cheapest eligible path today is a Cursor Teams Standard seat at $40 a month, not the $120 figure that most of the coverage still leads with. That changes the calculation for small teams considerably.

How to choose in one afternoon

Feature tables are how you make a shortlist. They are not how you decide. Four questions, in this order, will get you to an answer faster than any comparison grid, this one included.

Then run one real task through your top two, on the same day, and measure the thing you actually care about: how much of your attention it consumed and how much of its output you kept. That is a better signal than any benchmark, and it is the only test that catches the failure modes nobody writes about.

If your answer to question one was "code" and your answer to question two was "yes", the next page to read is the Grok Bot and Continuum comparison, which does the cost arithmetic properly and concedes the four things Grok Bot does better.

Questions people ask

Keep reading

Continue through Head to head with Grok Bot vs OpenClaw: managed cloud or a framework you maintain.