Two of the loudest agent products of 2026 were launched in the United States and are not available to everyone in Europe. Grok Bot, from xAI and Cursor, went into beta on 11 August 2026. OpenAI's dots were introduced on 29 September. Both give you a colleague-shaped thing: a named agent with a cloud computer of its own that keeps working after you close the laptop.

Founder2 is a different shape of product, built by a company in Waterford, Ireland. Comparing them is useful anyway, because the questions underneath the marketing are the same three: where does the work run, which model does the work, and who decides what the agent is allowed to do.

This is a dated comparison. Product terms, regional endpoints and residency documentation change often, and the useful thing to take from it is not a verdict but a method.

An agent product is a bundle of five decisions: where the agent runs, which model answers, what the agent can see, what the agent can change, and who is responsible when it is wrong. Marketing covers the first two and skips the rest.

Two products can sit in the same category and still be unusable for you, because they differ on the third, fourth and fifth questions. A founder in Dublin with a customer list, a Stripe account and a private repository is not in the same position as someone with nothing to lose.

Three always-on agents, three different answers to where the work happens.
  • Grok BotxAI and Cursor. Beta since 11 Aug 2026.
  • ChatGPT dotsOpenAI. Released 29 Sep 2026.
  • ClaudeAnthropic. The app stores data in the US.

The table below is the part that matters most for a European reader, and it is also the part that is easiest to get wrong, because "available in your country" and "processed in your country" are different claims.

Grok Bot's own documentation names two endpoints: a global one, where requests may be routed between regions, and a United States one that guarantees US processing at a 10% price premium for two text models. Files, collections and the server-side tools — web search, X search, code execution — sit outside that guarantee even on the US endpoint. There is no documented EU endpoint. The contract you end up with depends on which door you came through: xAI's data processing addendum, or Cursor's if the agent was included with a Cursor subscription.

ChatGPT dots has the sharper edge, but for a different reason. OpenAI's help centre states that dots roll out to individual Pro users "in markets excluding the European Economic Area, Switzerland, and the UK". If you pay for ChatGPT yourself and you live in Dublin, Lisbon or Brussels, dots are not available. Business Premium and Enterprise workspaces are the exception, because OpenAI lists dots across supported regions for them. Pricing reflects that split: a Business Premium seat is roughly €130 a month against roughly €26 for a standard Business seat, and the standard seat does not include dots at all.

It would be tidy to stop there, but it would also be wrong. ChatGPT Enterprise does offer European data residency and, for eligible workspaces, European inference residency. What dots do not get is that treatment: during the Enterprise beta, OpenAI documented that dots do not support data residency or inference residency, and its help centre notes that residency support for dots has a narrower scope than for ChatGPT itself. So the region you buy is the region your other work is processed in. The agent you bought it to run is not.

Anthropic's position is different again, and worth including for completeness. Claude's own privacy centre says data is stored in the US, and the API's inference location parameter accepts only a global value or a US one. Keeping Claude in Europe means going through Amazon Bedrock's EU inference profile or Google Cloud's EU multi-region endpoint instead.

Founder2 is operated by Leemer Group from Waterford, Ireland. The application and its database run on European infrastructure, connected Sentry and PostHog sources can be pointed at their EU regions from the integrations screen, and model traffic goes through OpenRouter, which offers EU in-region routing on its Business and Enterprise plans — requests sent to the EU endpoint are decrypted in Europe and routed only to provider endpoints in Europe, and fail rather than fall back to a global route.

The honest caveat is the same one we give in our own product notes: confirm the configuration for your own workspace rather than taking a marketing page for it. Residency claims differ from deployment claims, and the exclusions are where the risk lives.

What is documented, as of 14 November 2026. Check a provider's own terms before you rely on it.
CriterionGrok BotxAI / CursorChatGPT dotsOpenAIFounder2Leemer Group, Ireland
What it isA named agent with its own cloud computer that signs into your apps.A named agent with its own cloud computer that connects to thousands of apps.A plan, a task board and a co-founder, with a sandbox per workspace.
Who runs itxAI (SpaceXSI) or Cursor, depending on the subscription you bought.OpenAI.Leemer Group, an independent company in Waterford, Ireland.
Available to an individual in the EEAYes, on SuperGrok or a qualifying Cursor plan.No. Pro rollout excludes the EEA, Switzerland and the UK.Yes.
Documented processing locationUS. A US endpoint costs 10% more and covers two text models; files and search tools are excluded.Dots have a narrower residency scope than the rest of ChatGPT. During the Enterprise beta they do not support data or inference residency.EU in-region routing is available through OpenRouter; connected Sentry and PostHog sources can be pointed at their EU regions.
Which model answers youNone you choose. Routing is automatic and there is no picker.OpenAI's frontier family, expanded to GPT-6.1 Terra in November 2026.Your choice of nine open-weight models, set per workspace.
How writes are gatedThe Bot works in a computer it shares with your other Bots. xAI says separate Bots are not a security boundary.Per-workspace permissions, with admin controls for connected apps.Human approval before an external write, with the destination and the proposed content shown first.

Product terms change quickly. The dates here are the dates we checked.

The comparison people usually make is which model is smartest. The comparison that decides whether you can use the tool at all is which model you can choose.

Grok Bot does not offer one. Routing is automatic and there is no picker, so the model answering your business question is a decision somebody else made on your behalf. OpenAI's dots sit at the other end: they run on OpenAI's own frontier family, expanded in November 2026 to include GPT-6.1 Terra for parts of the work.

Founder2 ships nine open-weight models, set per workspace: Kimi K3.1 Pro, MiniMax M3.1 Flash, DeepSeek V4.1 Flash, Beam 1.0, MiMo V2.6 Flash, Muse Spark Open, Qwen3.8 27B, GLM 5.3 Flash and Inkling 2 Small as the default. The reasoning behind that choice is not that one of them is best. It is that you can change them without changing your plan, your history or your prices, and that a small task does not need the model you would use for a strategy memo.

There is a second reason, which is harder to admit: an open-weight model on a European endpoint is a way of making the residency claim less rhetorical. The model is not a black box that has to be trusted; it is a set of weights you can point at a provider you have chosen.

The models a Founder2 workspace can be set to, served through OpenRouter.
  • Kimi K3.1 ProMoonshot AI
  • MiniMax M3.1 FlashMiniMax
  • DeepSeek V4.1 FlashDeepSeek
  • Qwen3.8 27BQwen
  • GLM 5.3 FlashZ.ai
  • Muse Spark OpenMeta

Nine are available today. The list changes as new open-weight releases are evaluated.

Both always-on products are good at the part of the job where nothing irreversible happens. Research, drafting, summarising, reading a dashboard: these are all cases where the worst outcome is a wasted ten minutes.

The interesting case is the agent that changes something. A GitHub issue created in the wrong repository. A customer email sent to the wrong person. A refund issued at 2am. These are the actions worth designing for, and they are where the three products differ most.

Grok Bot's answer is architectural: a Bot works on a shared cloud computer, and xAI is explicit that separate Bots are not a security boundary from one another. Files, browser sessions and command-line credentials are shared across the Bots on your account. That is a reasonable design for one person with several teammates, and a bad one for an account with several people in it.

ChatGPT dots inherits the permission model of the workspace around it: an admin can enable dots, connect apps and control who does what, and specialist dots are being previewed with their own identity for access management.

Founder2 takes the narrow route. Writes to external systems — creating a GitHub issue, commenting on one — go through an approval step that shows the destination and the proposed content before anything happens. The point is not that approval makes an action safe. It is that the action has a visible, interruptible moment in it, and that moment belongs to a person.

The deeper difference is about the unit of work. An always-on agent is built around a task described in prose, running on a computer you do not watch. Founder2 is built around a business: a plan that persists, a task board that persists, and a set of connected sources that persist. The agent works against that state rather than owning a computer. If the context is wrong, the output is wrong either way — but with a persistent plan, you can see what the agent was working from and correct it.

Picture a founder with a Stripe account, a GitHub repository and a half-finished onboarding change, somewhere in the EU.

With ChatGPT dots, if their plan supports it and their workspace is eligible, they get an agent with a cloud computer that can sit in their tools. If they pay for ChatGPT individually, they get none of it. Either way, the agent's model is fixed and its region is not the one they chose.

With Grok Bot, they get a named teammate that can drive a real desktop. The computer is in the United States. Their session data goes with it, shared across every Bot on the account. The model is whatever routing selects.

With Founder2, they get a plan with the actual subscription numbers next to the actual target, a task board, and a co-founder that can read Stripe, GitHub, Sentry, PostHog and Slack snapshots. A proposed GitHub issue appears with its wording and repository visible before it is created. Models are theirs to pick, and they can pick a different one for a cheap task than for an expensive one.

None of that is automatically better. A founder who wants an always-on assistant that lives in their browser tabs does not need a plan document, and the Grok Bot design may suit them better. The point is that the two products are solving different problems, and the ones that read most alike in a launch post are the ones that differ most once you look at where the bytes go.

Before you sign a DPA, work through the exclusions. For Grok Bot: does the guarantee cover the files and the search tools, or only the text model? For ChatGPT dots: is the feature you actually want inside the residency boundary, or only the rest of the workspace? For any provider: what is excluded from "processed in the EU" — account data, logs, abuse-monitoring metadata, backups?

The exclusions are not a footnote. They are where a residency claim usually stops.

If you are in the EEA and pay for your own AI subscription, ChatGPT dots is simply not on the table today, and that settles the first question before quality enters it. If you want an always-on agent on a shared desktop and you work alone, Grok Bot is a coherent product with a clear model of its own.

If you are a founder running a business with a plan, a task list, revenue in Stripe and code in GitHub, and if the question of where your data is processed is a question your customers or your lawyer will eventually ask, then the choice narrows to products that keep your business state in one place and keep you in the loop when it changes.

Whichever you pick, write the residency question down and get the answer in writing from the provider, not from a summary. Our AI co-founder guide covers what to look for in the tool itself, and AI agent approvals covers how to review a proposed action before you let it through.