Coding agents
Claude Duet, OpenAI Dots or Grok Bot: comparing the always-on agents
Three agents that work in the background on a cloud computer, compared on vendor, model, connectors, safeguards, price and availability in France, from a developer's point of view.
In less than three months, three vendors have launched an agent that no longer just answers: it has its own computer in the cloud, watches your apps, works while you are away and comes to you for important decisions. Grok Bot arrived on August 11, 2026, OpenAI Dots on September 29 and Anthropic's Claude Duet on October 21. Meta is also in the race with Muse, an agent aimed more at consumers, which we leave aside here (Trending Topics).
For readers in France, availability is often the deciding factor. Claude Duet is US-only; Dots excludes the EEA for Pro subscribers, but not for Business Premium seats; Grok Bot publishes no country list, but its computers run in the US. Details below.
The comparison table
| Criterion | Claude Duet | OpenAI Dots | Grok Bot |
|---|---|---|---|
| Vendor | Anthropic | OpenAI | SpaceXSI (formerly xAI, legal entity SpaceXAI LLC); product run by Cursor, a SpaceX subsidiary |
| Launch | Oct 21, 2026, research preview | Sept 29, 2026, gradual rollout | Aug 11, 2026, beta |
| Model | Claude Opus 5.5, cannot be changed | GPT-6 Astra | Picked per task: xAI or third-party models (including Claude Opus 5.5), no picker |
| Where to use it | iOS and Android app, claude.ai | ChatGPT (desktop, web, mobile), Slack, Teams; texting in a limited US beta | macOS, Windows, Linux, iOS, Android apps; Slack (team bots); X (@bot) |
| Cloud computer | One virtual machine with its own browser | One cloud computer per dot; optional access to your computer | One computer shared by all bots on the account (one screen per bot) |
| Number of agents | One agent, several channels | One primary dot for now | Several bots in parallel (about 50 per account) |
| Connectors | Over 6,000 connectors and plugins | Over 4,000 apps through plugins | "Connect apps" plugins (GitHub, Google Workspace, Microsoft 365…) plus computer use |
| Safeguards | Approval before sending, paying, signing, deleting; securities trading and transfers blocked | Built-in rules and Custom Rules; auto-review; read-only proactive research | Auto Review (independent review model); "Ask first" rules; local commands need approval |
| Price | Included in Pro ($20/month) and Max (from $100/month), on the plan's usage limits | First dot included in Pro ($100, $200 or $500/month) and Business Premium ($125/seat/month) | Included in The X Subscription Premium ($30/month), Super ($100) and Ultra ($200), and in legacy Cursor plans |
| France / EU | No: US only | Pro: no (EEA, Switzerland, UK excluded); Business Premium: "all supported ChatGPT regions" | No country list published; computers in the US; @bot tagging unavailable in the EU |
Table sources: official pages from Anthropic, OpenAI and SpaceXSI, detailed in each section.
Claude Duet (Anthropic)
- How it works: a single agent, in an app separate from Claude, that sends a brief each morning, checks your accounts every few hours and only messages you when needed; it has a virtual machine with its own browser (Anthropic).
- Model: Claude Opus 5.5, with no choice (Claude Help Center).
- Safeguards: explicit approval before sending, paying, signing or deleting; securities trading and moving money between accounts blocked even with approval; the auto mode classifier (the one from Claude Code) and a dedicated classifier against sensitive data leaks.
- Price: included in Pro and Max at no extra cost, but background work draws on the plan's limits (Claude Duet, Anthropic).
- Availability: waitlist, Pro and Max in the US only; no Team or Enterprise for now.
Our detailed article: Claude Duet, Anthropic's proactive agent.
OpenAI Dots
- How it works: always-on agents in ChatGPT, each with its own cloud computer and browser; you start with a primary dot, and OpenAI envisions teams of dots later. You reach them in ChatGPT, Slack or Teams, or by voice call (OpenAI).
- Model: GPT-6 Astra (OpenAI, MacRumors).
- Connectors: over 4,000 apps through the plugin ecosystem. Access to your own computer is optional and off by default (OpenAI Help Center).
- Safeguards: background "proactive research" only uses read-only tools; Custom Rules let you allow an action, require approval or block it; auto-review checks sensitive actions; some tasks, such as changing a password, always stay with the user (OpenAI).
- Price: the first dot is included at no extra cost in ChatGPT Pro and Business Premium. Conversations with your dot do not count toward ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work do (OpenAI). Pro comes at $100, $200 or $500 a month (OpenAI Help Center); a Business Premium seat costs $125 a month, or $100 billed annually (OpenAI).
- Availability: rolling out to Pro users outside the EEA, Switzerland and the UK, to Business Premium in all supported ChatGPT regions, and as a beta for Enterprise when the admin turns it on (OpenAI Help Center).
Grok Bot (SpaceXSI / Cursor)
- Who makes it: the official pages live on x.ai, signed "© 2026 SpaceXAI LLC" under the SpaceXSI brand, the new name of the former xAI (SpaceXSI, The Next Web). But the product is run by Cursor: you sign in with a Cursor account, and the bots' computers run in Cursor's cloud (SpaceXSI Docs, Cursor). Cursor (Anysphere) has been part of SpaceX since August 14, 2026 (Cursor).
- How it works: a team of named bots that work in parallel, message each other, coordinate in group chats and learn workflows by demonstration (SpaceXSI). SpaceXSI sets practical limits of roughly 50 bots per account and six per group chat (SpaceXSI).
- Computer: a persistent cloud computer with a browser, filesystem and terminal, shared by all bots on the account: files, browser sessions and credentials are common; isolation is between users, not between bots (SpaceXSI Docs). The computers run in the United States (Cursor).
- Model: no single model and no picker. For each task, Grok Bot picks an xAI model or a third-party model; Cursor's help explicitly mentions Claude Opus 5.5 (Cursor). Elon Musk announced on October 7 that Grok Bot would use "the best back end model for any given task", including Claude Opus 5.5, Midjourney and Suno (The Next Web). Saying that Grok Bot "relies on Claude Opus 5.5", as Trending Topics wrote, therefore goes too far: Opus 5.5 is one of the possible models.
- Safeguards: Auto Review, an independent review model, evaluates risky actions (shell commands, app calls, computer use, routine changes) and can let them proceed, require approval or deny them; each user can add "Ask first" rules. Commands on your own machine ask every time by default (Cursor). Approvals from work started without you (a routine, another bot) expire after about 10 minutes (Cursor).
- Price: included in The X Subscription's Premium ($30/month), Super ($100) and Ultra ($200) plans (SpaceXSI), and in legacy Cursor Pro, Pro+, Ultra and Teams plans; a SuperGrok subscription can also be linked to a Cursor account. Usage resets weekly (Cursor).
- Availability: SpaceXSI publishes no country list for the app. The only documented regional restriction concerns tagging @bot on X, unavailable for accounts based in the European Union, the UK, Australia, Iceland, Liechtenstein and Norway (SpaceXSI Docs).
Official sources contradict each other on Grok Bot. The launch post limited the app to desktop and iOS, with an enterprise waitlist and usage "separate" from Grok and Cursor plans (SpaceXSI). Current documentation lists macOS, Windows, Linux, iOS and Android (SpaceXSI Docs), the product page says Enterprise is "generally available" (SpaceXSI), and on The X Subscription Grok Bot draws from the same weekly pool as Cursor and the Grok apps (Cursor). Every's comparison still shows legacy Cursor plan prices ($20, $60 and $200) (Every). We go with the most recent pages.
What matters to a developer
Grok Bot is the most "technical". Its computer exposes a terminal and a filesystem, it offers a GitHub connector, and its routines can run on a schedule or on an event; the design post illustrates triggers such as a pull request being opened or checks failing (SpaceXSI). SpaceXSI cites an engineering bot that reproduces a bug in the product UI, files the ticket and hands the fix to a debugging bot (SpaceXSI). Downsides: credentials and sessions shared across all your bots, no model choice, and for teams, Cursor's model allowlist does not apply to Grok Bot (Cursor).
Dots is the closest to Codex. OpenAI describes a dot that watches customer feedback, builds and tests small fixes, then brings you pull requests with videos attached (OpenAI). It can create tasks in Codex cloud environments and, with your permission, work on your computer (OpenAI Help Center). Downsides: no access for a Pro subscriber in France, and Codex tasks use your allowance.
Claude Duet is the most coordination-oriented. Anthropic mentions prototyping and web apps, but Duet cannot see the files on your computer, can read your work in Claude without changing it, and you cannot follow its browser (Claude Help Center). For the code itself, Anthropic's tool is still Claude Code; on OpenAI's side, see our Codex page, and for Cursor, the Cursor page.
Our take
- You are in France with a personal subscription: none of the three is clearly available. Duet and Dots (Pro) exclude Europe; Grok Bot does not explicitly exclude it, but processes data in the US.
- You are a French company: Dots through Business Premium is the only one of the three whose documentation mentions no European exclusion; check your data compliance before connecting mailboxes.
- You want an agent that touches code: Grok Bot (terminal, GitHub, event-triggered routines) and Dots (Codex) go further than Duet, which remains a coordination assistant.
Sources
- Anthropic: Introducing Claude Duet (Oct 21, 2026), verified 2026-10-22
- Anthropic: Claude Duet product page, verified 2026-10-22
- Claude Help Center: Get started with Claude Duet, verified 2026-10-22
- Anthropic: Claude pricing, verified 2026-10-22
- OpenAI: Introducing dots (Sept 29, 2026), verified 2026-10-22
- OpenAI Help Center: Getting started with your dot, verified 2026-10-22
- OpenAI Help Center: About ChatGPT Pro tiers, verified 2026-10-22
- OpenAI: Premium seats are coming to ChatGPT Business (Aug 10, 2026, updated Aug 25), verified 2026-10-22
- SpaceXSI: Introducing Grok Bot (Aug 11, 2026), verified 2026-10-22
- SpaceXSI: Grok Bot product page, verified 2026-10-22
- SpaceXSI: Designing Grok Bot for a world of persistent agents (Sept 3, 2026), verified 2026-10-22
- SpaceXSI Docs: Grok Bot overview, verified 2026-10-22
- SpaceXSI Docs: Use the computer and apps, verified 2026-10-22
- SpaceXSI Docs: Tag @bot on X, verified 2026-10-22
- Cursor: Grok Bot security, verified 2026-10-22
- Cursor: Which AI models power Grok Bot?, verified 2026-10-22
- Cursor: Grok Bot FAQs, verified 2026-10-22
- Cursor: Grok Bot plans and billing, verified 2026-10-22
- Cursor: Cursor is now a part of SpaceX (Aug 14, 2026), verified 2026-10-22
- The Next Web: Musk's Grok Bot will use Anthropic's Claude, Midjourney and Suno models (Oct 7, 2026), verified 2026-10-22
- MacRumors: OpenAI Launches Always-On 'Dots' Agents (Sept 29, 2026), verified 2026-10-22
- Every: personal agents comparison, verified 2026-10-22
- Trending Topics: Claude Duet, Anthropic's New A.I. Agent (Oct 21, 2026), verified 2026-10-22
Key takeaway: all three agents work on a cloud computer and ask for approval for sensitive actions, but only Dots, through Business Premium, does not explicitly exclude Europe; Grok Bot is the best equipped for code, Claude Duet the most cautious.
Spotted an error in this article? Report it · Our corrections policy