Grok Bot vs Hermes vs Dots vs Claude Duet

In ten weeks, every major AI lab shipped the same idea. xAI put Grok Bot into beta on 11 August. Meta launched Muse on 8 September. OpenAI announced Dots on 29 September, and Anthropic opened a waitlist for Claude Duet on 21 October. Nous Research’s open-source Hermes Agent had been doing a version of it well before any of them. Put Grok Bot vs Hermes side by side and you get the two ends of the category: a hosted team of bots on one side, software you run yourself on the other. The commercial offerings from OpenAI and Anthropic sit between them.

The category goes by a few names: always-on agents, proactive agents, personal agents. The common thread is that these products don’t wait for a prompt. Each one gets its own computer in the cloud (or yours), connects to your email, calendar and business apps, works while your laptop is closed, and messages you when something needs a decision. That is a different proposition from a chatbot, and it changes what you should be evaluating.

TL;DR: Claude Duet and OpenAI Dots are the most polished managed agents, but Duet is US-only for now and both sit on premium plans. Grok Bot is built for running several role-based bots at once. Hermes Agent is free, open source and runs anywhere, including Australia, but you own the security. For most Australian businesses in late 2026, the practical choice is Dots or Gemini today and Duet when it arrives.

What an always-on agent does that a chatbot can’t

A chatbot answers when you ask. An always-on agent has three things a chatbot doesn’t: a persistent workspace (a virtual machine with a browser, files and sometimes a terminal), standing access to your connected accounts, and a schedule. Claude Duet, for example, sends a morning brief, then checks your connected apps every few hours and only alerts you when something needs attention. Dots can be reached through ChatGPT, Slack or Teams, and you can open your dot’s computer at any time to see what it has been doing.

The business value is in the background work: chasing overdue invoices, triaging an inbox, preparing payroll inputs, monitoring a Shopify store, drafting follow-ups after a sales call. Anthropic’s own list of use cases for Duet reads like a small business owner’s to-do list, and that is deliberate. These products are aimed at the people who have too many systems and not enough hours, which is most of our clients at Avatar Studios’ AI and automation practice.

The risk is also in the background work. An agent that can send email, pay a bill or sign a document on your behalf is a new kind of access, and each vendor has taken a different position on how much it should do without asking.

Grok Bot vs Hermes vs Dots vs Claude Duet at a glance

Prices checked on 26 October 2026 against each vendor’s announcement and plan pages. All prices are in US dollars.

Two other entrants deserve a mention. Meta’s Muse is free for most use and runs in a dedicated “Muse Secure VM”, but it is rolling out in the US first. Google’s Gemini Spark, since renamed Tasks and folded into the Gemini app, is the one most Australians can use today: Google announced it for Google AI Pro subscribers in Australia, and it runs tasks on Google’s remote computers while your laptop is closed.

How each one approaches the job

Claude Duet is the most explicitly business-focused. It inherits the memory, skills and connected apps from your Claude account, which makes setup almost nothing if you already use Claude, and it has access to more than 6,000 connectors including Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot, DocuSign and Shopify. Its guardrails are the strictest of the group. It shows you the action and waits before it sends a message, makes a payment, signs a document or deletes anything. It will not trade securities or move money between accounts even if you approve it, and Anthropic says it never sees your passwords or sign-in tokens. Anthropic is candid about the gaps: personal errands such as booking tickets are weaker, and you can’t yet watch its browser. The bigger gap for Australian readers is that it’s a US-only waitlist, with no date for other markets.

OpenAI Dots is the most capable generalist and the most expensive entry point. Each dot runs on GPT-6 Astra with its own cloud computer and browser, connects to over 4,000 apps through ChatGPT plugins, and works inside ChatGPT, Slack and Teams, including by voice call. Control is rules-based: dots start with built-in rules for when to act and when to ask, and Custom Rules let you allow, require approval for, or block specific actions. Some tasks, such as changing a password, always stay with you. OpenAI is also piloting specialist dots for business and working with Microsoft to manage them through Agent 365, which matters for any business running on Microsoft 365. Your first dot is included with Pro or Business Premium, with extended limits for the first month after launch; OpenAI hasn’t published what heavier use will cost after that.

Grok Bot takes a different shape. Instead of one assistant, you get a set of named bots, each with a job, persistent context and its own task history. They coordinate in group chats and hand work to each other, and Team Bots let one person share a bot with their Cursor team. Each bot runs on a persistent cloud computer with a browser, file system and terminal. One detail from Cursor’s documentation is easy to miss: anything placed on the shared computer is available to all your bots. If your “finance bot” and your “marketing bot” share a machine, they share whatever is on it. Grok Bot usage is metered separately from your Grok or Cursor plan and resets weekly.

Hermes Agent is the open-source answer, and its traction in Australia is striking. Australians searched “hermes agent” about 12,100 times in September 2026, according to our keyword data, and it has held five-figure monthly search volume since at least June. (“grok bot” jumped to 40,500 searches in September, its first full month after launch.) The project also has more than 250,000 stars on GitHub. It runs in a terminal, on a $5 VPS, or in a serverless sandbox, and talks to you through Telegram, Slack, WhatsApp, Signal, Discord or email. You choose the model provider (Nous Portal, OpenRouter, OpenAI or your own endpoint), and it writes its own reusable skills after it completes complex tasks. The trade-off is that security is your job. The default “smart” approval mode uses a second model to judge risk, and Nous’s own security guide says that for maximum security you should run it on a separate machine or VM. If that sentence means nothing to you, Hermes is not the right first agent for your business. It is the same choice we described in our look at the OpenClaw agent framework, Hermes’ closest open-source rival.

Which agent fits which business

The right choice depends less on which model is smartest and more on where your work already lives and how much risk you can absorb.

A sole trader or small team on Google Workspace can start with Tasks in Gemini today. It sits on a Google AI Pro plan rather than a US$100-a-month tier, and it already has access to the Gmail, Calendar and Drive data that most of the agent’s value depends on.

A business standardised on Microsoft 365 and Teams should look hardest at Dots. The Teams integration and the planned Agent 365 integration point to admin controls that will fit an IT team’s existing processes, and Business Premium seats keep the agent inside a managed workspace rather than an individual’s personal account.

A business that handles payroll, invoicing and contracts through tools like QuickBooks, Gusto, Stripe and DocuSign is the profile Claude Duet was designed for, and its refusal to move money is a sensible default for a finance-adjacent agent. Join the waitlist, but don’t plan around a launch date that Anthropic hasn’t given.

A business with strict data sovereignty requirements, or a technical founder who wants to keep data on Australian infrastructure, is the case for Hermes. Self-hosting on an Australian server with a provider whose data processing terms you have checked is the only way among these options to know exactly where your data sits. Use our AI vendor data privacy checklist before you connect it to anything that matters.

Grok Bot makes the most sense where the work splits naturally into roles: a research bot, a content bot and a reporting bot that hand off to each other. That is an attractive model for agencies and marketing teams, provided you keep sensitive data off the shared computer.

What to set up before you switch any of them on

Whichever product you choose, the first month decides whether an always-on agent saves time or creates a mess. Start with read-only access. Let the agent brief you and draft work for a fortnight before it is allowed to send anything. Connect only the accounts it needs for a defined job, such as accounts receivable follow-up, rather than everything at once. Write down, in plain language, what it may do without asking, and match that to the product’s controls: Duet’s approval list, Dots’ Custom Rules, Hermes’ approval modes.

Watch the cost as well as the output. Duet’s background checks count against your Claude plan limits, Grok Bot has its own weekly usage pool, and self-hosted agents bill per token, so an agent that checks your inbox every 30 minutes is a running cost, not a one-off.

Finally, expect the rules of engagement to change. On 6 October, Sierra and Meta announced the Personal Agent Protocol, an open standard backed by Walmart, Stripe and Shopify that defines how personal agents sign in to and act on business websites. If your customers start sending agents to your website to check stock, book appointments or request refunds, your site will need to handle them. That is a strategy question as much as a technical one, and it is the kind of work our strategy and advisory team helps businesses plan for.

The verdict

Grok Bot vs Hermes is the clearest way to see the choice in front of you: a managed, multi-bot service where someone else runs the infrastructure, against open-source software where you control everything and carry the risk. Dots and Claude Duet are the polished middle ground, and Duet has the most thoughtful guardrails of the four, but Australian businesses can’t buy it yet. Our recommendation for most Australian SMBs is to pilot one managed agent on one well-defined, read-heavy job this quarter, measure the hours it saves, and only then give it permission to act.

If you’d like help choosing an agent, scoping a pilot or setting the guardrails, talk to Avatar Studios about AI and automation.

Frequently Asked Questions