The newest shape in AI at work is not a chat window, it is a roster: you name several bots, give each a job, and talk to them like colleagues in a group chat. Four of them, on the same five criteria.
Worth knowing what a "bot" means here, because the word is overloaded. In this page it means a persistent agent with its own name, its own memory, its own job, and its own way of being messaged. Not a chatbot widget on a website.
OpenAI Dots
OpenAI · cloud
Announced at OpenAI's developer conference in late September 2026. An always-on agent with an emoji-like avatar.
What it isYour "dot" is an always-on agent inside ChatGPT, powered by GPT-6 Astra. The first one is included with your plan at no extra cost.
How you reach itChatGPT on desktop, web and mobile, plus Slack and Teams, and voice. You can also create it only on desktop.
How it worksEach dot gets its own cloud computer and browser, and connects to over 4,000 apps through OpenAI's plugin ecosystem. You give it a goal and define what it may do alone.
GuardrailsIt is trained not to do certain sensitive things on its own, such as changing passwords or moving money, and it escalates decisions that need judgement back to you.
PriceNo standalone price: you need ChatGPT Pro or Business Premium. Pro sits in the US$100 to US$500 a month range depending on tier.
AvailabilityRolling out in eligible markets, excluding the European Economic Area, Switzerland and the UK. Enterprise workspaces need an admin to switch it on.
Model-agnosticNo. You get GPT-6 Astra.
The trade you are making: the smoothest possible experience, in exchange for the bot, its computer and everything it touches living in OpenAI's cloud.
Grok Bot
xAI / SpaceXAI with Cursor · cloud VM
Launched in beta on 11 August 2026. "AI teammates you can give real work to."
What it isNamed bots that each get a persistent cloud computer with a browser, filesystem and terminal, and sign into your tools the way a person would. No API or integration required.
How it worksBots click and type through the software you already use, including tools with no clean API. They keep working after you close your laptop and come back when something needs your approval.
LearningTeach a bot by doing the job once in front of it. It saves the steps as a routine and repeats them.
Team shapeMultiple bots run in parallel, message each other, share a thread, and hand work between themselves in a group chat.
PriceThere is no free tier and no standalone subscription. It is bundled with Cursor and SuperGrok plans, from roughly US$20/month at the entry tiers up to US$200 to US$300/month at the top. Its usage pool is separate from your normal Grok or Cursor limits and resets weekly.
Model-agnosticNo. It runs on Grok.
StatusStill labelled beta, with enterprise access on a waitlist.
The bots sign into your accounts. That is the feature and the risk: they hold live credentials to the systems you use, on a machine you do not control.
Hermes Bot Mode
Nous Research · your machine · open source
The same roster idea, running on your own laptop, where each bot is a full agent rather than a prompt.
What it isBot Mode turns your agent profiles into a roster of named bots in the desktop app. It ships built in and is on by default, so there is nothing extra to install.
What a bot getsIts own name, title, avatar, model, memory, skills, credentials and chat history, isolated from the other bots on a per-profile boundary on your disk.
How you reach itA Bots tab next to Sessions in the desktop app, plus the same bot from the command line if you want it.
Team shapeRoutines (scheduled recurring work, attached to the bot responsible for it), group chats where two to six bots deliberate and escalate to you, and direct bot-to-bot messages.
Where it runsOn the machine that owns the profile. One roster can span local, SSH, Docker or remote backends, and each bot keeps running where it lives.
PriceFree and open source. You pay only for model usage, which is US$10 a month on this machine.
Model-agnosticYes, per bot: one bot can run a cheap fast model while another runs a stronger one.
Honest limit: these bots run on your hardware, so a bot only works while your machine is on. That is the price of the data never leaving.
OpenMausBot
Independent open source · not affiliated with xAI
An open-source answer to Grok Bot: the roster idea, rebuilt local-first on the agent tools you already pay for.
What it isEvery bot in the sidebar is a real agent: Claude or Codex running locally under the hood, with its own personality, model, computer and connected apps.
How it worksA small server on your own machine owns each agent process and translates every provider into one stream. Transcripts and keys stay in a local folder.
Bring your own agentsBots run on the Claude, Codex and Grok command line tools already installed on your computer, using your existing logins and subscriptions. No new accounts, no proxy in the middle.
Give them handsEach bot can get a real computer, either a cloud Linux desktop you watch live or your own Mac, plus hundreds of app connections.
RequirementsmacOS, Windows or Ubuntu, Node 24 or newer, pnpm, and at least one agent tool installed and signed in. This is the most technical item on the page.
PriceFree and open source. A paid Pro tier adds an always-on cloud workspace with computers, voice and scheduled tasks.
Model-agnosticYes, by construction: it does not have its own model, it drives the ones you already have.
What the four actually disagree about
Where the bot's hands are. Dots and Grok Bot give every bot a computer in a vendor's cloud. Hermes and OpenMausBot keep the work on your machine, or on a machine you chose.
Who holds the credentials. Cloud bots sign into your apps from infrastructure you do not administer. Local ones use logins that are already on your disk. For anything client-facing, this is the question that matters, not how smart the bot is.
Whether you can change your mind. Dots and Grok Bot bind you to one model family and one subscription. Hermes and OpenMausBot are harnesses: when a better model appears, you point the bot at it and nothing else changes.
The pattern worth noticing: OpenAI and xAI both shipped this as a premium, cloud-only, per-seat product. The open versions do the same job on hardware you already own, and the honest difference is that you spend thirty minutes setting it up instead of thirty dollars a month.