What AI teammates actually cost per month: Grok Bot, Squad, Hermes Cloud, and running your own
The sticker is never the total. Grok Bot rides a $20 to $300 subscription plus metered usage; Squad is $99 plus the ChatGPT plan it recommends; a self-hosted squad is a small server plus the model bill; a managed pod is $29 with usage included.
The sticker is never the total, because every one of these products splits the bill into a service fee, a model bill and a computer, and each hides a different one. Grok Bot has no price of its own: it rides a Cursor or SuperGrok subscription (Cursor Pro at $20 a month is the cheapest door since August 26, 2026) with weekly usage included and extra usage metered. Squad is $99 a month to Squad plus the AI plan you bring, and they recommend ChatGPT at $100 to 200, so their own all-in figure is $199 to 299. A self-hosted OpenClaw or Hermes squad is free software plus a $5 to 25 server plus whatever the model costs, plus your time. A managed pod is $29 a month with the first $10 of model usage included. Same idea, four bills.
The table, as of September 2026
| Route | Service fee | Model bill | The computer | A realistic month |
|---|---|---|---|---|
| Grok Bot | None of its own; requires a Cursor plan (Pro $20/mo, Teams $40/seat) or a SuperGrok plan | Weekly usage included with the plan; extra “billed from model and token cost” | Included: one shared computer per account | $20 to $300 depending on the plan, plus overage |
| Squad | $99/mo flat, unlimited teammates | The plan you bring; ChatGPT at $100 to 200 recommended | Included: one dedicated computer per squad | “$199 to 299 all-in” by Squad’s own figure, plus $10/mo per agent inbox |
| Hermes Cloud | A Nous Portal tier (Free, Plus, Super, Ultra; prices not on the cloud page) | Credits; “$10 minimum in credits, or an active subscription” to deploy | Included; “scales to zero when idle,” you pay while it works | Preview; not yet priceable from the public page |
| Self-hosted OpenClaw or Hermes | $0 | API usage, typically $5 to 30/mo for one busy personal agent; or a subscription you already have | A VPS at $5 to 25/mo (a 2 vCPU / 4 GB Hetzner box is about €6.50) | $10 to 60, plus the hours you spend as the administrator |
| Managed pod (Everpod) | $29/mo, which includes the computer | Up to $10/mo included; bigger tiers at $59 and $99 include $40 and $80; your own key adds nothing on top | Included: one private machine per customer | $29, or the plan you bring plus $29 |
Grok Bot: a feature of a subscription you may already have
There is no standalone Grok Bot plan; access comes with an eligible Cursor or SuperGrok subscription, and the eligible list widened on August 26, 2026 to include Cursor Pro, SuperGrok and every Cursor Teams plan. Cursor’s pricing page now lists “Grok Bot access” on Pro at $20 a month and Teams at $40 a seat, with “higher” and “highest” limits on Pro+ and Ultra. When we read xAI’s own page in mid-August the top tiers were Cursor Ultra at $200 and SuperGrok Heavy at $300 a month. The usage model is the part to price: xAI’s docs say subscriptions “include weekly usage,” that eligible accounts can add “on-demand usage billed from model and token cost,” and that Grok Bot’s usage is separate from the plan’s own Grok or Cursor usage. So the monthly number is the plan you would have anyway, plus whatever an always-on roster burns past its weekly allowance, which nobody can quote you in advance. If you already pay for Cursor Pro, Grok Bot’s marginal price is close to zero until the allowance runs out.
Squad: the fee is the small number
Squad’s $99 is deliberately flat: unlimited teammates, “no per-task AI charges from us,” a seven-day money-back guarantee. The larger line is the one they hand you: the work runs on a consumer AI plan you connect, and their recommendation is ChatGPT at $100 to 200 a month, with the option to “stack multiple accounts with fallbacks” when one runs dry. Their own resources pages quote “$199 to 299 a month all-in” and compare it to a junior hire. Add-ons are modest: $10 a month per teammate for a real email inbox, and research tools on credits (1,000 a month included, 5,000 for $5). What is not on the pricing page, and worth asking about before the payroll login goes in, is whether an unattended squad on a consumer ChatGPT plan is within OpenAI’s terms; Squad’s public pages do not address it. What Squad is covers the product itself.
Hermes Cloud: not yet priceable
Nous Research’s hosted Hermes is in preview. The public cloud page names the Portal tiers (Free, Plus, Super, Ultra) without prices, says all paid tiers include monthly credits and 200-plus models, that deploying a cloud agent needs “$10 minimum in credits, or an active subscription,” and that the agent “scales to zero when idle” so you “only pay while it works.” That is a different meter from everyone else’s (usage, not a flat computer), and until the tier prices are on the page it is the one route on this list we cannot put a monthly figure on.
Doing it yourself: the cheapest bill and the largest invoice
OpenClaw and Hermes are free. The computer is a small VPS, about €6.50 a month for a 2 vCPU, 4 GB Hetzner box or around €23 for the dependable tier above it, and the model bill is whatever you use: metered API access runs $5 to 30 a month for one busy personal agent, or nothing extra if you connect a ChatGPT subscription you already pay for. A multi-agent squad multiplies the model line, not the server line; two agents on one 4 GB box is routine. The invoice that does not appear on any page is your time: installing, mounting the right folders, keeping the Gateway off the internet, patching, backing up, and building the board the products sell. For a developer that is an evening and then an hour a month. For anyone else it is the reason the products exist.
The managed pod: the computer, priced
Managed AI agent hosting is the fourth shape: the open software, a machine of your own, someone else’s operations. Everpod’s pod is $29 a month for one private, always-on machine running OpenClaw, with up to $10 of model usage included so the agent works from the first minute without an API key errand; the $59 and $99 tiers are the identical pod with $40 and $80 of usage, and bringing your own key or a ChatGPT subscription adds nothing on top. For a squad that runs on your own plan, the pod is mostly the computer, and $29 is what the computer plus its care costs. What it is not is a board or a roster product: OpenClaw runs as the project ships it, multi-agent features included, and the mission-control layer is yours to choose.
How to read the four bills
- Whose plan pays for the model? Grok Bot: the plan’s allowance, then metered. Squad: yours, entirely. Self-hosted and pod: yours, or an included allowance you can exceed on your own key.
- Is the computer a line item or hidden in the fee? Hidden in Grok Bot and Squad, explicit for a VPS, the fee itself for a pod.
- What is per seat, per agent, per account? Grok Bot’s Teams pricing is per seat; Squad’s inboxes are per agent; everything else here is per computer.
- What does overage look like? Metered tokens with no Grok Bot-specific cap (per xAI’s docs); a second stacked plan (Squad); your API bill (self-hosted); a hard cap you raise on purpose (a pod’s included usage).
Prices above were read from the vendors’ own pages on September 5, 2026, except where the prose says otherwise. In this category they have changed monthly; check the page before you decide.