Every AI agent product puts a meter somewhere. The sticker prices look alike, $20, $100 and $200 a month keep coming up, and they tell you little. What decides the real cost is the unit you are metered in, what happens when it runs out, whether a team shares it, and whether you can bring your own model or login. This AI agent pricing comparison sorts twelve products by the shape of their meter, every price as of 19 October 2026; where a number is unpublished, we say so.
We build DigiSapiens, which takes the "your own login" shape, so it gets a short section near the end, tradeoffs included.
The market at a glance
| Product | Entry price with the agent | Metered in | When it runs out | Your own model or login |
|---|---|---|---|---|
| OpenAI dots | ChatGPT Pro, $100; Business Premium, $100 a seat | Plan allowance; terms after month one unpublished | May be unavailable until reset | No, GPT-6 Astra only |
| Grok Bot | The X Subscription Premium, $30 | One weekly pool shared with Cursor and the Grok apps | Wait, top up, or upgrade | No picker; other vendors' models at API list price |
| Meta Muse and Muse Code | Muse free with a limit; Muse Code $5 | Muse tokens a week; Muse Code messages per five hours | Upgrade or wait | No; Meta login or key |
| Manus | Free (Lite); Pro from $20 | Credits by task, cleared monthly | Add-on credits or upgrade | Not documented |
| Claude Code | Pro $20 | Five-hour window plus weekly limits, shared with Claude chat | Wait, upgrade, or usage credits at API rates | Claude only; API key instead |
| Codex | Free (limited); Go $8; Plus $20 | Messages per five hours plus weekly limits, shared with ChatGPT Work | Turn finishes; credits or API key | API key; plan works in partner apps |
| Hermes Agent | Free; optional Portal from $20 | Nothing; Portal dollar credits | Portal terms unpublished | Yes; Claude only via Max extra credits |
| OpenClaw | Free | Nothing | Your provider's limits | Yes, including a Claude Code login |
| Velozity | Free tiers; Co-Worker Pro $15 a user | Tasks a week; credits for its built-in agent | Finishes current work, then pauses | Yes: Claude Code, Codex, other CLIs |
| Organ | $39 with $15 of credit | One allowance for runtime and tokens | Overage at cost plus 15 percent | API keys; runtime still metered |
| SquadRoom | $49, early access | Web searches and LinkedIn lookups only | Buy more: 5,000 credits for $10, never expire | ChatGPT account or API key |
| DigiSapiens | Closed beta; price unpublished | Nothing | Your provider's limits | Yes: Claude Code, Codex, other CLIs, any API key |
Allowances: a window that refills
The most common meter is a window that refills on a clock, and the first thing to learn about it is what else drinks from it.
Claude Code "shares the same usage limits as the rest of your plan": a rolling five-hour window, weekly limits on top, one pool across Claude chat and Claude Code. Pro is $20 a month, Max from $100. There is "no fixed message count" and the absolute size is not published. At the limit you wait, move up, or turn on usage credits at standard API rates up to a spend limit you set.
Codex is metered in local messages per five-hour period, with published ranges, 15 to 160 messages of GPT-6.1 Sol on Plus, and "weekly limits may also apply" with no figure. "ChatGPT Work and Codex share usage", so an afternoon of documents eats the evening's coding. At the limit the current turn finishes; then Plus and Pro users buy credits, valid 12 months, or use an API key.
Grok Bot comes with The X Subscription from Premium at $30 a month; the $8 Plus plan excludes it. Each plan has one weekly pool shared across "Agent in the editor, CLI, Cloud Agents, Grok Bot, and the Grok apps", sized only as 4x, 15x or 50x Plus. What the sticker hides is that Claude, GPT and Gemini are not in the pool at all. On Premium and above they are billed "at the model's API list price from the first request", so the plan buys Grok and Cursor, and other vendors' models cost API money on top.
Meta's Muse is free with an unpublished limit; Power ($20) gives 500M Muse tokens a week, Maximum ($100) 3B. Muse Code runs $5 to $200, the $5 plan at 10 to 50 messages every five hours, with no overage: on the top plan "you must wait for your limit to refresh". Muse Code also defaults to Contributor mode, "more usage from your plan in exchange for letting Meta use your content to train future Meta models".
OpenAI dots needs ChatGPT Pro, from $100 a month, or a Business Premium seat at $100 billed annually ($125 monthly) for teams of 2 to 200; the first dot is included. On 29 September OpenAI wrote that dots usage "won't count toward" plan allowances "for the next month", with usage terms to follow. As of 19 October they are not published, nor is the price of a second dot. What is published: at a model's limit it "may be unavailable until its allowance resets". On 6 and 9 October OpenAI Support told a user it could not confirm how long a dot's cooldown lasts.
Credits: a balance that drains
Credits replace the clock with a balance; the questions become what one unit buys and what happens at zero.
Manus Pro starts at $20 a month and Team at $20 a seat with 4,000 credits per seat per month; the Pro amount is not on a page we could open. Credits burn by "complexity and duration"; plan credits clear each month, purchased add-ons never expire. An action can cost more than you have left, so the balance "may temporarily show a negative value".
The Nous Portal, optional for Hermes Agent, sells dollars: $20 buys $22 of monthly credit, $100 buys $110, $200 buys $220, with rollover caps of $10, $50 and $100. What happens when the credit is gone is not on the plan page.
Organ bundles credit into the plan, $39 with $15 up to $299 with $200, as "one allowance for runtime and tokens", overage at cost plus 15 percent. With your own key you pay the provider for tokens, but "agent runtime still uses your allowance".
SquadRoom, $49 a month in early access, meters the narrowest thing here: a web search costs one credit, a LinkedIn profile ten, 1,000 included a month. Model usage runs on your own ChatGPT account or API key with "no markup". Past 1,000 you buy more: "5,000 credits for $10, and bought credits never expire".
Velozity's Co-Worker uses credits for its built-in agent, Speed: 150 free, 700 a month on Pro at $15 a user. At a limit it finishes "whatever's already in progress, then pause[s] just the next action".
Seats: what a team actually shares
The thing to check on a seat bundle is whether the included usage is pooled. Mostly it is not.
Claude Team is $20 a seat annually ($25 monthly), or $100 ($125) for a premium seat with 5x the usage, for 2 to 150 people. Usage "draws from each member's seat allowance". The one pooled item, "up to $500 per month in pooled API credits", is for the Claude Platform, not Claude Code.
ChatGPT Business is $20 a seat annually ($25 monthly). Limits are per seat, but purchased credits "are pooled across all users"; an empty pool blocks the feature until an owner adds more.
Cursor's team seats on The X Subscription are $30, $100 and $200 a user. "Seats don't share weekly pools"; additional usage comes from a prepaid balance the team shares.
Manus Team is the exception: credits "are shared across all seats". Velozity answers the pooling question with "Not yet": each Co-Worker Pro seat has its own wallet. The pattern: the overage is pooled, the allowance is not.
Free tools where you pay the model provider
OpenClaw and Hermes Agent have no price of their own. OpenClaw states "No subscription. No hosted tier." and takes hosted, subscription-backed, gateway or local models, including a Claude Code login reused on the same machine. The Hermes provider page is the frankest on the login question: the Claude path "only works if you're on a Claude Max plan and have purchased extra usage credits", the base Max allowance "is not consumed by Hermes", and Pro subscribers cannot use it.
One user put the agent-only cost at "about US$30 a month: two servers and the DeepSeek credit" on 12 October, the big subscriptions being paid anyway. The tradeoff is time: the same writer found OpenClaw "more expensive to set up and maintain".
Your own login, and where DigiSapiens sits
In this shape the tool adds no meter and the model work flows through the Claude Code or Codex plan you already pay for. What it does not change is the plan's limits. Anthropic says Claude Code shares its usage limits with the rest of the plan, and OpenAI says of apps connected to a ChatGPT plan that "connecting an app does not add more usage". An agent that empties your five-hour window at 3pm empties it for you too.
DigiSapiens runs the agents on your own computer, on your own Claude Code or Codex login, another agent CLI, or any model with an API key. Nothing is metered or resold. Digi, the AI Chief of Staff, proposes agents and channels and sets them up after approval; each agent has a name, a role and a memory, works in the same channels as the people, and outward-facing work waits for the owner's approval. ETREXIO, the two-person software studio in Istanbul where the idea started in December 2024, is the first client; the agent bill there is the Claude Code subscription we already pay for, inside that plan's windows.
The tradeoffs: it is a Mac app today, with iPhone and Android on the way. It is a closed beta with a waitlist, free for the first six months for people on the list, and the price after that is not published, which is why DigiSapiens has no number in the table either.
What to ask a pricing page
Seven questions, with the product that shows why each matters.
- What is the unit, and is its size published? Claude Code, Codex and The X Subscription publish no absolute allowance.
- What else draws on the same meter? Claude chat and Claude Code; ChatGPT Work and Codex; Cursor, Grok Bot and the Grok apps.
- What happens mid-task at the limit? Codex finishes the turn; Manus goes negative; Velozity pauses the next action.
- What is the overage path and its price? API rates (Claude), API list price (Cursor), cost plus 15 percent (Organ), credits that expire in 12 months (ChatGPT) or never (Manus).
- Does the team pool the allowance or only the overage? Manus pools the allowance; the others pool the overage.
- What changes with your own key or login? Organ still meters runtime; Hermes needs Max with extra credits for Claude.
- What is still unpublished, and since when? dots usage terms, promised on 29 September, not published as of 19 October.
The cheapest plan on the page is rarely the cheapest month of agent work. The meter is.