For One Person
For a TeamArchie vs an agent on a cloud computer.
With Archie, your agent works on your own computer. Compared here with Dots, Cowork, Grok Bot, Symphony, Muse and Cue.
Wix and Manus do not say where their agents runSource 8Source 17.
Archie is in beta. Sources checked August 20 to September 30, 2026.
Your computer is the one you can switch off.
Your agent, accounts and approvals stay with you.
Your agent is yours. What it is made of, what it remembers and what you built for it live on your computer, with account tokens in its password storeSource 5. Your words go to your chosen AI company by default. You can save all of it to one file and restore it there.
If you stop paying, nothing on your computer is deleted, and a personal agent keeps working at 20 jobs a day on an AI account of your ownSource 6. If we close, the Terms oblige us to publish a final version of Archie that needs no license checkSource 7.
Anthropic, OpenAI, Google, Groq, xAI, DeepSeek or Mistral, on your own account, switchable any timeSource 5.
An email reply waits until you send it or set a time. A calendar change waits for approval in a later message. Neither gate can be turned offSource 5. Read each before you approve.
A cloud computer keeps working while yours is off.
Dots keep working when your computer is offSource 18. A Grok Bot runs on a persistent cloud computerSource 3. Archie needs your computer awake, though a timed routine can wake it once set up.
What each plan and each piece of work costs.
Archie is free at 20 jobs a day on your own AI account, or $30 a month without the limitSource 6. AI usage is extra: up to $5 a month for light use, more for heavy useSource 6.
Our free starting credits pass through our server on the way to Anthropic while we are payingSource 5. Email features need an AI account of your own.
Compare accounts, approvals, availability and plan prices
| Archie | Dots, Cowork, Grok Bot, Symphony, Muse and Cue | |
|---|---|---|
| The computer it works on | Yours, already paid for | Theirs: a dot’s “own cloud computer”Source 18, Anthropic’s serversSource 2 and a cloud computer per Bot at xAISource 3. Wix does not say where Symphony’s agents runSource 8. Muse runs on “its own dedicated computer in the cloud”Source 16. Each Cue agent has “its own computer”, and Manus does not say where it runsSource 17 |
| While your computer is off | Nothing runs. Archie can wake a sleeping computer for a routine set for a time | Dots, a Grok Bot and Symphony’s agents keep workingSource 18Source 3Source 8, which is what their computer buys you |
| Signed into your accounts from | Your computer, with the tokens in its password storeSource 5 | Their computer: a Bot signs into your tools on its own cloud computerSource 3. Dots sign in to websites in a cloud browser that “has its own sessions, separate from the browser on your computer”Source 18. A Cue agent has “its own email, phone number, wallet, and computer”Source 17 |
| Where your working files and sessions live | On your computer. No server of ours holds themSource 5 | On their side: Cowork saves sessions and files to your Claude accountSource 2; Muse keeps a person’s conversations and data on the cloud computer Meta runs for themSource 16. A dot works on files on its own cloud computerSource 18 |
| Where you talk to it | In Archie, or in Telegram, Discord, Slack, Matrix, or iMessage on a MacSource 5 | In their apps: Cowork in the Claude appsSource 2, a Bot on desktop and iOSSource 3, Muse on iOS, Android and muse.ai, in the USSource 16. Dots answer in ChatGPT, Slack, Teams or a callSource 18. Cue is on the web, desktop and mobile, with iOS after App Store reviewSource 17 |
| Which AI answers | Your pick of 7 companies, on your own account, switchableSource 5 | The company whose agent it is. Symphony is “powered by Wix AI”Source 8, with no way offered to pick another, and the Manus pages we read name no model for CueSource 17 |
| Approval before it acts for you | Email and calendar changes are held until you approve, and the gate cannot be turned offSource 5 | Cowork, a Bot, Symphony, Muse and Dots ask by default. Meta says “By default, Muse will not take many important actions, like sending an email, without your approval”Source 16. Cowork also offers a Skip mode that checks nothingSource 1. One Dots rule is “Take action without asking”Source 18. Manus’s pages describe no approval step for Cue, which is not the same as there being noneSource 17 |
| Where you see what the AI costs | An estimate in the app, counted from each reply. The real bill is on the AI company’s own dashboardSource 6 | Cowork and a Bot work inside their tier’s limits, and Dots inside an allowance “for deeper work” that the plan includesSource 18. Symphony spends credits at a cost “calculated after each action, based on its complexity and the tool used”Source 9 |
| What you pay | Free at 20 jobs a day on an AI account of your own, or $30Source 6 a month for the software without the limit. AI usage is at cost either way: up to $5Source 6 a month for a handful of tasks a day | Dots come with ChatGPT Pro, from $100Source 18 a month, and Cowork with a Claude plan, from $20Source 1. A Grok Bot is from $200Source 4 with Cursor Ultra or $300Source 4 with SuperGrok Heavy. Symphony is free, then $16Source 9, $40Source 9 or $80Source 9 a month. Muse is free, with paid plans Meta’s page does not priceSource 16 |
Prices checked August 20, 2026; Symphony September 11; Cue September 29; Dots September 30.
Cue has no price yet: it is free in early access, with an invite codeSource 17.
Check the AI cost per reply, credit or action
| Agent | Their unit | What it costs |
|---|---|---|
| Archie | One reply from your agent, with every pass it takes inside it | $0.01 when it follows another closely, up to $0.11 when it starts from nothing, measured on an agent with twelve add-ons and everything connected, at BalancedSource 6. A reply that hands work to a specialist or uses the Deep setting costs more. Billed by the AI company to you |
| Lindy | A credit. An “everyday ask” is 2 to 250 of them, “deep work” 250 to 1,000 | $29.99 a month for 3,000 credits, so an everyday ask is $0.02 to $2.50 and deep work $2.50 to $10Source 10 |
| Zapier Agents | An activity: one action, one search, or one question answered from a knowledge source | $400 a year for 1,500 activities a month, about $0.02 eachSource 11 |
| Microsoft Copilot Studio | A credit. An answer is 2 of them, an action 5 | $200 a month for 25,000 credits, or $0.01 each pay as you go, so a reply that takes two actions is about $0.10 to $0.12Source 12 |
| Salesforce Agentforce | A conversation, or an action paid in Flex Credits | $2 a conversation, or $0.10 an actionSource 13 |
| Claude Code, an agent on a key like ours | A developer’s working day | About $13 an active day and $150 to $250 a month, on Anthropic’s own figuresSource 14 |
Each row is in its own unit: a Lindy credit, a Copilot credit and a reply are different sizes of work, and the companies say so. Ours is a whole reply; most of theirs is one step.
Claude Code is a coding agent and works far harder than an assistant; it is here for the shape of the bill, an agent on your own key. One person’s published three-month log of OpenClaw, another agent on a key, came to $187, $94 and $78 a month as its owner moved to cheaper modelsSource 15.
Their prices checked September 12, 2026; Archie measured in August and priced September 12.
Sources for every figure and claim on this page
claude.com/pricing, read August 20, 2026: “Includes Claude Cowork” on the Pro card, $20 a month or $17 billed annually. The permission modes, including Skip (“Claude doesn’t pause to ask and nothing checks its actions automatically”), are in Anthropic’s own guide.
“Claude’s work runs on Anthropic’s servers, in an isolated environment, and your sessions and files are saved to your Claude account”. From Anthropic’s getting-started guide, read August 20, 2026, which lists the Claude desktop, web and mobile apps.
In beta on desktop and iOS. Each Bot “runs on a persistent cloud VM with a browser, filesystem, and terminal”, signs in to your tools there, and comes back “when something needs your approval”. From xAI’s announcement and its documentation, read August 20, 2026.
SuperGrok Heavy at $300 a month and Cursor Ultra at $200 a month, from the Grok Bot plan cards, read in a browser and transcribed on August 20, 2026. Those pages refuse automated requests, so no script of ours can re-check them. The same cards list Cursor Teams Premium at $120 a seat.
The custody claim, the password-store claim, the 7 providers, the 5 chat apps and the approval gate are all on our Trust page. It lists everything Archie sends and to whom, and shows how to watch it yourself with a network monitor.
Our own price, and our own measured figure for light AI use (a handful of tasks a day), both broken down on the pricing page. Heavier use costs more. The app keeps an estimate; the AI company’s own dashboard has the real bill.
Our Terms require us to publish a version that needs no license check if we cease operating. The wording is in the Terms of Service and the reasoning is on the Trust page.
An AI agent platform for small businesses with a “mobile-first interface”. “Symphony is available on iOS and Android. Download the app and take your business with you wherever you go”, and it runs in a browser too; you use it with a Wix account, though it needs no Wix website.
It is “powered by Wix AI and built on 20 years of real business expertise”; the page names no model and offers no model choice. Wix does not publish where the agents run, so we say only how you reach it. From wix.com/symphony and the Symphony overview, read September 11, 2026.
Free, then Basic at $16 a month ($12.80 billed annually), Pro at $40 ($32) and Max at $80 ($64). The free plan needs no credit card and carries 500 monthly credits and 100 daily credits; credits “reset monthly and do not roll over”. Basic carries 2,000 monthly credits and Pro 5,000; Max publishes no figure.
The unit, in Wix’s words: “Each action you take with agents and tools consumes AI credits. The exact amount is calculated after each action, based on its complexity and the tool used.” From wix.com/symphony/pricing, read September 11, 2026.
Plus at $29.99 a month per user with “3k credits / user / mo”, and the ranges “Everyday Asks 2-250 Credits” and “Deep work 250-1,000 Credits”, from lindy.ai/pricing, read September 12, 2026. The per-ask dollars are our arithmetic: $29.99 over 3,000 credits is a cent a credit.
“Agents Pro $400 billed annually” with “up to 1,500 times per month”, from zapier.com/l/agents-pricing, read September 12, 2026. Zapier’s help article counts a trigger, an action, a search, or a question answered from a knowledge source as one activity each. $400 over twelve months of 1,500 is 2.2 cents, our arithmetic.
“1 capacity pack = 25,000 Copilot Credits” at “$200 per credit pack/month (billed annually)” from Microsoft’s pricing page. The rate table, generative answer 2 credits and agent action 5, and the pay-as-you-go rate of $0.01 a credit, are from Microsoft Learn; both read September 12, 2026. A reply that takes two actions is 12 credits, our arithmetic: $0.096 prepaid, $0.12 metered.
“$2 USD per conversation” and “$500 USD per 100,000 Credits” from salesforce.com/agentforce/pricing, and “One Agentforce action consumes 20 Flex Credits ($0.10 USD)” from Salesforce’s Flex Credits announcement, read September 12, 2026.
“Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month”, from code.claude.com/docs/en/costs, read September 12, 2026. A coding agent, on the page for the shape of the bill rather than the size of the job.
A self-hosted OpenClaw agent tracked by its owner: “Month one: $187” on Claude Opus, $94 in month two after moving to Sonnet, and $78 in month three. From a dev.to post of March 30, 2026, read September 12, 2026. It is one person’s log, the only kind of figure that exists for an open-source runner with no price of its own.
Meta’s personal AI agent, announced September 8, 2026. It “runs on its own dedicated computer in the cloud, contained so no one else’s agent can reach it”. It is “rolling out in the US on iOS, Android, and muse.ai, and coming soon to AI glasses”.
On approval, Meta’s help center says “By default, Muse will not take many important actions, like sending an email, without your approval”. On privacy, Muse “doesn’t share a person’s conversations or the data in their VM with Meta’s ad systems”, and people “can opt out of their interactions being used to train Meta’s AI models”.
On price, only that Muse is “free for most of what people need, with subscription plans for people who want to do more”. Meta publishes no figure we could read, so we print none. From Meta’s announcement, read September 15, 2026.
Cue is “a new standalone app for your personal agents”, “built on the same infrastructure as Manus”. Each agent has “its own email, phone number, wallet, and computer, so it can send messages, pay within the budget you set, and see a task through on its own machine”. It “can also take your calls and leave you a summary in Cue”.
It is “in early access and free to use with an invite code”, on the web, desktop and mobile, with iOS to follow after App Store review. Of agents working together in a group chat, Manus says “You set the direction and make the final call.”
Manus does not say where an agent’s computer runs, and describes no approval step and no model for Cue, so we say none. From the Manus 2.0 announcement of September 28 and its help center, read September 29, 2026.
OpenAI’s “always-on agents”, announced September 29, 2026 and “Powered by GPT-6 Astra”. “Each dot works on its own cloud computer, while your computer and its contents stay separate unless you choose to connect it.” Connecting yours “is optional and starts turned off”. A dot “can keep working even when your computer is off”, and you reach it “in ChatGPT, Slack, Teams, or a call”.
On approval: “Dots start with built-in rules for when to act independently and when to ask for approval. Custom Rules let you allow specific actions, require approval, or block them.” One of those rules is “Take action without asking”. On training, OpenAI says it does not use content from Business, Enterprise or Edu workspaces to improve its models by default, and that personal plans can control it.
On price: “Your first dot is included in your Pro or Business Premium plan at no extra cost”. ChatGPT’s Pro card reads “From $100 / month” and lists “Dot, your always-on agent”. Pro users get it outside the European Economic Area, Switzerland and the UK. The plan “includes an allowance for deeper work”; OpenAI publishes no price per task.
From OpenAI’s announcement, ChatGPT’s guide to dots and its controls page, OpenAI’s help center and ChatGPT’s pricing page, read September 30, 2026. Three of those refuse scripts, so we read them through a service that returns a page’s own text.
Nothing here says these products are careless with your content: each company documents its own isolation, defaults and approvals, and we quote them. The claim is only about where the work happens: on a computer their company runs for four of them, somewhere Wix does not say for Symphony, and on yours for Archie.
Nothing here says Dots act without asking you. OpenAI publishes that they have built-in rules for when to ask, and does not list those rules.
For Cue, Manus says each agent has a computer of its own and does not say where. Nothing here says Cue acts without asking you, because Manus’s pages do not say either way.
Nothing here says Symphony trains on one customer’s data to answer another: we quote what Wix says its AI is drawn from, and nothing more. Each company’s own security documentation is the place to judge their side properly.