Eight agents with names, or one employee with receipts.
Atoms (by the MetaGPT team, formerly MGX) gives you a cast of eight named AI specialists and a workspace to build apps in. We give you one employee that builds a business and writes down every single thing it does. Both are real products; the difference is what you can verify.
Their pricing and claims as read on atoms.dev on 11 August 2026. If they change, tell us and we will update this page.
The shape of the two products
On Atoms you type a prompt and a team of mascots — a researcher, a product manager, an architect, an engineer, an SEO specialist, an ads specialist, a data analyst and a team leader — assembles an app in a chat workspace, and you drive it from there. Here you describe a business once, a real site goes live on its own address in about three minutes, and from then on one employee works a scheduled shift on it and files a report. Theirs is a workshop you operate. Ours is a worker you employ.
| Atoms | First Employee | |
|---|---|---|
| Price | $20/mo Pro, $100/mo Max | $19 — one plan |
| What is metered | Credits — 100/mo on Pro, 500/mo on Max; buy more when they run out | Nothing to top up mid-task: a plan is a monthly allowance of shifts, and a shift is a unit of finished work |
| Share of your revenue | None | None on self-managed; 15% only if you choose the managed service, in settings |
| The free tier | 15 credits a day in their workspace; public projects carry the Atoms badge | The whole founding build — researched, named, written, published, verified live — and the site stays up, free, forever |
| Proof of what the AI did | Testimonials, star counts and papers — no public per-build audit trail | An append-only receipt ledger, written as each action runs, and a homepage that replays a real build from its own receipts |
| What 'done' means | An app in your workspace, published when you press Publish | A company is only marked active after we fetch its public pages over the open internet and they answer |
Their numbers are from atoms.dev/pricing with the yearly toggle off. Credits are their unit of AI work; how much one credit buys is their definition and can change.
The part we think matters most: verification
Atoms' trust section leans on scale — countries, community size, GitHub stars, academic papers, a Product Hunt badge. All of it may be true, and none of it tells you what their agents actually did for any one customer.
We built the product the other way up. Every action our employee takes is one line in an append-only ledger — the feed you watch during the build IS the ledger, so the story and the record cannot disagree. The case study on our site is rendered from those rows at request time, not written by a marketer. When we say the build works, we mean our server fetched the finished site over the public internet and looked at it.
In fairness to them
If what you want is to build software — a SaaS app, an internal tool, a game — Atoms is genuinely good at that, and parts of it are ahead of anything we offer: full-stack code generation, a hosted backend, code export and GitHub sync, a visual editor, and a mode that runs your prompt across several models at once. They grew out of MetaGPT, a widely used open-source framework, and the engineering pedigree is real.
We do not give you an IDE, and our employee will not build you an arbitrary app. It builds and works on a business: the site, the pages, the copy, the leads, the follow-through. If you want a code workspace, use theirs. If you want to look up from your own job and find the work done and accounted for, that is what ours is for.
Three differences that do not show up in a feature table
- An employee is not a chat. You do not drive it prompt by prompt; it works its shift on its own and reports, and the standing instructions you give it persist.
- An allowance is not a meter. A shift either happens or it does not — you are never mid-task wondering what a credit is worth or watching a balance drain.
- A ledger is not a feed. Ours cannot be edited after the fact, shows a line only for work that ran, and is the same data your dashboard, our homepage and our case study all render.