AtmosphereAGI
Product·the AtmosphereAGI team

Getting started with AtmosphereAGI: from sign-up to your first finished task

The AtmosphereAGI task page showing an intent bar, a live task card, and downloadable output files

What you are signing up for

AtmosphereAGI is a managed AI agent platform with a deliberately simple shape: you type what you want in a single intent bar, a live task card shows the agent working, and when the run finishes you download the files it produced. If you have seen Manus, the UX will feel familiar — that is intentional. What is different is the billing: tasks are billed at Anthropic's sticker rate, passed through with zero markup. This guide walks you from a fresh account to your first finished task.

Step 1: Create an account

Sign-up takes under a minute. Once you are in, you will land on the task page: an intent bar at the top and your task history below it. There is no onboarding maze — the product is the intent bar, and everything else supports it.

Step 2: Choose how you want to pay

There are two lanes, and you can pick either on day one.

Prepaid credits

The starter pack is $25, and you can top up in $25, $50, or $100 increments. Credits convert 1:1 — $25 buys $25 of usage — and they never expire. There is no subscription and no monthly minimum: buy a pack, run tasks against it, top up when you choose. Because pricing is pass-through, your spend maps directly to Anthropic's published rates, and each step of a run is line-itemed at the same rate Anthropic would charge you directly.

Bring your own Anthropic key

If you already have an Anthropic API key, you can plug it in instead. Your key is encrypted at rest with AES-256-GCM, only the last four characters are ever shown back to you, and deletion is immediate and irreversible. In this lane you pay Anthropic directly for inference, and AtmosphereAGI a small flat fee for the runtime. This is the natural choice if you already manage an Anthropic account and want your agent spend consolidated there.

Step 3: Run your first task

Pick something real but low-stakes — a research summary, a drafted email, a small document. Write the instruction the way you would brief a capable assistant: state the goal, name the inputs, describe the output you want. A brief like "research the top five project management tools for a three-person design studio and give me a one-page comparison with a recommendation" will outperform "research project management tools" every time, because the agent knows what done looks like.

When you submit, the task card goes live. You can watch the run in progress and see the spend counter tick in real time — there is no mystery about what a task is costing while it is happening. If you would rather not watch, you do not have to; the run continues either way.

Step 4: Collect your deliverables

The end state of a task is not a wall of chat to scroll back through. It is files: the report, the spreadsheet, the draft, whatever the task called for, ready to download and use. This "files as deliverables" principle is the core of the product — a task is finished when you have something you can attach to an email, open in a spreadsheet app, or hand to a colleague. Finished runs stay in your history, so you can revisit or replay them, and public share links let you send a finished run to someone who does not have an account.

What to try in your first week

The platform ships with eight seed agents covering common shapes of work: email, research, slides, a web worker, voice notes, logo, code, and numbers. The email agent is free forever, which makes it a zero-risk first stop. A reasonable first-week arc: day one, one email draft and one research task; midweek, a document or slide task with a source file; end of week, try a scheduled task for something you want recurring. By then you will have calibrated two things — how to brief the agent, and what a typical task costs you — and both are easier learned by doing than by reading.

Understanding your costs

Because billing is pass-through, the mental model is simple: a task costs what its Anthropic inference costs. Sonnet 4.6 input runs $3 per million tokens and output $15 per million, and your invoice line-items each step at those rates. The live spend counter during a run and the per-step line items after it mean you are never guessing. Combined with credits that never expire, there is no pressure to "use up" anything before a renewal date — there is no renewal date.

Common first-timer questions

Do my credits expire? No. Prepaid credits are 1:1 and never expire.

Is there a subscription? No. The starter pack is $25 and top-ups are $25, $50, or $100 whenever you want them.

Can you see my Anthropic key? It is stored encrypted with AES-256-GCM; only the last four characters are ever displayed, and you can delete it at any time, effective immediately.

What if I want to stop?Delete your key if you added one, and your remaining credits simply sit there — they never expire, so nothing is lost if you come back later.

Sign up, run one small task, and judge the product by the file it hands you at the end. That is the whole pitch.