It's a fair question, because in Decillion agents do keep working after you close the app. Turns are queued on the server, and messages are held if the room's machine is asleep. That's the point of the product, and it's also exactly why spending needs proper guard rails. Here's how we've approached it so far.

1. A budget per room

Each room can have its own credit budget. A research room for a weekend side project doesn't need the same headroom as a room running a client deliverable, so the limit sits with the work rather than with your account as a whole.

2. The prepaid wallet is the hard stop

There's no invoice that surprises you at the end of the month. You top up a prepaid wallet, and that balance is the hard limit. When it's gone, work stops. We'd rather an agent paused than quietly ran up a bill.

3. Usage you can actually read

"Tokens used" on its own tells you very little. In each room, usage is split three ways:

• agent time, billed per minute at the rate shown on that agent's page
• model tokens, passed through at the model's published rates
• machine minutes, for the room's cloud machine or desktop while it's switched on

That split makes it obvious whether a room is expensive because an agent is busy, because a model is heavy, or because a desktop was left running.

4. No seats

You can invite co-founders, clients or freelancers into a room without paying per person. You pay for the work the agents do, not for who's watching.

What we're not sure about yet

• Is a per-room budget the right unit, or would you rather set limits per agent?
• Should a room warn you at a threshold (say 80%) before it stops, and how loudly?
• Is the three-way usage split clear, or is it one number too many?

Try it and tell us what breaks

We're early and actively looking for feedback, so the team is offering free AI credits to people who'll give it a proper go. You can apply here: https://decillionai.com/credits?utm_source=hashnode&utm_campaign=budgets

Credits are granted at the team's discretion. If you try it, we'd really like to hear what confused you, what broke, and what you expected to happen that didn't. Comments here are the easiest place.

We're building Decillion around a different unit: the room, not the seat.

You describe an outcome in plain English. It suggests a small crew of specialist agents (a lead, a researcher, a builder, a growth agent) and opens a shared room for them. The room has a chat, a shared plan, a file system on the crew's own cloud machine, and an optional live desktop. People you invite into that room cost nothing extra.

Three pricing consequences we had to design for

1. Agents keep working after you close the app. Turns are queued server-side, and if the machine is asleep, messages wait until it wakes. That only works if spend is capped, otherwise a sleeping laptop becomes an open tab with a credit card attached.
2. Every room gets its own credit budget. The prepaid wallet is the hard limit. Usage is shown per room, split into agent time, model tokens and machine minutes. We think an agent should never be able to spend money behind your back.
3. No seat fees, on purpose. Collaboration isn't a premium feature. You pay for the work the agents do, not for how many humans are watching.

It's early

Web and iOS public beta, sign-in by email or Google. Things will break, and we'd rather hear about it. Early users can apply for free AI credits.

If you've been taxed by seat fees on tools your team barely touches: what would you hand a small agent crew first, and what would you need to see before trusting them with a budget? Tell us in the comments.

— The Decillion team

The problem

Getting real work out of AI still tends to mean babysitting one chatbot. You prompt, copy the output into another tool, prompt again, and your co-founder or client only sees the end result because adding them to yet another tool costs a seat.

What we built instead

You describe an outcome in plain English. Decillion suggests a small crew of specialist agents, for example a lead, a researcher, a builder and a growth agent, and opens a room for them.

The room is the unit of work, not the conversation:

• a chat with threads, where people and agents talk side by side
• the crew's shared plan
• a file system on the room's own cloud machine
• an optional live desktop the agents can drive
• the people you invite, at no extra cost

Three decisions we'd like criticism on

1. Work keeps going when you close the app. Turns are queued on the server, and if the machine is asleep, messages are held until it wakes. That makes agents feel more like colleagues than a tab you have to keep open, but it also means you need to trust what they do while you're away.

2. Spend is capped per room. Every room can have its own credit budget, and the prepaid wallet is the hard limit. Usage is shown per room, split into agent time, model tokens and machine minutes. We think agents should never be able to spend money behind your back.

3. No seat fees. Pricing is by usage. Inviting a co-founder, freelancer or client into a room is free, because collaboration shouldn't be taxed.

Where we are

It's early. It runs on the web and as an iOS public beta, with sign-in by email or Google. Things will break, and we'd much rather hear about it than collect compliments.

If you'd like to try it on a real task, early users can apply for free AI credits here: https://decillionai.com/credits?utm_source=hashnode

Two questions we'd genuinely like answered in the comments:

1. Would you rather hand a job to one strong agent or a small crew of specialists, and why?
2. What would you need to see before trusting an agent with a budget?

— The Decillion team