Friday refresh for this pillar: plan numbers now match live pricing (Free → Max), we added a wallet-vs-overage section (the $5 pack is a purchase, not a silent meter), and we tightened bridges to CloudAxis for product overview depth — signup economics stay here.
Pay-per-use rewards infrastructure. It punishes exploration.
OpenAI, Anthropic, and Google sell tokens by the fraction of a cent. That model is correct for production APIs with dashboards and ops owners. It is a bad default for a founder who wants an assistant that researches, browses, and schedules without a finance review every Thursday.
Priya’s $187 bill was not a bug in Stripe. It was the design working as advertised: keep running, keep charging. The second cost never shows on the invoice. She spent the next week asking shorter questions. Three follow-ups she would have sent in March never left the draft box. That is metered anxiety — documented in cloud storage and mobile data for years, now showing up in AI assistants.
Horror stories in builder communities follow the same shape: a forgotten overnight loop, a teammate with raw API keys, a document reprocessed ten times. For consumer and SMB assistants, “you will pay what you used” is not a feature. It is a trust tax.
Hard caps, soft caps, and metered — what actually happens at the limit
A hard cap pauses the feature when the allowance hits zero. No automatic overage invoice. A soft cap warns, then keeps going with extra charges or throttles. Pay-per-use never stops for your budget — only for your card.
CloudyBot hard-caps AI credits and cloud browser minutes separately. Live self-serve numbers (verify on Pricing before you buy): Free is $0 with 50 AI credits at signup plus earn bonuses (~100 day-one possible) and 10 browser minutes; Growth $29 / 2,500 credits / 500 browser minutes; Pro $59 / 5,000 / 1,000; Max $199 / 18,000 / 3,500. Paid plans bill immediately. Free needs no card. When a meter is empty, that feature stops until you upgrade, buy an explicit top-up, or the period resets.
| Model | Surprise-bill risk | At the limit | Best fit |
|---|---|---|---|
| Pay-per-use | High — swings map to invoices | Keeps running; cost rises | Forecastable workloads with someone watching spend |
| Soft cap | Medium — overage lane after warnings | Often continues with charges or throttles | Teams that want elasticity if finance signs off |
| Hard cap (CloudyBot) | Low — spend bounded by tier + explicit top-ups | Requests pause until upgrade, pack, or reset | SMB assistants where trust beats uncapped throughput |
Three September bills for the same workload
Same fictional team: four people, weekday competitor checks, some research chat, light browser automation. Same rough usage. Different pricing machines.
| Setup | What they planned | What September charged | How it felt |
|---|---|---|---|
| Raw metered API | “About $40–60” | $187 after one runaway weekend job | Trust cracked; prompts got shorter |
| Soft-cap SaaS | $79 seat + “fair use” | $79 + $64 overage line item | Felt unlimited until finance Slack |
| Hard-cap Growth | $29 on day one | $29 (+ optional $5 pack they clicked) | Paused two days before reset; boring invoice |
Insider stance you can disagree with: for a team under twenty people, I would rather hit a ceiling on the 27th than receive a polite overage email on the 3rd. Boring invoices ship more agent work than clever meters. If you are comparing hosted assistants on price alone, start with affordable hosted AI assistants compared and keep the ceiling column visible.
Wallet top-ups are not soft caps in disguise
This is where buyers get confused — and where we got questions after the May version of this post.
CloudyBot lets you buy a $5 starter pack (100 AI credits) into a credit wallet. Referral bonuses land in the same wallet when enabled. Chat draws from the monthly plan bucket first, then the wallet. New chat is refused only when both are empty. That looks like “extra usage,” so people ask: is that not soft-cap metering?
No. The difference is consent and timing. Soft caps and pay-per-use invent a charge after the work already ran. A wallet pack is an explicit purchase before or during the period — same move as buying another coffee punch card, not the same move as an uncapped token runaway. Browser minutes still hard-stop. WhatsApp and other meters pause on their own counters. Details stay on Pricing; cost modeling helpers live at the AI cost calculator and token counter.
Friction we will own: people still hit caps mid-deadline when they ignore the dashboard until day 28. Mitigations that work — budget long research like calendar blocks, watch both AI and browser meters (browser-heavy duties are a different burn rate; see cloud browser automation), and treat two consecutive months at the ceiling as a signal to move Free → Growth or Growth → Pro, not as a moral failure.
Why we still pick hard caps (and when we would not)
Hard caps are frustrating when you are mid-thread and the ceiling arrives. We have heard that. The design bet is which failure mode you can survive: transparent blocking, or a bill that rewires how you use the product forever.
Hard caps win when:
- You want a number you can put in a spreadsheet on day one of the month
- More than one person can trigger spend (shared assistant, scheduled duties)
- Always-on or overnight work is in play — meters punish sleep; ceilings do not. Pair with the scheduling intent post why your AI agent should work while you sleep; for always-on OS framing, bridge to CloudAxis on always-on agents that run while you sleep.
Hard caps lose when you are a power user past Max’s published self-serve envelope every month and you have finance tooling for raw APIs. We are honest about that. Pay-per-use with a firm budget alarm can fit infrastructure teams better. Self-hosting shifts the cost into ops tax instead of invoices — fair comparison in hosted vs self-hosted and the product overview on CloudAxis: CloudyBot on CloudAxis.
Predictable pricing is also why scheduled competitor and SEO duties make sense on a Free or Growth ceiling instead of an open meter. If chat-only tools are enough for your team, stay there — the ops fork is spelled out on our ChatGPT alternative page. For how duties, browser, and files fit together without the OS deep dive, use how it works.
FAQ
What is the difference between hard caps and metered overages?
Metered billing charges for each increment of usage and can produce unexpected invoices. Hard caps pause usage at the published plan allowance so monthly subscription cost stays predictable unless you choose to upgrade or buy a pack.
Does a credit wallet mean CloudyBot has soft caps?
No. Wallet credits are prepaid units you buy or earn. Soft caps invent post-hoc charges after work runs. Chat stops only when monthly credits and wallet are both empty — and browser minutes still hard-stop separately.
Why meter browser time separately from AI credits?
Live browsers have a different cost profile than tokens. Separating the meters keeps each ceiling honest. Check current allowances on Pricing.
What should we do if we hit a cap during a deadline?
Upgrade, buy an explicit credit pack, or wait for the billing-period reset. Split giant jobs across sessions earlier in the month so the dashboard is not a fire drill on day 29.
Where do I compare plan ceilings side by side?
On Pricing, plus the companion posts predictable AI pricing for teams and how pricing caps prevent bill surprises.
Further reading
- CloudyBot pricing — live AI credit + browser caps
- Predictable AI pricing for teams
- How pricing caps prevent AI bill surprises
- Affordable hosted AI assistants compared
- Cloud browser automation
- Why your AI agent should work while you sleep
- Can AI replace Zapier? We tested it — hybrid glue vs agent cost tradeoffs
- AI cost calculator
- How CloudyBot works
- CloudyBot product overview (CloudAxis)
- Always-on agents while you sleep (CloudAxis)
Know the ceiling on day one. Free to start — hard caps, real browser, no surprise overages.
Get started free →No card on Free · Growth from $29/mo · Paid plans bill immediately and pause at the limit