How to protect the free tier of a cloud IDE
Cloud IDE free tier abuse holds shared capacity for days and leaves disks behind. The arithmetic, the basic fixes that break, and what to do first.
"Starting your workspace" is the dullest screen in your product, and on a good day a developer sees it for about 11 seconds.
The screenshot in your support inbox shows it at 14 minutes. It's 10:20 on a Tuesday, and it's from a team lead who pays you for 30 seats: "Half my team can't open their environments and we release today. Should I tell them to work locally?"
You open the fleet page. All 20 hosts are full and 80 workspaces are waiting for a slot. So you sort what's running by uptime, and the top of the list is free accounts you've never heard of. Project name test. Running for 57 hours of a 60-hour allowance. No editor connected since it started. The account was created minutes before the workspace, and the rows below it look the same.
That's a made-up Tuesday, so swap in your own numbers, but the arithmetic of cloud IDE free tier abuse works the same on any fleet. Here's what we're assuming:
- A host costs $1.28 an hour and holds 32 workspaces, so a workspace-hour costs $0.04.
- You run 20 hosts, which is 640 slots. A normal weekday peak uses 560: 380 paying developers and 180 free ones.
- The free tier is 60 hours a month and a 10 GB disk at $0.08 per GB a month, kept for 90 days after last use.
A 30-day month has 720 hours, and 720 ÷ 60 = 12, so it takes 12 free accounts to keep one workspace running all month. Now say 8 people each hold 20 slots that way, which is 160 slots:
- 160 slots × 12 accounts = 1,920 new accounts a month, or 8 a day per person
- 160 slots × 720 hours × $0.04 = $4,608 a month in compute
- 1,920 accounts × 10 GB × $0.08 = $1,536 a month in new disks, on top of last month's
Your peak now wants 720 slots (560 + 160) and you have 640, which is where the 80 waiting workspaces came from. Your scheduler treats every request the same. Paying developers are 68% of your real demand (380 ÷ 560), so about 54 of those 80 are people who pay you. One of them works for that team lead, whose 30 seats at $25 are worth $9,000 a year (30 × $25 × 12), and who's deciding today whether to trust you on a release day.
Disks pile up for three months before the first ones expire:
- Month 1: $4,608 + $1,536 = $6,144
- Month 2: $4,608 + $3,072 = $7,680
- Month 3: $4,608 + $4,608 = $9,216
That's $23,040 in a quarter. If 900 seats at $25 bring in $22,500 a month, the month 3 bill is 41% of your revenue, everything 369 seats pay you. If you spend $40,000 a month, you've lost about 17 days of runway. And the 1,920 signups a month you were pleased about never opened a file.
Stopping doesn't stop it. Block every one of those accounts on the last day of month 3 and the compute ends that night. The disks don't, because you told free users their work would be there for 90 days, and you can't yet say which of them it's safe to break that promise to. Month 4 still costs $3,072 and month 5 $1,536.
So that afternoon you're choosing. Five more hosts cost 5 × $1.28 × 720 = $4,608 a month, which means buying capacity for those 8 people to the dollar. You could queue free accounts at peak, and then the 180 real free developers, your next paying teams, get the spinner. Or you could close the free tier that most of your 900 seats started on.
Why cloud IDE free tier abuse holds on so long
An API request is over in milliseconds, and the next one can use the same machine. A workspace keeps its 2 vCPUs and 4 GB until something stops it, and nobody else can have that slot in the meantime.
Your hosts are oversold too, on purpose. Thirty-two workspaces at 2 vCPUs each is 64 vCPUs on a 16-core host, which works because developers spend most of the day reading and typing. A scraper doesn't read. Eight free workspaces running flat out use all 16 cores (8 × 2), and the other 24 developers on that host compete with them for every build.
Free compute attracts work that has nothing to do with coding: a scraper, a chat bot, a game server, a miner. The limit sits on the account, so a new account gets a new meter, wherever you put the limit.
The obvious fix is to give less away. Cut the free tier to 30 hours and one always-on slot needs 24 accounts instead of 12. If those 8 people keep going, compute stays at $4,608 and your disks double:
- 160 slots × 24 accounts = 3,840 new accounts a month
- 3,840 accounts × 10 GB × $0.08 = $3,072 for each month of signups
- Three months of disks: $9,216, up from $4,608
Your real free developers got half the hours, and your bill went up.
Basic fixes for free workspaces, and how each one breaks
Start with signup, where the next account comes through. Gmail ignores dots and plus-tagged mail reaches the same inbox, so one mailbox can be many addresses, and a verified address only proves someone can open an inbox.
| Basic fix | What it catches | What it misses | Who it wrongly blocks | Upkeep |
|---|---|---|---|---|
| Email verification, unique emails | Typos, dead addresses, exact repeats | Any real second inbox | Developers whose code lands in spam | Deliverability work |
| Disposable-domain blocks | Known throwaway services | New domains, ordinary mailboxes | Developers using privacy forwarding | A list someone updates forever |
| CAPTCHA | Simple scripts | People signing up by hand | People using screen readers | Fees and lost signups |
| Phone or card requirement | Casual second accounts | Anyone with another number or card | Students, and anyone who wants to try first | Fees and fewer signups |
| One per IP address | Repeat signups on one connection | Any other network | Bootcamps, offices, hackathon rooms | Exceptions and tickets |
| VPN blocks | Known VPN ranges | Everything off your list | Developers on company VPNs | Another list to maintain |
Say a 40-person coding bootcamp starts its first lesson on one connection. With a one-per-IP rule, 39 of the people you most want on your free tier get refused.
Inside the product, three more:
Idle timeouts catch the forgotten workspace, and real developers barely notice a 30-minute one. They measure activity, though, so anything still working counts as active, and the same rule can end a long test run someone left going over lunch.
Daily caps hold a single account down, and a 4-hour cap is kind to real developers. It doesn't change the sum, because 720 hours a month is still 12 accounts.
Manual review finds the obvious rows on your fleet page. At 2 minutes an account, 1,920 accounts is 64 hours every month, and the reviewer is guessing whether a workspace with no editor attached is a scraper or somebody's overnight job.
Idle timeouts come closest, and they do nothing about the next account.
What to do first on your own fleet
Give paying developers first claim on slots. Here, 380 paying workspaces need 12 hosts at peak (380 ÷ 32, rounded up). Reserve 13 and the free tier gets the other 7, which is 224 slots. The queue moves instead of disappearing: 340 free workspaces want 224 slots, so 116 wait, and about 61 of those are real free developers (116 × 180 ÷ 340).
Shorten how long you keep untouched free disks. Going from 90 days to 30 takes the pile from three months of disks to one, or $4,608 down to $1,536. Email the owner first, because a real developer back after six weeks will want their project.
Write what a refused developer reads. Offer a paid plan and a way to reach you, because some of the people you turn away will be real.
Those steps protect paying developers and shrink the tail. They also leave the question you had on the fleet page, whether this free claim should go through at all, and your records can't answer it. A brand new account arrives with nothing tying it to the last one, while a whole bootcamp on one connection can look worse than someone opening 8 accounts a day. Resetting the meter with a new account is the part these fixes leave open.
Evaluate each claim before the first free hour
Before a new account gets its 60 hours, one thing is worth settling: should this claim go through? AbuseZero was built for exactly that, and it's very good at it.
You've just watched a domain list go stale and an IP rule land on a bootcamp. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing. It checks mailbox aliases, disposable domains and mail routing. Where your VPN list stops, it sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. With multi-layer AI risk analysis, AbuseZero examines 140,000+ signals, detects linked claims and flags abuse that single checks miss, and it uses a proprietary algorithm, with AI that checks the claim again. Spammy bots, browser automation and malicious scripts get blocked while genuine AI agents and assistants are allowed, which counts for a lot in a product full of coding agents. Creating a fresh account isn't enough to erase previous activity.
You don't have to build your own abuse prevention system. Your server makes one API call before the first free workspace starts and gets a clear allow or deny, under rules you control. Eligibility can be once per user or once per calendar month. Observe mode shows you what would be denied before anyone is blocked, and the quickstart gets your first benefit protected. Your application still starts the workspaces and enforces the hours.
Frequently asked questions
What is cloud IDE free tier abuse?
It's one person opening more free accounts than you meant them to have, to keep workspaces running well past one account's allowance.
Should a cloud IDE require a card for the free tier?
It cuts casual repeat accounts, and it costs you the students and developers who want to try first. Here's the trade-off in detail.
Is every second free account abuse?
No. People keep work and personal accounts, and whole classes share one connection. Write down what you mean to give each person before you refuse anyone.
Protect your free hours
Start with Builder
Pay as you go. Protect your first benefit today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.