One rate card. Three sizes of headroom.
Every plan meters the same way at the same unit rates. The fee does not buy cheaper units — it buys a bigger monthly allowance and the paid capabilities. On Free, usage stops at the allowance; it can never become a bill you did not expect.
| Included each month | Free$0Allowance is a hard cap | Pro$19per month · 2× allowance | Scale$99per month · 10× allowance | Past your allowanceSame rate on every plan |
|---|---|---|---|---|
| LLM promptcredits | 400K | 800K | 4M | $0.05 / 1M |
| LLM completioncredits | 200K | 400K | 2M | $0.40 / 1M |
| DataUnitsSQL · vector · graph | 1M | 2M | 10M | $2.00 / 1M |
| ObjectStoragebuckets · cold tables | 3 GB | 6 GB | 30 GB | $0.0199 / GB-mo |
| HotStoragereserved RAM / NVMe | 1 GB | 2 GB | 10 GB | $0.11 / GB-mo |
| Serverless computeIgnite vCPU | 12 h | 24 h | 120 h | $0.0432 / vCPU-h |
| CI build minutes2-vCPU runner | 300 min | 600 min | 3000 min | $0.008 / min |
| Git + registry storage | 2 GB | 4 GB | 20 GB | $0.05 / GB-mo |
| Egress | Unmetered | Unmetered | Unmetered | $0 |
| At the allowance | Usage stops | Keeps going, billed | Keeps going, billed | Billed monthly in arrears |
| Also included | Community support | 3 domains · 10 auth pools · 10K users | 25 domains · 50 M2M clients · 50K users · SSO · branch governance | — |
LLM rates are per 1M credits — tokens × the model multiplier, so a 1× model bills 1 credit per token. See the per-model rate sheet. Need committed volume or a sovereign deployment? Talk to us.
Estimate your monthly usage bill.
Nine meters, four groups, no bundles to decode. Set a month of usage and see the usage bill on a paid plan (the $19 / $99 fee comes on top) — and whether the monthly allowance already covers it.
The fine print, in plain English.
Do I need a credit card to start?
No. Sign up and build — every meter includes a monthly allowance, and on Free that boundary is a hard cap: usage stops, so you can never be charged by accident. When you're ready, upgrading to Pro or Scale is an explicit Stripe Checkout inside the console — a card on file by itself never changes your plan and never bills.
How big is the free allowance?
Sized for tutorials, demos, and small side projects — roughly 265 LLM requests on a 1× model, 100K vector searches (1M DataUnits), 12 vCPU-hours of compute, 300 CI minutes, 3 GB of object storage, 1 GB each of git, registry, and hot storage. It renews every month, and Pro and Scale include the same allowance before overage billing starts.
What does the monthly fee actually buy?
Capabilities, not cheaper units — the rate card is identical on every plan. Pro ($19/mo) unlocks overage billing past the allowance, 3 custom domains, a 10 GB hot-storage reservation ceiling, 5 auth pools, 10 M2M clients, and 10,000 monthly active users. Scale ($99/mo) raises those to 25 domains, 100 GB, 10 pools, 50 M2M clients, and 50,000 MAU, and adds SSO and git branch governance. Beyond 50,000 monthly active users, contact sales for Enterprise.
What is a DataUnit?
One work currency for every query on the data plane — SQL, vector, and graph. It's computed from data facts (bytes scanned, vectors returned, edges expanded), not CPU time, so a query's cost is predictable before it runs. Anchor: 1 DU = one 768-dim vector returned (minimum 6 DU per read). One number to watch instead of three products' worth of meters.
What's the difference between ObjectStorage and Reserved HotStorage?
ObjectStorage is bytes at rest — buckets, git repos, and cold table data — measured monthly at $0.0199/GB. Reserved HotStorage is RAM/NVMe capacity you explicitly reserve for low-latency serving, billed per GB-hour while the reservation is held, whether or not it's full. Reserve nothing beyond the included 1 GB and you're pure serverless: you pay only DataUnits plus cold ObjectStorage.
What happens at the allowance boundary?
On Free, the corresponding API calls return a quota-exceeded error until the next monthly cycle — no overage charges, no auto-upgrades, ever. On Pro and Scale, usage continues and the overage bills in arrears at the unit rates above, with budget alerts at 50/75/90% of a spend target you set. Paid plans are never denied for using more — that's what the fee buys.
Can I downgrade later?
Yes, to Free at any time, from the console. You pay any outstanding fee and usage already incurred, and the hard caps apply again from then on. Your data and configuration are unaffected — only the paid capabilities (extra domains, larger reservations, SSO, governance) stop being available for new provisioning.
Cheaper than the boxes you’d glue together — and capped, so you can’t be surprised.
Four principles, then the apples-to-apples math.
