Terms of service

In force from 3 October 2026.

1. What this covers

These terms cover gagarin.cloud — a platform for deploying and running applications, comprising the website, the CLI (gg), the dashboard at my.gagarin.cloud, the API at api.gagarin.cloud, and the MCP server at mcp.gagarin.cloud, which is how a coding agent reaches Gagarin. Together we call all of that the Service. “We” is Gagarin; “you” is whoever is using it.

Most of what happens here is done by an agent acting on your behalf. That does not divide the responsibility: what your agent does with your credential is your doing, and what we owe you we owe you whichever of you typed it.

Signing up, deploying, or paying an invoice means you have read these terms and accepted them in full.

If you do not accept them, stop using the Service.

2. What you get

Access to compute: deploying containerised applications, resources (postgres, qdrant, valkey, iggy, rabbitmq, s3), routing including your own domains, and a meter that counts what you actually used.

Every project also has a memory: somewhere your coding agent can leave notes about the project for the next session to read. It is metered like everything else — by how many notes are in use, at the rate on the pricing page — and reading is not charged. How many notes a project may keep in use at once is also capped: reaching the cap refuses the write and says so, rather than quietly dropping anything, and archiving a note that is no longer true makes room again. It is part of the platform, not a resource: it is backed up with our own database, not separately, and cannot be restored on its own.

Resources are unmanaged today, with two exceptions: postgres, qdrant and iggy resources are backed up nightly and each backup is kept fourteen days, and every type’s credentials can be rotated on demand. No point-in-time recovery, no replication, and no backups of any other type — a rabbitmq keeps its queues on its volume, but we take no backup of it. What we do and what we deliberately refuse to do is on the “What we will never build” page, and it is kept current.

You run your own applications here at your own risk, and the architecture and the limits are disclosed in the documentation named above rather than hidden from you.

3. What we do not promise

The Service comes as it is. We do not promise:

• that it will do what you were hoping;
• that using it will get you any particular commercial result;
• that it will run without interruption or without error — yours or ours.

We are not liable for:

• losses of any kind, direct or indirect, lost profit included;
• data you lost or had damaged on an unmanaged resource;
• what third parties do or fail to do, our own cloud provider included;
• temporary technical failures, and restrictions on access.

What you run here, and what you do with what comes out of it, is your call and your risk.

4. Staying inside the law

The Service is not here to encourage, organise or help along anything unlawful. Use it within the law that applies to you. Whatever you put on it — the applications and the data both — is yours to answer for.

5. Who owns what

What we publish (including the CLI gg, under the MIT licence) comes under whichever licence we published it with. What you put on the Service — your applications, your code, your data, the notes in your projects’ memory — stays yours. We acquire no rights in any of it, and we do not train anything on it.

6. Cutting off access

We may suspend or restrict your access if:

• you break these terms;
• something is being abused, including load that threatens everybody else on the platform;
• the law or our payment provider requires it;
• an invoice goes unpaid.

Losing access does not wipe out anything you already owe.

What counts as abuse

Named rather than left to judgement, because you should be able to tell before you deploy whether we are going to stop you. Do not use the Service for:

• Cryptocurrency mining, or anything else whose purpose is to consume as much compute as it can get.
• Sending bulk or unsolicited email, or relaying mail on someone else’s behalf.
• Phishing — impersonating a person or an organisation to collect credentials, payment details or anything else.
• Attacking anybody: denial of service, port scanning, brute-forcing credentials, or exploiting systems you do not own.
• Hosting or distributing malware, or command-and-control for it.

This is not only a rule about you. Gagarin runs on infrastructure we rent, and a provider that finds abuse coming from our account suspends the account — which takes every other customer down with it. That is the actual risk, and it is why this list is short and firm rather than long and negotiable.

What we enforce automatically

Two of the above are prevented rather than policed, so you find out immediately rather than by receiving a suspension notice:

• Outbound connections on the mail ports (25, 465 and 587) are blocked for every service. If your application needs to send email, use a provider’s HTTPS API — which is what we do ourselves. This is not negotiable per-account.
• Sustained CPU at your service’s ceiling is measured. A service that sits at its limit for an hour is flagged for a person to look at. Being flagged is not a suspension: builds, imports and busy databases all do this legitimately, and a human decides.

What a suspension actually does

It stops your services. It does not delete anything. Your volumes, your data, your project memory, your domains and your certificates all survive untouched, and lifting the suspension starts everything again within a few minutes. You keep read access throughout — gg status and gg logs keep working, and the reason is printed above the table — because being cut off from the explanation as well as the service helps nobody.

If your account has simply run out of credit, adding some lifts it by itself. Anything else is a decision we made, and we will tell you what it was. If you think we got it wrong, write to support@mail.gagarin.cloud — we would rather be corrected than be wrong about somebody’s production.

7. Paying, and getting money back

We charge for what you actually used — the time your services ran, the size each one ran at, and the disk they held — at the rates published on the Pricing page when the usage happened, in US dollars. The smallest unit we bill is a minute: run something for part of a minute and you are charged a whole one, and nothing more than that. If you change a service’s size, the new rate applies from that moment rather than to the whole month.

A new account starts with a balance, stated on the Pricing page. What you use comes out of that. Nothing is charged while there is a balance to draw on.

Some of a balance may be credit rather than money you paid: the starting balance, credit we add by hand, and credit from referrals. Credit spends on the Service exactly like money you paid, but it is not money. It cannot be paid out, withdrawn, refunded, or moved to another account, and it does not count as a payment for anything that depends on you having paid. Referral credit is governed by the Referrals page, which is part of these terms: if the top-up that earned it is refunded or disputed, it is taken back, and if a referral is abused, we may void it and take back what it earned. Taking credit back can leave a balance at zero, in which case it is treated like any other empty balance below.

You may put a payment method on the account. If you do, you are authorising us to charge it automatically whenever the balance falls near empty — currently at about three days of your usage — for $20 or roughly two weeks of your usage, whichever is greater. Each charge is added to your balance. Nothing recurs on a fixed schedule, and we do not charge you in advance of a balance running low unless you ask us to.

You can switch the automatic charge off at any time, from your billing page, and keep the card on file. While it is off we do not charge the card unless you ask us to, we email you when the balance is running low, and if the balance reaches zero running services are stopped as described below. Switching it back on restores the automatic charge. Only the person signed in to the account at the console can change this; it cannot be changed with an agent’s credential.

Asking is the “top up” button on your billing page. You pick an amount — no less than $20, no more than $2,000, in whole dollars — and we charge the card on file straight away and add it to your balance. There is no separate payment screen, because there is no new payment method to collect. It is the same card and the same balance as the automatic charge above, and it works whether or not the automatic charge is switched on; what you choose is how much and when.

If a charge fails — not enough funds, a declined card, that kind of thing — we retry, and we will email you. If the balance reaches zero, running services are stopped. A stopped service is not charged for running while it is stopped; the storage it keeps — volumes, object storage and project memory — is still charged. Data and storage volumes are retained; see clause 6 and the next paragraph.

Storage costs us whether or not a service is running, so volumes belonging to an account that has been at a zero balance for 30 days may be deleted. We will email before that happens. Restoring a balance at any point before then brings everything back.

We do not refund resources you have already used. The exception is our own fault: if you were charged for something that was never actually provided, tell support within 14 days of the charge and we will look at it.

If you give somebody the editor role on your project, you are letting them run up your bill — deploying services, attaching resources, and so on. We do not tell the two of you apart when charging, and we do not refund what they spent, including if they spent it in bad faith. Arguments between you and somebody you handed access to are between the two of you.

Talk to us before you go to your card issuer. Do not open a chargeback without contacting support first.

8. Your data

Covered separately in the Privacy policy, which is part of these terms.

9. Changes to these terms

We may change them. The current version lives on this page with its date on it. Carrying on using the Service after a change means you accept the new version.

10. Getting in touch

Anything at all, payments and refunds included: support@mail.gagarin.cloud.