Hosting for the apps you build with AI agents
An agent can produce a working application in an afternoon. Then it needs somewhere to run, a database, a private path between the two, a certificate, backups and someone watching it. That part is what we do, and each piece of it is one action or one API call.
The loop
Write, push, review it running, merge
The faster an agent writes, the more the bottleneck moves to everything after the code. These four steps are the whole cycle.
The agent writes it
Claude Code, Cursor, Copilot, an in-browser builder, or your own hands: whatever you work in, the output is a repository. Nothing about your project has to be shaped around us, and there is no library of ours to add.
Push, and it deploys
Connect GitHub, GitLab or Bitbucket, choose the branch, and every push builds into your own container. 14 framework presets, or a Dockerfile if the presets do not fit.
Review it running
Each pull request gets its own live URL. When an agent has written more code than you want to read, running it is the faster review. Included from Pro upwards.
Merge, or roll it back
Merge and the production container rebuilds. If a commit turns out badly, redeploy the previous one in a single action. Your data lives in a separate database, so code and data roll independently.
What the app needs
A database, and a private path to it
Three engines, each in a container of its own with memory and CPU you set. Attach one to a site and it joins that site's private network in the same action.
PostgreSQL
version 15The default choice for an application database.
Redis
version 7Cache, queues, sessions and rate limiting.
ClickHouse
version 25.8Analytics over events, logs and product usage.
The database is not on the internet
Each customer gets their own subnet. An attached instance gets a private address on it, the app reaches it there, and your laptop can join the same network with a WireGuard peer config when you need to run a query by hand. Private networking is included on Pro, Studio and Agency, from £59/mo. A live standby with automatic failover is available for PostgreSQL and Redis at £7.50/mo per database.
Let the agent do it
A key scoped to the job, not to your account
If you would rather not click through the setup yourself, create a key in the dashboard and tick what it is allowed to touch. Hand it to your agent or your build pipeline. It is a key for the job, not a way into your account.
Writing an integration rather than pointing an agent at it? The API reference is here.
A key cannot widen itself
Creating, rotating and revoking keys requires a signed-in browser session. A scoped key cannot mint a broader key, so a leaked deploy key stays a deploy key.
Expiry and a label
Set an expiry when you create it, and give it a label naming the job or the agent it belongs to. Up to 10 keys can be active at once.
Rotate in place
Rotation issues a fresh token with the same name, scopes and expiry. Swap it into your agent, then retire the old one.
Recorded
Key creation, rotation and revocation land in your audit log with who did it and when.
Already handled
The work nobody wants to build twice
These are included on every plan, on UK infrastructure we run ourselves, at the same fixed price.
Questions from people shipping agent-written code
Answered plainly, and only about what runs today. If you want the whole thing as a walkthrough, read how to deploy an app your AI agent built.
Does it matter which AI tool wrote the code?
No. We deploy from a git repository, so anything that produces one works: Claude Code, Cursor, Copilot, an in-browser builder, a teammate, or you at two in the morning. There is no G7Cloud SDK to install and no project layout to adopt.
Can an AI agent deploy on its own?
Yes, if you want it to. Create a key in the dashboard, tick what it is allowed to touch, and give it to the agent. Most people start by doing the setup themselves and hand over the repetitive parts later, which is a perfectly sensible order to do it in.
What stops an agent doing something I did not intend?
A key only carries the permissions you ticked, and it can never create another key or widen its own reach. You can give it an end date, replace it, or revoke it outright, and each of those is written to your audit log. Your billing and account settings are never in its reach at all.
Which databases can I run next to the app?
PostgreSQL 15, Redis 7 and ClickHouse 25.8, each in its own container with its own memory and CPU. Attach one to a site and it joins that site's private network in the same action, reachable at a private address rather than a public one.
How is this billed if my app gets busy?
One fixed price per plan, in GBP. There is no per-request charge, no bandwidth meter and no invoice that arrives larger than you expected because something went viral overnight. Plans run from £12/month, and month 13 costs what month 1 cost.
Do I get preview environments on every plan?
PR preview environments are included from Pro upwards, and on the Solo Plus and Solo HA single-site plans. Git push-to-deploy itself works on every plan that includes a hosted site, from Starter at £12/month.
Can I reach the database from my laptop?
Yes, over the private network. Add a WireGuard peer for your machine, import the config, and your local tools connect to the private address directly. No database port is ever exposed to the public internet.
What happens to the code if I leave?
It stays yours. Your app is built from your repository, which you own and control, and your data lives in a database you can back up and download. There is nothing proprietary in the middle to extract yourself from.