AI will write your app in an hour. Who watches it for the next year?
In 2026 data leaked from thousands of apps built in Lovable, Bolt, v0 or Cursor — not because of the hosting, but because of what stayed in the code: open APIs, leaked keys, missing headers. Grove Cloud is hosting from a security company: every deployment goes through the same scanner Grove Tech AI uses on thousands of Czech websites.
Six layers included in the price
206 checks
Security headers, TLS, keys leaked into the build, malware and phishing, exposed admin panels and APIs, dependencies with known CVEs. The result appears next to your app within minutes.
Defender
Runtime protection for the AI layer: prompt injection, data leaking through model responses, suspicious requests. The same SDK as @grovetech/defender.
A dedicated container
Each dynamic app runs in its own container with a memory limit. Static sites are served from a shared server with no code execution at all.
Daily, off the node
Daily backups of data and configuration, versioned images for rollback.
Germany, EU law
Hetzner Nuremberg. Support in Czech and English, invoicing in CZK, contract under Czech law.
Deep test on request
When you want more than a scan: pentest with 33 active probes and a report for your customers or an audit.
Honestly: a scan finds what can be observed from the outside and from the build. Flaws in business logic, or in the data the app stores itself, only surface in a pentest or a code review. We never promise "100% secure".
Multiple AI agents in one app? You see each one separately.
Automatic protection in Grove Cloud covers every model call without touching your code. Since Defender 0.29.0, add an x-grove-agent header to the call and the dashboard splits calls, spend and interventions per agent. The header is stripped before the request reaches the provider.
Three agents (support-bot, billing-agent, research-crawler), deployed via Grove Cloud from a public repo, OpenAI key added as an environment variable. After 15 calls: two agents 100% clean, research-crawler — deliberately fed poisoned input — flagged 5× for prompt injection. Spend read from the usage of every response.
The test app is public: github.com/grovetechai/grovecloud-test-agents
// OpenAI SDK
new OpenAI({
defaultHeaders: {
"x-grove-agent": "support-bot"
}
});
// fetch
fetch(url, { headers: {
"x-grove-agent": "billing-agent"
}});The automatic layer covers Node (fetch) and Python (httpx and requests — i.e. the OpenAI, Anthropic and Mistral SDKs and LangChain). What we do not see and do not sell: streamed responses, HTTP clients other than fetch/httpx/requests, and local models. For those, use the full SDK in code.