Backing infrastructure for the apps your agent writes. A database, a game server, or any container, from one catalogue it calls as tools.
We host it.We don't watch it.
How a Kraftify box gets deployed
We host it. We don't watch it.
The pitch is that we host it and do not watch it. Here is what that means in practice.
No logs, no retention
Console output streams to you and drops at the socket. Chat is never relayed and query logs are never written. What remains is a retention manifest four rows long, and one of those rows exists because tax law says so.
Secrets never arrive
Env values are set by whoever already holds them. We record a hash so you can tell whether the value in force has changed, and there is no endpoint that returns a value, because there is nothing stored to return.
The agent is capped and scoped
Every create is checked against your monthly ceiling first, and nothing in the tool surface can raise it. Tokens are scoped, console write is not exposed to agents, and read-only mode exists for the week you spend deciding what to trust.
The complete list is one page: what we store.
Everything your agent can reach.
Agents, databases, game servers and containers: one catalogue, the same tools, capped by the same ceiling.
FAQ
The questions people actually ask.
Short answers first, docs linked where it matters.
Start
Postgres, Redis, a Minecraft server, or any container on a TCP port. One catalogue, same flags, same tools, same bill: an agent asking for a database and an agent asking for a game server call the same function with a different argument. The container kind is the escape hatch, so nothing is off the list by being unlisted.
Start
It doesn't, from us. The agent creates the target, then calls kraftify_set_env, which records that DATABASE_URL is set and nothing else. The value is never sent to us and no endpoint returns it. We keep the key name and a set-flag so the console can show it is configured. See What we store.
Start
No, and that is the guarantee we would point at first. Every account has a monthly ceiling, enforced server-side on every create and start — refused, never queued — through the same code path for the console, the CLI, and the MCP surface alike. Neither surface has a path around it. The agent can read the ceiling first, so it tells you a request is too large instead of failing into it. See Billing.
Start
Funding. A region is a rack, a network, and someone awake at 3am when a node goes bad, and we will not sell a region we have not paid for. No preorders, nobody charged. The week the money lands, the region goes up and the status page says so first. To speed that up, book a call.
Game servers
Yes. A Minecraft target pins a version and loader, and modpacks arrive as templates, so every install from one template is identical. If a pack needs something unusual, the container kind runs anything on a TCP port.
Game servers
Not directly: they are separate targets speaking different protocols. Run one of each, or bridge them with a proxy in a container. The limits come from the Bedrock side, which supports fewer protocol features, so test crossplay before inviting everyone.
Game servers
Vanilla with a few friends fits in 4 GB. Large modpacks want 8 GB, and big public servers want 16 GB. Those are exactly the small, medium and large sizes, and you can resize up without losing the world. See Deploy a target.
Game servers
Scheduled snapshots of data and config: the 24 most recent hourly and 7 daily, pruned automatically. Restoring stops the target, swaps the files, and starts it again. Snapshots live on the same node, so they restore fast but do not survive node loss. See Backups.
Game servers
Every target gets a hostname on a Kraftify domain, and your own domain can point at it instead. See DNS. Per-region DDoS mitigation will be specified at launch and carried on the status page: no region is funded yet, so there is nothing to measure.
Data and trust
Your content data is four fields we cannot read: email, password hash, env key names and set-flags, and the billing record. Console and query output streams to your client and is discarded at the socket. No command history, no chat history kept, no crash dumps, no query log, no plugin telemetry. Account and operations data (OAuth profile, sessions, snapshot, abuse records) is listed in full in What we store.
Data and trust
Your files and traffic, no: there is no console in our tooling that opens them. Automated abuse screening does run on domain requests and abuse reports — that is how CSAM, phishing, and attack infrastructure get caught — and an engineer debugging your box sees kernel logs and resource metrics. Anything beyond that would have to come from a court.
Data and trust
None yet, because there is nothing running to promise it on. When regions open, the status page carries live status and incident history, and any SLA will be written there first, not here.
Data and trust
You pick the region at deploy from the funded list. Planned regions are US East and Europe, both still waiting on funding. See Region status.
Data and trust
World and target files export directly, and env values were always yours since we never held them. Delete the target, then the account, and only the seven-year billing record stays, because tax law requires it. See What we store.
Data and trust
Three of us build this under one alias, noncera. A no-logs host should not be a place where people go looking for faces.
Billing and support
The target is snapshotted, you get seven days to export it over S3 or FTP, then it is deleted. We keep nothing just in case, because an indefinite backup is a copy we hold, which is the exact thing we do not do.
Billing and support
Pro-rata, on request, within the first 30 days, no interrogation. Past 30 days we won't refund a month already consumed, but you can leave whenever and we keep nothing after.
Billing and support
No support inbox yet — no email on our side. The docs and status page answer most things, and booking a call is the fastest way to reach us.
Still stuck? Read the docs and check the status page.
No regions live yet.
Leave an email and we will tell you when a region opens.
Point it at your agent.
Nothing is live yet. This books a call, not a sandbox.
- regions
- none live yet
- agent spend
- capped at your ceiling
- retained
- 0 bytes of console