AI Policy

Last updated: 4 July 2026

This policy explains how CreeperHost uses AI-assisted systems, including Theo in CreeperPanel, crash-report explanation, support-ticket assistance, diagnostics, and internal support tools.

It is deliberately written plainly. AI is not just a technical feature. It affects trust, work, privacy, infrastructure, and the people around it.

The short version

AI at CreeperHost is a tool for gathering information, explaining technical detail, and helping customers and staff move faster. It is not a replacement for people.

Our team are not placeholders waiting to be automated away. CreeperHost is built by people with deep experience in games, hosting, Minecraft, modding, modloaders, infrastructure, support, and the strange edge cases that only show up after years of running real services for real players.

For customer-facing AI features, our position is local-first and CreeperHost-controlled. We do not silently send customer personal information, support tickets, account history, service data, server logs, diagnostic output, or chat content to third-party hosted large language model providers for AI processing. If that ever changes for a specific feature, we will clearly tell you first or ask you to choose it.

AI-assisted tools support CreeperHost staff. They do not replace human review for cancellations, refunds, account closures, abuse handling, billing disputes, or other decisions that could have a serious effect on you or your service.

A personal note from Paul

This part is written in the first person because the responsibility is personal as well as technical.

I am not completely comfortable with AI. I do not trust it by default, and I do not think anyone should.

That is not a reason to ignore it. CreeperHost has always been early to look inside new technology, especially where it might help players, server owners, modders, and our own team solve difficult problems. Pandora's box is open now. Pretending otherwise will not protect anyone.

The principle behind our approach is simple:

Fear should never stop progress, but progress should never harm people.

That means we will explore AI carefully. We will build the parts we depend on. We will keep people responsible for decisions. We will be honest about what the system can and cannot do. We will not present AI as magic, and we will not treat customers or staff as test subjects for careless automation.

What we use AI for

We may use AI-assisted systems to:

• Help customers understand information already available in CreeperPanel.
• Explain crash reports, logs, service status, diagnostics, or common configuration problems.
• Summarise support tickets and account or service context for staff.
• Convert customer-provided scheduling instructions into technical formats.
• Help gather time-sensitive diagnostic information when a support request is opened.
• Draft support-ticket text from information the customer has provided or approved.
• Help staff find relevant information faster so they can spend more time solving the real problem.

Theo, our CreeperPanel assistant, is designed around this same idea. He can help inspect service state, explain likely issues, and prepare certain panel actions. He is still new, and where he is marked beta or prerelease, he should be treated that way.

We refer to Theo as "he" because that is how we treat him inside CreeperHost: a named assistant with a job, limits, and responsibilities, not an anonymous automation layer. He is part of the way our team works, not a substitute for the people responsible for him.

Model choice and purpose-built tooling

Theo is not trying to be ChatGPT, Claude, or any other large public frontier model. He is not meant to be the best general-purpose chatbot in the world, and we do not want to pretend otherwise.

Our priority is different. Theo is built for CreeperPanel, our services, and our support workflows. That means he may sometimes be less polished than a much larger public model at open-ended conversation, but he has something a generic chatbot does not have: carefully limited first-party context, purpose-built tools, service-aware checks, confirmation flows, and direct knowledge of how our panel actually works.

That is the trade we are choosing. We would rather have a narrower assistant we can understand, operate, restrict, and hold accountable than a more impressive general model running somewhere we do not control.

When Theo is useful, he should be useful because he can safely gather the right facts and prepare the right panel action, not because he sounds clever.

What we do not use AI for

We do not use AI to replace our team.

We do not use customer personal information to train public AI models.

We do not use AI to make final decisions about cancellations, refunds, account closures, abuse handling, billing disputes, or other matters with legal or similarly significant effects.

We do not expose broad destructive control to AI systems. In CreeperPanel, this means Theo is not given arbitrary access to:

• Delete, edit, rename, copy, unzip, or extract arbitrary files.
• Destroy, uninstall, or remove services.
• Run arbitrary console commands.
• Kill system processes.
• Change root passwords or SSH keys.
• Destroy or grant account/API credentials.
• Submit billing cancellations, refunds, orders, or payment changes.

Where Theo can prepare a service-impacting action, he must do so through an allowlisted panel action with a clear confirmation step. If the confirmation is missing, expired, ambiguous, vetoed by safety review, or the system is not confident enough, the action should not run.

Our infrastructure position

For customer-facing AI features, we do not want the sensitive part of the work running on rented AI hardware in someone else's data centre.

Theo's physical hardware was built by me personally, in-house, for CreeperHost. He runs in our own data centre, behind our own physical security and two steel doors. That matters to us because trust is not only about software. It is also about where the data goes, who can touch the machines, who operates them, and who is accountable when something goes wrong.

We may use OpenAI-compatible APIs as a technical interface. That does not mean customer AI requests are automatically sent to OpenAI or another hosted AI provider. Our customer AI features are designed to run on CreeperHost-controlled infrastructure unless we clearly tell you otherwise.

Data and privacy

AI-assisted systems may process information relevant to the request, such as support ticket contents, chat messages, account details, service details, IP addresses, server logs, diagnostic output, billing/support history, crash reports, or current service status.

We use that information to provide support, operate services, explain technical issues, and help customers control their own services. We do not use it to train public AI models.

We try to keep AI access bounded to the task. Where a feature needs to share diagnostic text, create a paste, attach support material, or prepare a visible customer action, the system should use a specific server-side flow rather than handing raw credentials or broad access to a model.

Our Privacy Policy gives the formal data-protection position. This AI Policy explains the practical philosophy behind it.

Ask Theo chat history

Ask Theo keeps recent transcript history in your browser's session storage so CreeperPanel can show the conversation if you navigate around or refresh during the same browser session. That browser copy is on your device. It is not a CreeperHost chat-history database.

On the server side, Theo keeps the active chat session in memory so he can understand the current conversation. If the conversation gets large, older messages and tool records may be compressed into a structured summary, but that summary is still held in RAM as part of the active session.

We do not currently store Ask Theo chat transcripts or compressed session memory in a database for later training, profiling, or review. The in-memory session can disappear when it expires, when the service restarts, or when infrastructure is rescheduled. That is intentional.

Support tickets, diagnostic pastes, text attachments, or other records you explicitly create through Theo are separate from chat history. Those are handled as support or service records because you asked the system to create them.

Human responsibility

CreeperHost has people who know this field deeply. Our team includes long-standing names from the games industry, MMO server reverse engineers, Minecraft mod developers, Minecraft modloader developers, infrastructure engineers, and support staff who understand the difference between a quick answer and the right answer.

AI can help those people find facts faster. It can help a customer get unstuck at 2am. It can help preserve useful context before a log rotates away. It can help turn a messy support request into something clearer.

It cannot replace judgement. It cannot replace responsibility. It cannot replace the experience of people who have spent years understanding how these systems fail in the real world.

Approvals and service-changing actions

Read-only checks may happen when needed to answer a question, such as checking service status, looking at recent logs, listing available backups, or gathering diagnostic signals.

Actions that can change a service need a clear approval flow. The customer should be shown what will happen before it happens.

For Theo, confirmation is not a single blind "yes" check. Service-changing confirmations are reviewed in layers before an action can run, and unclear or mismatched confirmations are blocked.

Where model review is needed, two separate local reviewers must both agree that the customer's confirmation matches the requested action. Those reviewers use different models and are deliberately run on separate local GPU stacks, with different GPU manufacturers and driver stacks, so one model, runtime, driver, or hardware path is not the only thing standing between a request and a service-changing action. If either reviewer disagrees, is unavailable, fails to return a valid decision, or thinks a human should look at it, the action is blocked.

Approving an action still matters. If Theo asks you to approve something, check the request. If the request looks wrong, cancel it and contact support.

Environmental responsibility

AI should not be wasteful theatre. We care about efficiency because power, cooling, hardware, and water all matter.

Running our own infrastructure does not give us permission to be careless. It gives us responsibility. We design these systems to use bounded context, smaller utility models where appropriate, queues, timeouts, and specific tools rather than throwing unnecessary compute at every problem.

Useful AI should reduce waste, not create a new kind.

If something feels wrong

Theo and other AI-assisted tools can misunderstand requests, make poor suggestions, or miss important context. If an AI-assisted answer does not feel right, stop and contact our team.

We would rather have a human look at the problem than let an uncertain automated answer push you into a bad decision.