AI Ops
AI that acts — not just chats
NexCMP is built by the same team that runs onie.net, a cloud platform where AI already does real, governed work today — not just chatbot Q&A. As we bring NexCMP to market, we are integrating that same AI-operations layer: a chat assistant that can act on your account once you confirm, an agent that can safely run real commands on your own server, a control surface for AI tools like Claude and Cursor, and a supervised AI system that helps triage DDoS traffic. Part of this already runs natively inside the NexCMP engine; the rest runs today at onie.net and is being integrated — every feature below is labeled honestly with exactly where it stands.
Verified positioning
“Between April and September 2026, our team shipped a continuous line of AI-operations features handling real customer actions — chat-based account actions (April), an MCP server and an SSH agent that executes real commands (May–July), a support-ticket handoff bot (August), and a supervised AI-driven DDoS auto-mitigation system now in production (September) — placing this work in the same category of action-taking, agentic AI-ops as the hyperscalers and CMP vendors we reviewed in our market research.”
Independently fact-checked 2026-09-27 — every date above is tied to a fact-checked "first live" record for the corresponding feature; the line makes no ranking claim against any named competitor.
Features
Six ways AI already operates here
AI Chat Assistant
Ask questions about pricing or your account, or ask it to start, stop, create or remove a server — the assistant drafts the action and hands you a confirm button; nothing changes until you tap it. It also knows your live account context (your resources, current balance, active promotions) so answers are grounded in real data, not guesses.
How AI is kept safe
Only a fixed, hard-coded list of account actions is reachable at all. Anything that changes or removes a resource always waits for a confirm tap — never triggered by text alone. The assistant is instructed never to invent account numbers, balances, passwords or IDs. Any image you upload is re-processed to strip hidden or malicious data before the AI ever sees it.
AI SSH Agent
Open a server's console and ask the AI agent to check its health, diagnose an issue or run a fix — safe, read-only commands run right away, and anything with real risk pauses and shows you a Run/Skip choice.
How AI is kept safe
The agent can only reach a server you already own (or one explicitly shared with you). A fixed set of dangerous command patterns — deleting data, sending data out, self-replicating scripts, shutting the machine down, tampering with access — always pauses for your decision, even in autonomous mode. Server output is treated as untrusted data, never instructions. Each turn is capped at a fixed number of action rounds, on a small capped budget. Has been through an adversarial security review and passed every scenario tested.
MCP Control
Connect Claude, Cursor or any MCP-compatible AI client to your account and manage your cloud in natural language — list resources, start or resize a server, or clean up something you no longer need, across hundreds of tools covering nearly the whole platform.
How AI is kept safe
Every tool call that deletes or destroys something requires an explicit "confirm" flag — without it, the server refuses outright, at the code level. The approval prompt before a tool runs comes from your own AI client, so you are always the one clicking allow. Access is scoped to your personal account token — never an administrative one — and administrative endpoints are blocked outright in code.
AI DDoS Analyst
Suspicious traffic patterns are first flagged by a plain, deterministic detection engine (fixed thresholds, no AI) — the AI's job starts after that: it analyzes the traffic window and related signals and returns a verdict (attack / suspicious / normal) with a confidence score. Only above a high confidence threshold does the system apply a temporary block; the block always carries an expiry, is re-checked on a schedule, and is lifted automatically once traffic returns to normal.
How AI is kept safe
The AI never decides on its own that something is an attack — a separate, deterministic detector opens the case first. It only takes the extra step of applying a block when confidence clears a high bar, and every block is time-limited, auto-reviewed and auto-lifted. Every verdict and every action is logged and reported to a human operator.
Chat → Ticket / Human Handoff
Ask the chat assistant to open a support ticket — even before you've logged in — or ask to speak with a real person, and the conversation is handed off automatically.
How AI is kept safe
Guest requests (before login) require a bound, single-use security token so they can't be replayed from elsewhere. Requests are rate-limited per email and per real visitor address. If the email already belongs to an account, the system only opens a ticket — it never resends an account-reset email to someone else's inbox.
AI API Key Gateway
A metered marketplace lets you buy pay-per-use access to mainstream AI models through one unified gateway, at prices below list, without managing multiple provider accounts.
How AI is kept safe
Spend and budget limits are enforced with the same atomic accounting used for billing, so two simultaneous requests can never double-spend a budget or push an account past its cap. Every account's active/blocked status and remaining balance is checked before a single AI request is let through.
How AI is kept safe
The same five rules, every feature
Confirm before acting
Anything that changes or deletes a resource always stops for an explicit human decision — a button tap, a Run/Skip choice, or a required "confirm" flag. No AI feature here silently executes a destructive change from a sentence alone.
Whitelist, not free-form
The AI can only ever call from a fixed, hard-coded list of possible actions/tools. It cannot invent a new action, reach an arbitrary internal system, or step outside that list.
Anti-hallucination guardrails
The AI is explicitly instructed — and in several places enforced in code — never to invent account numbers, balances, passwords, IDs or server identifiers; it can only report values that came back from a real, live lookup.
Scope lock
Every action is restricted to resources you actually own (or that were explicitly shared with you) — never another customer's server, storage or account, and never an administrative capability.
Logged and reviewable
Every AI-triggered action is recorded, and higher-risk automated actions (like a temporary DDoS block) are reported to a human operator and re-checked on a schedule rather than left to run unsupervised indefinitely.
Why this matters
See it work
Demos
Every demo below is simulated and uses only RFC 5737/3849 example addresses or invented example values — never real customer data.
Chat-to-server
Ask the chat assistant to create, stop or remove a server. It replies in plain language with a one-tap confirm button — the action only runs once you tap it.
Simulated demo — illustrative data.
Control your cloud from Claude/Cursor (MCP)
Connect an MCP-compatible AI client to your account and ask it to list or resize resources — any delete or destroy call requires both your client's own approval prompt and an explicit confirm flag.
Simulated demo — illustrative data.
AI agent in the server console
Open a server's console and ask the AI agent to check disk space or restart a service — safe commands run immediately with their real output shown; anything destructive pauses for your Run/Skip choice.
Simulated demo — illustrative data.
Supervised AI DDoS analysis & auto-mitigation
When traffic on a protected address turns suspicious, AI analyzes the pattern and returns a verdict with a confidence score; only above a high threshold does it apply a temporary block, which is reviewed on a schedule and lifted automatically.
Simulated demo — illustrative data.
Hourly billing, transparent to the line
Watch a billing tick run live: usage is charged to the hour with an append-only ledger, so re-running the same tick never double-charges a customer. Not an AI feature — the deterministic core engine underneath the platform, shown for context.
Simulated demo — illustrative data.
White-label reseller
Create a reseller, customize their storefront branding and colors, add a custom domain, and set resale pricing above a protected floor. Not an AI feature; this capability is in beta today, not yet serving live customers.
Simulated demo — illustrative data.
Honesty check
What AI can't do yet
- No AI feature here executes a destructive action (delete, stop, restart, unblock) without an explicit human confirmation — a button tap, a Run/Skip decision, or a required confirm flag.
- The chat assistant does not let you pick a specific data-center region when creating a server through chat today — it always provisions to the platform's current default region.
- The MCP server does not yet support single-click OAuth sign-in for AI clients — access today is set up with a personal account token you generate yourself and paste in.
- The SSH agent cannot run interactive commands that need a live terminal (a text editor, a package manager prompting yes/no) — it will tell you to run those yourself. It is also capped at a fixed number of action rounds per turn, and only acts on a server you already own.
- The DDoS system's AI does not detect attacks by itself — a separate, deterministic engine always opens the case first; the AI only analyzes and concludes afterward. Fully-automatic blocking is an internal operations capability today; a limited, suggestion-only version for customers is a separate, smaller-scope feature.
- Billing is not AI-powered — it is a fully deterministic engine; where AI appears near billing, it only summarizes or explains a bill in plain language, never calculates or decides an amount.
- None of the features on this page train a custom or proprietary AI model — they route to established third-party AI models through a shared gateway.
Frequently asked questions
Ask about the AI roadmap
Tell us what you'd want AI to do on your own infrastructure and we'll tell you honestly what's ready today.