Future18 servers come with automation that watches, repairs and protects them, and an AI Agent you can talk to in the HWorker panel, on WhatsApp and on future18.com. This page explains, in plain words, how each of them decides what to do, what it may and may not do, and how you stay in control. It is part of our Terms and Privacy Policy.
1. The principles
- Facts first. Answers start from your server’s real data — logs, settings, DNS, service status — not from a general guess.
- Least power. The AI Agent works only inside the account, servers and websites your login may use, with the same permissions as your role (a read-only login gets read-only help).
- You approve changes. Nothing that changes how your server behaves happens without your “YES”; critical changes are reviewed by a Future18 engineer.
- Tested, and put back on failure. Every change is tested before it is applied (for example a web-server configuration test) and the previous state is restored if anything fails.
- Everything is logged. Each check, command and action is recorded and visible in your panel.
- A person is always near. Ask for a person at any time; long investigations become a support ticket automatically, with everything already checked.
2. How an answer is made
Every question goes through the same steps, and the first step that can answer correctly, answers:
| Step | What happens |
|---|---|
| 1. Guard | Limits on how many messages can be sent in a short time; a request for a person goes straight to our team; attempts to manipulate the agent (“ignore your instructions…”) are noted and refused. |
| 2. Direct answers | Questions Future18 has answered once and for all (prices, policies, how-tos) get our exact, approved answer. |
| 3. Understanding | The agent works out what you want and which website, mailbox, domain or server it is about. If that is unclear, it asks one short question instead of guessing. |
| 4. Learned answers | An answer that already solved the same question several times (three by default), and that customers rated helpful, is reused. |
| 5. Checks and playbooks | For common problems — a website showing an error 500 or 502, a slow site, a database connection error, an expired certificate, mail not arriving — the agent runs a fixed set of checks on your server (logs, PHP settings, DNS, certificate, running apps) and follows a playbook written by our engineers. |
| 6. AI model | If the steps above cannot answer, a large language model from a third-party AI provider reads a compact summary of the facts and may use the panel’s own tools — first a smaller model, then a larger one for harder cases — within a fixed number of rounds. |
| 7. A person | When the agent cannot solve it, or a playbook says so, the case is handed to Future18 support with the chat and everything that was checked. |
While it works you see live progress (what it is reading), and a Stop button ends the run. When a change you approved finishes, the result — done, or failed with the reason — is sent back into the same chat (WhatsApp or the panel). AI answers can be wrong; check important advice before you rely on it.
Something it cannot do yet. The AI Agent says so, and asks whether to add your request to its learning model. Only with your yes is the request (in one sentence, with your words) passed to Future18's team, who decide whether to build it.
3. What the AI Agent may do
The AI Agent uses only the tools of the HWorker panel — the same jobs you can start yourself. Each tool has a fixed risk level:
| Level | Examples | What happens |
|---|---|---|
| Read | Website status, logs, PHP settings, DNS, storage use, database sizes, WordPress administrators, following an email through the mail log, Shield alerts | Runs straight away; nothing is changed |
| Low | Issue or renew an SSL certificate, change a PHP setting or version, add or edit a DNS record, restart a service after its test, test WordPress plugins one at a time, re-activate a website or mailbox your team paused, suspend a website, stop / start / restart an app, back up a website now, install WordPress | Proposed to you; runs after your “YES” |
| Sensitive | Reset a password (the link goes to you or the account owner, never shown in chat), add or delete a mailbox or website, change a website's domain, restore a backup, import a database | Proposed to you; runs after your “YES”. Passwords are delivered by a one-time link or into your Vault — never shown in a chat |
| Critical | Stop a process, restore a quarantined file, anything needing root or a code change | Sent to a Future18 engineer for review; you see the outcome in the panel and by email |
The AI Agent never writes or runs code of its own on your server, never sees card details, and never acts on another customer’s account. On WhatsApp, only the number verified for your login can approve actions; a number that matches only a billing phone gets read-only help.
4. F18 Healer — keeping services running
- Watches the web server, PHP, databases, mail and other services every minute.
- A service that is down is restarted only after its configuration test passes — a broken configuration is never loaded. A failing test goes to our engineers with the output.
- A database that is slow to answer on the network is not restarted while it still answers on its local socket or a ping (a restart at peak time would drop every connection).
- At most a few restarts in a time window; then it stops and asks for a person, so it never loops.
- Cleans only safe things when a disk fills (package cache, old journals, rotated logs, old temporary files) and reloads PHP gracefully under memory pressure.
- Reads each website’s error logs: when PHP runs out of memory, the website’s memory limit is raised to 512 MB; other repeated errors are reported to you.
- Keeps SSL certificates renewed and makes sure the web server uses the renewed files.
5. F18 Shield — protecting the server
- Counts failed sign-ins and probes for known weaknesses in the server’s logs; successful sign-ins never count.
- An address that crosses the limit is blocked for a while; repeat offenders for longer. Addresses from your own country, Future18’s own systems and trusted services (such as WordPress.com / Jetpack and Cloudflare) are never blocked.
- Watches for programs that send unusual amounts of traffic out of the server and for files that look like web shells, which are quarantined (not deleted) so they can be restored if they were safe.
- Blocking a whole network needs your approval.
6. WhatsApp protection
Every message to our WhatsApp business number from a number that is not a Future18 customer is checked before it reaches our team or the AI Agent. The number is blocked, using WhatsApp’s own blocking feature, when the message:
- contains attack content — code or database injection, attempts to make the AI ignore its rules, or a request for a one-time code someone received — or an executable file;
- pretends to be WhatsApp, Meta, Future18 or Hosting Worker (in the profile name or the message);
- is a file type we cannot accept, or a message type WhatsApp marks as unsupported;
- is clearly promotional — at least two signs such as SEO or backlink offers, “limited offer” sales language, crypto or earn-from-home schemes, loan or prize messages, bulk-messaging offers or many links — or a group invite link. A single sign (such as asking us about a discount) is never enough.
Customers (a login’s mobile or an account’s billing phone), Future18 staff, numbers we wrote to, people who left their number in our website chat, and messages from our own ads are never blocked. Our team sees every blocked number with the reason and can unblock it; an unblocked number is never blocked automatically again.
7. Your data and the AI
- The AI model receives only what the answer needs: your question, a short summary of the relevant facts and the output of the tools it used.
- AI providers act as our processors under contracts that do not allow them to train their models on your data.
- Conversations and actions are stored with your account and kept as our Privacy Policy describes.
- Your thumbs up or down on an answer helps decide which answers are reused (step 4).
- The AI Agent always says it is an AI when asked.
8. Questions and complaints
If you think the AI Agent or any automation made a wrong decision about you or your server, write to support@future18.com or open a ticket in your panel and a person will review it. Privacy grievances go to our Grievance Officer at legal@future18.com.