Let AI help run your server. You hold the switch.
Ask in your own language and get an answer from your own data. When a change is needed, the assistant writes a card, and nothing happens until a person presses Approve. Off until you turn it on, and in every tier.
Ask. Propose. Approve.
The assistant can read what you are allowed to see. It cannot change anything by itself.
1. Ask
Ask in your own language: why mail is not arriving, how much disk is left, which accounts are close to their limit.
2. It proposes
For a few changes, such as adding or deleting a DNS record or suspending an account, it prepares a card. The panel writes the card, not the model, so it says exactly what will happen.
3. You approve
Nothing runs until a person presses Approve, confirms, and types their password. The person’s own permissions apply, and a proposal expires after 30 minutes.
Proposed change — nothing has happened yet
Delete the DNS record A old → 203.0.113.10 in the zone of example.com.
Check this card: it, not the assistant’s message, says exactly what will happen.
30 minutes left to decide
Watch a change wait for you
Recorded on a test server. The assistant’s answer is scripted for the recording; the card, the two confirmations and the result are the panel’s own.
Inside the panel, or from your own AI tool
An assistant inside the panel
A chat panel for administrators, resellers and customers. It sees only what the person asking may see. The operator chooses the provider and the key, sets the limits (30 questions a day and 2,000,000 tokens a month to start with), and decides who may use it.
Connect your own AI tool
A tool such as Claude Code connects through an MCP address on the panel. That address answers 404 until the operator allows it. A token can be limited to some areas and set to read-only, and deleting or suspending anything becomes a request that the token’s owner approves.
Built so that you can say yes.
Off until you switch it on
A new install has the assistant and the MCP address turned off. The server owner turns them on and chooses who may use them.
It sees what you see
Answers come from your own panel data, through your own login. A customer’s assistant never sees another customer’s account.
The card is written by the panel
The model’s reply is free text; the card is not. It states in the panel’s own words what will happen, and warns you when the assistant read messages written by other people.
Limits you set
Daily and monthly limits per person keep the bill predictable. A token can be read-only and limited to the areas it needs.
Your key stays yours
The provider key lives in a file only root can read, on your own server. It is never shown again, not even to an administrator.
A trail you can read
Questions to the assistant, proposals, approvals and refusals are written to the audit log.
Your provider. Your key. Your server.
Choose from ten providers, or point the assistant at any OpenAI-compatible or Anthropic-compatible address. Run a model on the server itself with Ollama, and the questions stay on your machine.
- Anthropic (Claude)
- OpenAI
- Google Gemini
- Mistral AI
- Groq
- DeepSeek
- xAI (Grok)
- OpenRouter
- Together AI
- Ollama
- Any compatible address
Choosing the AI, and keeping the bill small
The assistant runs on your own key from the provider you choose. Nimbopanel never sees your key or your questions: they go from your server straight to the provider.
You pay the provider, by use
Cost grows with the number of questions and the length of the answers. Prices change often, so open the provider’s price page before you decide, and also set a spending limit in the provider’s own dashboard. Nimbopanel adds no charge.
Which provider?
For good answers with the least work, choose a hosted provider such as Anthropic, OpenAI or Google. For one key that reaches many models, choose OpenRouter. To keep every question on your own server, run Ollama there: no per-question fee, but it needs enough memory and smaller models answer less well. You can change provider later.
Start small
Keep the starting limits (30 questions per person a day, 2,000,000 tokens a month for the whole server) and let only administrators use it at first. 0 means no limit, so leave that for later.
Decide what the AI may read
Tick “Keep support tickets away from the AI service” if you do not want customers’ messages sent to a provider. Every person sees a privacy notice before their first question.
It looks widely. It proposes narrowly.
It can look at
Hosting accounts and their usage, domains, DNS records, email accounts (never passwords), databases, backups, server load and services, the mail queue, resellers and support tickets. Each person sees only their own part.
It can propose, and you approve
Adding a DNS record, deleting a DNS record, suspending an account and bringing an account back. Each one is a card that names the account and the record, and waits for you.
What to know before you rely on it
It can be wrong
An answer is a model’s best reading of your data. For a change, the card is the source of truth, not the answer.
Not every vendor is tested yet
We have tested the assistant against a stand-in provider, not against each vendor’s live service. Try it with your own key on a test account first.
You pay your provider
Questions are billed by the provider you chose, to the key you entered. Nimbopanel adds no charge, and the free tier includes it.
Questions, answered
Is the AI assistant on by default?
No. A new install has both the assistant and the MCP address switched off. The server owner turns them on, enters a provider key, and chooses who may use them.
Can the assistant change my server by itself?
No. It can read what the person asking is allowed to see. For a few changes it prepares a card, and nothing runs until a person presses Approve and confirms. A proposal expires after 30 minutes.
Which AI providers can I use?
Anthropic, OpenAI, Google Gemini, Mistral, Groq, DeepSeek, xAI, OpenRouter and Together AI, or a model running on the server itself with Ollama. You can also point it at any OpenAI-compatible or Anthropic-compatible address.
What is MCP, and do I need it?
MCP is an open standard that lets AI tools such as Claude Code work with other software. Nimbopanel offers an MCP address so your own tool can look at your panel. You do not need it for the assistant inside the panel, and it answers 404 until the operator allows it.
Which provider should I choose?
For the best answers with the least work, a hosted provider such as Anthropic, OpenAI or Google. For one key and many models, OpenRouter. To keep every question on your own server, Ollama. Try one on a test account first; you can change later.
Does it cost extra?
Nimbopanel adds no charge: the assistant and the MCP address are in every tier, the free one included. The questions themselves are billed by the AI provider you chose, to the key you entered.
Try it on your own server
The free tier includes it. Switch the assistant on when you are ready, and off again whenever you like.
Start free