The control panel for managed AI

Customers ask for AI. Say yes, and run it on your servers.

AI Admin Panel lets hosting providers and MSPs run AI apps and agents for every customer, on servers they control. Each customer gets their own account, their own apps and only the controls you allow.

Get started See how it works

Free to try with up to three customers and five services.

19 in
  1. 01Your servers
  2. 02Your customers
  3. 03Your rules
  4. 04From zero to agents in three steps
  5. 05What you can check today
Fig. 1.1Front elevation. Select a unit to open its section.

§ 01

You want to say yes. Then the work starts.

A customer asks if you can run an AI agent for them. Then another wants a workflow tool, and a third wants a chat app for their team.

Each request becomes a separate server, set up by hand. Every small change goes through your support desk, and after a while nobody is sure who owns what.

AI Admin Panel puts all of it in one panel, on a server you control. Every customer and every app gets a clear owner.

Fig. 1.2Your team runs the panel, your customers use their apps, and you choose where each app's AI runs. If an app uses an outside AI service, its requests leave your server. A local model keeps those requests on it.

§ 02

Who uses AI Admin Panel

Table 2.1 · Who uses AI Admin Panel

§ 03

Select a unit to pull it out. Arrow keys move between units.

You decide where your customers' AI runs.

AI Admin Panel installs on your own Linux server. From there you put AI apps live from the catalog, from Git or from a Docker Compose file. The catalog lists 50 apps in version 2.17.0. They range from agents like OpenClaw and Hermes to workflow tools like n8n, chat apps and local models.

You also see how your apps are doing. The panel flags what needs attention, on the server and in each app, and you open an app's logs from there.

Note

The catalog lists what we package. Check each app with your own setup before you sell it.

Every customer gets their own account and sees only their own apps.

Set up plans that decide how many apps and how much capacity each customer gets. Then invite each customer's team. Their portal covers the routine things: checking an app, restarting it, seeing what their plan includes.

If you distribute AI Admin Panel to other providers, the Partner Center issues and tracks their licences.

Note

Plans, roles and resellers inside the panel come with the Web Host edition. Your billing system still sends the invoices.

A list of customers, each with a plan and a status.Open full size
Fig. 3.2Customers, each with their own account and plan.

You decide who may do what, and which AI service your apps use.

Switch on only the optional customer features you are ready to support, such as their own AI key. For apps that support it, you choose the AI route when you deploy. It can be the customer's own key, your metered key with a monthly budget, or a local model. Other apps take an AI key in the deploy form or set their AI provider inside the app.

Caution

If an app uses an outside AI service, its requests leave your server.

The AI models page, listing local models with their status.Open full size
Fig. 3.3AI models: local models on your own server, pulled and managed from the panel.

  1. Install it on your server.

    One command on a supported Linux system, then a setup wizard in the browser.

    Shown as text · Copy it and run it on your server
    curl -fsSL https://get.aiadminpanel.com -o install.sh && bash install.sh

  2. Add your customers.

    Create their accounts and invite their teams. In the Web Host edition, you also give each one a plan.

  3. Put their AI apps live.

    Pick an app from the catalog or deploy from Git. Where the app supports it, choose its AI route.

See how it works

Table 3.5 · What you can check today
ReleaseVersion 2.17.0, released on 18 October 2026, with public documentation.
Catalog50 apps in the catalog, each one named on our templates page.
API and MCPA documented API, and MCP access for your team and, where you allow it, for your customers.

Note

We are early, and we have no customer numbers to show you yet. So we show you the product instead.

§ 04

See the panel

Screens from an earlier build of the panel, with made-up customers.

The services list, with each app's status, type and CPU and memory use.Open full size
Fig. 4.1Every service you can see, with its health and resource use.
One service's metrics tab, with CPU, memory and network charts over the last hour.Open full size
Fig. 4.2Live CPU and memory trends for each service.
One service's overview page, with its ID, type and address, and tabs for logs, metrics and environment.Open full size
Fig. 4.3One service: logs, metrics and environment in one place.
The audit log, a table of timestamped events with actor, action and resource, and filters above it.Open full size
Fig. 4.4The audit log: filter events by actor, action, resource and date, and export them to CSV.
The roles page, listing seven default roles, each with its number of permissions.Open full size
Fig. 4.5Roles and permissions, with sign-in through Keycloak.
The panel's dashboard with service health, alerts and quick actions.Open full size
Fig. 4.6The dashboard: what needs attention, service health and quick actions.
A list of customers, each with a plan and a status.Open full size
Fig. 4.7Customers, each with their own account and plan.
The AI models page, listing local models with their status.Open full size
Fig. 4.8AI models: local models on your own server, pulled and managed from the panel.

§ 05

50 apps in the catalog of version 2.17.0.

  • OpenClaw

    2 cores, 3 GB

    Build and run AI agents that use tools, through a gateway with its own password. AI route chosen at deploy. Review its tool settings inside the app.

  • Hermes Agent

    1 core, 2 GB

    Nous Research's agent, with a web dashboard for multi-step tasks. AI route chosen at deploy. Chat channels connect in its own dashboard.

  • Ollama

    2 cores, 4 GB

    Runs open models such as Llama, Mistral and Gemma on your server. A GPU isn't required, and speed depends on your hardware.

  • vLLM

    4 cores, 8 GB

    An OpenAI-compatible engine for serving models under heavy use. It needs a GPU.

  • LiteLLM Proxy

    1 core, 1 GB

    One OpenAI-compatible endpoint for many AI providers, with keys, budgets and spend tracking. AI route chosen at deploy.

  • Open WebUI

    1 core, 1 GB

    A chat interface for local and OpenAI-compatible models, with several users, document upload and web search. AI route at deploy: a local model.

  • Qdrant

    1 core, 1 GB

    A vector database for semantic and document search, with a web dashboard. Its API key is generated at deploy.

  • n8n

    1 core, 512 MB

    Visual workflow automation with many ready-made integrations. Incoming webhooks work once the app is public.

  • Langflow

    1 core, 2 GB

    A drag-and-drop builder for AI agents and document search. AI providers are set up inside the app.

  • Dify

    2 cores, 6 GB

    A visual builder for chatbots, document search and AI agents. AI providers are set up inside Dify.

23 AI apps, 6 databases, 10 developer tools and 11 web apps.

See all 50 apps

§ 06

Portrait of Lukas Hertig.
Fig. 6.1Lukas Hertig, founder

Why I built it

I spent about twenty years in the hosting industry, working with hosting providers and the people who build on them. Then I started an AI agency. Every customer needed their own separate setup for AI, and I couldn't find a panel that did that. So we built one.

More about us

§ 07

What we're working on

Table 7.1 · What we're working on
No.ChangeStatus
7.1Deploying from private GitHub repositories. Connecting a GitHub account is in preview today.No date given
7.2Restore for more kinds of app.No date given
7.3Connections to billing systems.No date given

We don't give dates. We'll tell you when each one ships.

§ 08

Questions people ask

Do I have to replace cPanel or Plesk?

No. Your hosting panel keeps running your websites and email. AI Admin Panel runs the AI apps, on a separate server you set up for it.

What does the free edition include?

Up to three customers and five services, with the catalog, monitoring, notifications, the customer portal and the audit log. Power User removes those limits and adds custom domains and S3 backup. Web Host adds organisations, roles, projects, resource and service plans, and resellers. You can try either paid edition for 14 days.

Which Linux systems does it run on?

Certified versions of Ubuntu, Debian, Rocky Linux, AlmaLinux, RHEL and CentOS Stream. Our documentation lists each version.

Does our customers' data stay on our server?

The apps and their data run on your server. Some things leave it: requests to an outside AI service, backups to your storage, and emails and notifications. So do licence checks, updates, certificates and DNS, code fetched from Git, model downloads and any chat app you connect. Profile pictures load from Gravatar, using a code made from each user's email address. Each app can also make its own connections. For apps that support it, a local model keeps the AI requests on your server. Our documentation describes the panel's main connections. The licence agreement also lets the panel send us usage statistics, such as counts of customers and services.

Can my customers break things?

In the panel, they get the actions their plan and role allow, and server-level tools stay with your team. Inside each app, that app's own settings apply.

Thirty minutes with me

I'll show you the panel with made-up customers, and you get an honest answer on whether it fits your business.

Book 30 minutes Get started