PromptPartner DE

Custom and open-source agents, on a server you control

We build agents for one job each and run open-source agents on your own server. You choose the model for every job, and you decide what may leave the house.

Your server: agentsIllustration with sample data
  1. Quote agentlocal model
  2. Order intake agentyour own key
  3. Assistant, sales teamcloud model
  4. New skill: PDF readerchecked, waiting for approval

Your IT admin approves

For firms that want to keep their AI in their own hands

  • ControlYou decide what runs on the server, who may open it and what it may change.

  • A second wayThe job is kept apart from the platform and the model, so it can move when one cloud stops.

  • Secret filesWhat must stay in the house goes only to a model that runs under your firm's name.

Agents built for one job

A custom agent takes one job off your desk: a quote, an order confirmation, a service report. It pulls data from several of your systems, works by your rules and hands the draft to one of your people to approve.

  • ERP
  • Mailbox
  • Files
  • CRM

Your quote agent

Works by your rules

skillsplugins

Draft: quoteExample

Your estimator approves itNothing goes out without a yes

Set up before the first run

  • AccessOnly the folders and systems you allow.
  • GuardrailsRules that set what it may do.
  • LogWhat it did, written down.
The agent works between your systems with your rules, and your estimator approves the draft.

Open-source agents, run on your own server

Some jobs suit an open-source agent better than a new build. We run it on your server, where from the start only people who sign in to your server can open it. Each agent runs in containers of its own.

You choose the model for each job.

  • A local modelThe requests stay on your server. On ordinary hardware it's slower, and it's the place for secret files.

  • Your own keyThe agent uses your account with the model provider you pick, and the provider bills you directly.

  • A cloud modelFor public and internal work that needs a large model.

Your server runs on AI Admin Panel, software we build ourselves. AI Admin Panel is for companies that must run AI on their own servers. For everyone else, we use the tools that fit, often Claude or ChatGPT Enterprise.

A personal assistant for each person who needs one

It sorts the mail, prepares the day in the calendar, finds files, keeps track of tasks and works in the chat apps your team uses. We build it on the open-source OpenClaw or Hermes Agent, and always with four safeguards.

  • Each person gets their own instance, on your server.
  • We check every skill before it is installed.
  • Nothing leaves the house until a person approves it.
  • You choose the model.
The personal assistant

OpenClaw and Hermes Agent are open source under the MIT licence (github.com/openclaw/openclaw, github.com/NousResearch/hermes-agent).

A woman works at a laptop at her desk, a small glowing orbit symbol beside her

Risky changes wait for your IT admin's yes

When an agent wants to delete or overwrite data on your server, the call waits for an admin. The server's own assistant works the same way, one change per approval.

  1. The agent asks

    It wants to delete or overwrite data.

  2. The call waits

    It stops on the approvals page.

  3. Your IT decides

    An admin approves it, or it lapses after an hour.

  4. It goes in the log

    Who did what, to which app, and when.

Facts about the server software from aiadminpanel.com (home page, Security, Managed AI agents), read on 15 November 2026.

Data and security

If you stop, the agents, the rules and the data stay with you

Everything we build sits on your server from the first day: the agents, the skills and plugins that carry your rules, and your data. There's nothing on our side to hand over, and backups go to your server or to storage you choose.

And when a cloud goes down, as Microsoft 365 did on 12 November 2026, a job that's kept apart from the platform can move to your server.

Source for the outage: The Register, 12 November 2026.

Where your agents run

Questions about agents on your own server

Do we need a server room of our own?

No. The server can stand in your building or with a host you choose, and if you want, we look after it every month.

Can an assistant on OpenClaw or Hermes Agent work with our mail?

Yes, and we always set it up with four safeguards. Each person gets their own instance, on your server. We check every skill before it is installed. Nothing leaves the house until a person approves it, and you choose the model.

Which model does the agent use?

The one you choose for each job: a local model on your server, your own key with a provider you pick, or a cloud model. If you change the model later, the job and its rules stay the same.

What happens to our agents if we stop working with you?

They stay with you, on your server: the agents, the skills and plugins with your rules, and your data. There's nothing on our side to hand over.

Who sees what an agent did?

Your IT. Risky calls wait for an admin's approval, and the log shows each recorded change as a plain sentence: who did what, to which app, and when.

Which files must never leave your house?

Tell us in 30 minutes. We'll say whether a server of your own makes sense for you, or whether the tools you have will do.

30 minutes with Lukas. You leave with at least one recommendation you can use, whether or not we ever work together.