Managed AI agents

Run an AI agent for each customer, private by default.

Your customers want AI agents that work for their team.

Set up in a hurry, an agent can sit on an open port, holding the customer's keys.

AI Admin Panel deploys catalog agents for each customer, on your servers, behind your panel sign-in.

Public install opens soon. Leave your email address, and we'll tell you when it does.

§ 01 · Each customer apart

Give each customer their own agent, private from the start.

A customer asks for an agent.

Set up by hand, one customer's agent can end up beside another customer's data.

Pick OpenClaw, Hermes Agent or another agent from the catalog. It starts private, behind your panel sign-in, on that customer's own network. Its passwords and AI keys sit in that customer's vault compartment.

An OpenClaw agent's service page, marked Private, with buttons to open the chat, restart and stop.Open full size
Fig. 1.1An OpenClaw agent on its service page: private behind your panel sign-in until you make it public.

§ 02 · Own sign-in

Customers open their agents with their own sign-in.

The customer's team wants the agent every day.

If every login and restart goes through you, the agent becomes one more ticket queue.

Invite the customer's team to their own portal. They see only their own apps, open each agent from there and restart it where you allow it.

The service page of a Hermes Agent app, marked private behind the panel sign-in.Open full size
Fig. 2.1A Hermes Agent service: its dashboard opens only for people signed in to the panel.

§ 03 · Day to day

Know when an agent needs you.

An agent stops answering on a Monday morning.

The customer waits while someone works out which server it runs on.

The panel flags what needs attention in each app and opens its logs. Alerts reach your team by email, Slack, Discord or Telegram.

Autofix restarts an unhealthy app up to three times, then alerts your team. You switch it on or off.

§ 04

Agents and agent builders in the catalog

  • OpenClaw

    Tool-using agents with a gateway API, in a secure profile.

  • Hermes Agent

    An agent runtime with its own web dashboard.

  • OpenHands

    A coding agent that writes and tests code in its container.

  • DeerFlow

    Plans long tasks, researches and writes reports.

  • n8n

    Visual workflow automation with many integrations.

  • Dify

    A visual builder for chatbots, RAG and agents.

See every app in the catalog

§ 05

Good to know

  • The catalog lists what we package. Test each agent before you sell it.
  • Some agents also ask for their own login.
  • Hermes Agent and OpenHands always stay behind your panel sign-in. Publishing an agent can't be undone yet.
  • Requests to an outside AI service leave your server.
  • The panel and the server itself remain trusted.

All known limits in version 2.17.0

§ 06

Questions about agents

Can an agent answer on WhatsApp or Telegram?

OpenClaw links one WhatsApp number by QR code, with one allowed sender and groups off. It also takes a Telegram bot token. Hermes Agent connects channels in its own dashboard.

Which AI does an agent use?

For agents that support it, you choose the route at deploy: the customer's key, your metered key or a local model. Others take a key in the deploy form or inside the app. About managed private AI

Can an agent reach the server it runs on?

Each agent runs in containers of its own, under a privilege baseline. They reach the server itself only on web ports and DNS, plus ports you list.

Thirty minutes with me

I'll show you the panel with made-up customers, from an agent's deploy to the customer's first sign-in. You get an honest answer on whether it fits your business.