How to manage AI agents for each client

Last updated: 10 November 2026

The cleanest way to manage AI agents for many clients is to give each client their own dedicated server and build your skills and interface once, then roll them out to every client server. With CRHQ, the agent harness, each client's agents and data stay on that client's server, while you publish shared themes and skills from your own agency server.

This guide shows the setup step by step: how the servers fit together, what each client sees, how updates, backups and AI models work per client, how to report to clients, and how to onboard your first one.

The problem: rebuilding for every client eats your margin

Every client who wants agents needs the same groundwork. They need a server, an agent harness, access to an AI model, backups, sign-in for their people and someone to call when it breaks. If you build that by hand for each client, the second client costs almost as much time as the first, and every fix has to be made again in each place.

The work that clients pay for sits on top of that groundwork: the skills, the schedules and the know-how about their business. The setup below is designed so you build the groundwork pattern once and spend your hours on the part clients value.

The setup: one server per client, plus your own

Here is how the pieces fit on CRHQ.

                 Your agency server (never locked)
     builds and publishes: your Canvas theme, shared skills, agents
                               |
          publish a new version to your organization
                               |
        +----------------------+----------------------+
        |                      |                      |
  Client A server        Client B server        Client C server
  your brand + domain    your brand + domain    your brand + domain
  own agents, data,      own agents, data,      own agents, data,
  schedules, AI access   schedules, AI access   schedules, AI access

Your agency server is where you build and test. You make a theme, a skill or an agent once for your organization, and it rolls out to every server, with new versions arriving automatically. Each client server runs under your brand and your domain and holds that client's own agents, files, memory and schedules. Each server has its own database, file system and browser, and agents on one client's server cannot reach another client's server (only your organization's Fleet Agent works across your servers, at the request of a person in your organization).

There are a few ways to arrange client work, and each has a place.

People and access: what the client sees

When a client signs in, they see your name and logo in the app and on the sign-in page, and the support agents introduce themselves by your name. The client server runs on your domain, and CRHQ sets up the domain with you.

You decide how much of the system each client can touch. With Canvas you can hide any page or Settings section, and you can lock the interface on client servers so clients cannot switch to another one. Your own main server is never locked. One simple pattern is to give the client the pages they use every day, such as a dashboard, their tasks and their reports, and keep settings like models and billing hidden. For the client's own people, use roles instead of admin access, because an admin always sees and manages everything on their server.

If a client's customers will talk to an agent, use a contained agent. It runs each conversation in a sealed space, reads only the knowledge you grant, and cannot reach the rest of the server, including files, other conversations, memory and credentials.

Updates, backups and AI models per client

Updates

CRHQ runs the harness and every update on every client server. Your Canvas theme is maintained by CRHQ, so improvements reach it automatically, and updates never break your pages. When you publish a new version of your own theme or skill, it reaches every client server in your organization.

Backups

CRHQ keeps daily, weekly and monthly database backups on a separate cloud storage provider, plus automatic server snapshots from the hosting provider. Security patches are applied automatically on every server.

Model access

Each server connects to the AI subscriptions you choose: Claude through Claude Code, ChatGPT through Codex, Cursor (which includes Gemini and Grok), Z.ai GLM and open-source models. You can choose the model for each new session and give different agents different models. With more than one connected account, auto-rotation moves work to an account with capacity when five-hour or weekly limits are reached, for Claude, Codex and Z.ai. Subscriptions are billed by the provider, and the provider's plan limits and terms apply.

Monitoring and reporting to the client

CRHQ monitors the health of every server and alerts its own operations team when something is wrong, so your part is watching the work the agents do.

1. Read the agents' conversations. For contained agents on a website, chats waiting for a person show up in that server's settings.
2. Read the agents' conversations, which show each step an agent took on the server.
3. Build a status page in the client's Canvas. A partner's Canvas on crhq.ai shows a live dashboard of agents at work, conversations, scheduled jobs, server health and recent work.
4. Schedule the report. A background job can run weekly, write the report and post it to Slack or to a status board the client opens.

Agents can produce reports, charts and presentations you can open and share, so the monthly report can be a file the client keeps.

Steps to onboard client #1

1. Agree the scope with the client: the jobs the agents will do, who approves their output, and what knowledge the agents may use.
2. Build and test the agents, skills and schedules on your own server first. A 7-day trial costs $19.
3. Build your branded interface with the Theme Builder agent, from your logo or website.
4. Share the theme with your organization, set it as the default and lock it for client servers.
5. Book a call about white label to set up your domain and add a dedicated server for the client.
6. Connect the AI subscription the client's agents will use.
7. Install your shared skills and add the client's own knowledge, such as their brand notes, product facts and past reports.
8. Turn on the schedules, for example a weekly report every Monday.
9. In Settings, open "Users and roles", add the client's people as users with a role instead of admin access, then walk them through what they see.
10. Read the first two weeks of conversations, including each step the agents took, fix the skills, then send the first monthly report.

When client #2 signs, steps 3 and 4 are already done, and step 7 starts from skills that already work.

FAQ

Answers checked on 8 November 2026.

Should each client get their own server?

Yes, for client work. On CRHQ each client gets a dedicated server under your brand and domain, with their own agents and data, and agents on one client's server cannot reach another client's server (only your organization's Fleet Agent works across your servers, at the request of a person in your organization). Settings inside one server control what people see, so separate clients belong on separate servers.

How do agencies keep each client's AI agents and data separate?

They give each client their own server and keep shared work, such as themes and skills, in their own agency organization. Each CRHQ server has its own database, file system and browser. Public-facing agents run as contained agents, which read only the knowledge you grant.

Can a client see or change their own agents?

You decide, through roles. In Settings, under "Users and roles", you add the client's people as users and choose for each role which pages it sees, which agents it may use and whether it may create or edit agents and skills. An admin always sees and manages everything on their server, so give client staff a role instead of admin access. With Canvas you can also hide pages and Settings sections for everyone on a server and lock the interface on client servers. Your own main server is never locked.

How do I roll one skill out to all clients?

Build the skill once for your organization and roll it out to every server. When you publish a new version, it arrives on each client server automatically, so every client runs the same version.

Who pays for the AI model, me or the client?

The AI subscription is billed by the provider, separately from the CRHQ server. We recommend that each client connects their own AI account on their own server, so usage limits and the bill stay with the client. Consumer plans are individual accounts under each provider's terms, so check those terms before using one account for several clients. Contained agents use the same connected account, within the spending caps you set.

Run your first client this month

Try CRHQ on a dedicated server for 7 days for $19, then from $299 a month. Start a 7-day trial or Schedule a demo to set up white label for your clients.

Related: White label · Referral program · Rent out agents · Pricing · AI agent management as a service