Your own AI superpower.A whole team's work, in one system you control.
Marketing. Programming. Design. Analytics. Systems administration.
One idea often needs several specialists. With your own system of AI agents, you can handle many of those tasks yourself: describe the result you want, direct the work and review what's done.
It doesn't just give advice. It does the work and builds things.
We install Cardloop on your server, set up the agents and the connections you need, and teach you to use it in person.
Setup, configuration and training: $1,999
On a computer in your office or a remote server, reachable from your computer and phone. On-site across the Sacramento area, remote anywhere in the US.
The whole path from idea to result changes.
Create something new
Say you want to launch a new service. You need to study competitors, shape the offer, build a page, prepare the launch materials and connect the tools behind it. The research becomes the basis of the offer. The offer becomes the page and the materials.
- Research
- Offer
- Page
- Materials
- Tools
Run your daily work
Every day, agents check your mail, work through supplier and government portals, update your spreadsheets and prepare documents for your signature.
- Portals
- Spreadsheets
- Documents
- Your OK
You set the direction, review the results and make the decisions. Agents do the work at every step.
The skills of different specialists. At your disposal.
Marketer and content creator.
Research your audience and competitors, develop versions of your offer, prepare ad copy, posts and video scripts.
Developer and web developer.
Build or change a website, write code, put together a calculator, an internal tool or an automation for your task.
Designer.
Develop a visual concept and prepare images, layouts and materials for a website, a presentation or ads.
Analyst and researcher.
Gather information from different sources, compare options, work through spreadsheets and prepare conclusions.
Systems administrator and integrator.
Help with technical settings, connecting services, deploying applications and automating processes.
Documents and operations assistant.
Sort files, pull out data from invoices and contracts, reconcile spreadsheets, prepare documents and organize recurring office tasks.
These six are only examples.
Your system can grow a whole team of its own, with as many agents and roles as your work needs.
We start from your work and set up the capabilities it needs. Important results need your review, and some tasks need a licensed professional.
How agents do the work
- A live browser
Opens sites and portals, signs in with you, fills in forms and applications while you watch.
- Your email
Reads what comes in, drafts replies and sends the ones you approve.
- Your programs
Connects through APIs to almost any system, from spreadsheets to QuickBooks and your CRM.
- Its own code
Writes the scripts, pages and tools a task needs, and runs them on your server.
All of it runs in Cardloop, a workspace for AI agents and their tasks: Claude, OpenAI Codex with your ChatGPT plan, Grok and local models in one place, with a task board and shared memory, from your computer or phone.
Everything inside your system
The full list of what Cardloop can do. Open a group to see it.
AI agents and models
- Three agent engines: Claude Code (the default), OpenAI Codex with your ChatGPT plan, and xAI Grok Build. Pick one for each chat or task.
- Local models (through Ollama) run the whole agent loop on your own machine.
- Move a conversation to another engine and the context goes with it.
- A second opinion from another model family, such as Gemini, in the middle of a task.
- Deep mode: maximum reasoning, with the work split across sub-agents that check each other.
- Eleven sub-agent roles: architect, executor, debugger, researcher, three reviewers, skeptic, test writer, docs writer and a quick helper. You can edit them.
Tasks and the board
- A task board in every project. Move a card to In progress and an agent works on it to the end.
- Each card runs in its own copy of the project. The result waits for Check (runs the tests), Apply or Discard.
- Plan mode: the agent only reads until you approve its plan. Ask mode: approve each step from your phone.
- Project autopilot, if you want it: it notices what needs doing, proposes the work and does it after your OK.
- Hit a usage limit and the task pauses, then continues on its own when the limit resets.
- Background work while you're away, waiting in the chat with a notification.
- Self-repair, if you want it: an error in the logs becomes a repair card with a check.
- Finished cards are archived only on objective evidence, and a specs tab tracks larger pieces of work.
- Every scheduled job on the server (cron, systemd, n8n and agent jobs) in one tab.
- House rules for agents, written in plain text and checked before every action.
Memory and search
- Project memory: lessons from past work are saved and read back in the next session.
- A context pack at the start of each new session, so agents don't start from zero.
- One search box (Ctrl+K) across every chat, board, file and memory note in all your projects.
- A daily journal of what was done across all projects.
- A health check for every project and an editor for the rules each project's agents follow.
The workspace
- A live chat with the agent: every step and file change shown as it happens, a model switch on the fly, a prompt library, skills, a message queue and a stop button.
- Roll your files back to how they were before any message.
- Choice buttons, pictures, video and files right in the chat.
- A live browser, built in or your own Chrome: saved sign-ins, view-only by default, and you can control it together from your phone.
- A terminal, a file browser for the whole server, logs, a timeline, a spending tab with costs per model and an encrypted password vault.
- Many projects in folders, each with its own engine and model.
Phone and access
- An Android app and an installable web app.
- A notification on your phone when work is done.
- Access through your own domain (Cloudflare Tunnel), a private network (Tailscale) or a quick temporary link.
- Sign-in with a password plus a code from your phone.
Running safely
- Always on, as a service on your server.
- Updates itself, checks itself after the restart and rolls back on its own if something breaks.
- A load meter for the server.
- The workspace's own passwords are kept out of the agents' processes, and a short list of irreversible server commands is blocked.
- 5,400+ automated tests and the OpenSSF Best Practices badge (passing).
Connected on top, for your tasks
- Email: Gmail and your company mailbox. Read, search, and send with your OK.
- Google Drive and Canva.
- A scheduler on your own server for posts to your business pages.
- Image and video generation.
- Domains and DNS on Cloudflare.
- Around 80 skills: design, video and animation, n8n workflows, research, test-driven development, code review, websites, a landing page from an Instagram account and a knowledge base from a book.
- Our own Cardloop video was made inside Cardloop: script, screen recording, voice and edit.
We take on all the launch work.
For agents to do your tasks, the working environment has to be prepared and its parts connected.
That installation and configuration work is what our $1,999 service covers.
- 01
Server preparation.
We help you choose a suitable office computer or remote server, then prepare it: system updates, the software Cardloop needs, an encrypted vault for passwords and keys, and an automatic restart after a power cut. Agents get a machine of their own, separate from your work computer.
- 02
Cardloop and AI agent installation.
We install Cardloop, configure the agent providers you choose and connect your accounts.
- 03
Browser, tools and access.
We set up the live browser the agents use, your own address (for example ai.yourcompany.com) behind a password and a code from your phone, and the phone app.
- 04
Connections to your services.
We connect the services agreed for your setup and configure the permissions and credentials your first workflows need.
- 05
Testing with your actual work.
We run through the agreed workflows, troubleshoot the setup and check that the pieces work together.
- 06
Personal training and handover.
We walk you through the system, explain how to operate it and help you start using it for real tasks.
Learn to work with the skills of a whole team.
We show you how to turn an idea into a clear task for agents, give them the materials they need, guide the work and judge the result. Training runs on your own tasks. As you get comfortable, you can grow from single tasks to your own workflows.
Professional installation, configuration and training
$1,999You pay for our work: installing, configuring and testing your own AI agent system on your server, and training you to use it.
Includes
- server preparation
- Cardloop installation
- agent configuration
- remote access
- agreed service connections
- first workflow testing
- personal training
Before starting, we agree on the setup scope, the services to connect and the workflows to configure.
Hardware or hosting and any paid AI plans, API usage or third-party services are separate. Custom development and ongoing support are quoted separately.
Questions about AI agent setup
What is AI agent setup and training?
It is a service: we install your own AI agent system on a server you own, connect it to your services and teach you to use it. The system is Cardloop, an open-source workspace built by our team, running Claude, OpenAI Codex, Grok or local models. Setup, configuration and training cost $1,999.
How much does it cost to set up AI agents for a small business?
$1,999 for setup, configuration and training, a fixed price for the scope we agree before starting. Hardware or hosting, AI plans and API usage are paid separately, on your own accounts. Ongoing support is optional and quoted separately.
What can AI agents do for my business?
They do the work instead of only answering questions. Agents research, write, build web pages and internal tools, work through supplier and government portals in a live browser, handle email, sort documents and connect to your programs through APIs. You set the task, review the result and approve anything that goes out.
Can it replace my accountant or lawyer?
No. Agents draft documents, do the research and keep the paperwork moving: they can keep your books in order, prepare tax paperwork, draft contracts and letters, write ad copy and posts, and build your website and tools. Legal and tax advice and signatures stay with a licensed professional. You direct the work and decide what to use, and the responsibility for the results is yours.
Who is responsible if an agent makes a mistake?
You are. The system runs on your server, under your accounts and at your direction, so you review important results and approve what goes out. We are responsible for the work we agreed to do: the system installed, connected and passing the workflows we tested together.
Which AI models does it use? Can I use my ChatGPT plan?
Claude by default, OpenAI Codex through your ChatGPT plan, Grok, and local models that run on your own machine. You can use one or several and pick the engine for each task. Each runs on your own account with that provider.
What is Cardloop?
Cardloop is an open-source workspace for AI agents, built by our team and published under the MIT license. Each task is a card on a board: an agent works on it, and the result waits for your review. It runs on your server and you control it from your computer or phone. Our team runs its own work on it every day.
Cardloop is free. What am I paying for?
Our work around it: preparing your server, installing and configuring the agents, the browser and access, connecting your services, testing on your real tasks and training you in person. You can install Cardloop yourself from GitHub, or hand the whole job to us.
Do I need technical experience?
No. We handle the installation and configuration, then teach you how to use the system. You'll still need to give direction and review important results.
Do I need to buy a server?
The system needs its own machine: a small computer in your office or a hosted server. We help you choose an option based on your needs before you buy anything.
Can it work with my existing tools?
Agents can work with services through available integrations, APIs and a live browser, so almost any program or website you use is within reach. We review your tools and confirm the first connections before starting.
Does everything stay on my server?
Cardloop runs on your server. When you use cloud AI providers or connected services, the data a task needs is sent to them for processing. Local models keep the processing on your own machine. We explain the data flow for your setup.
Do you set it up outside Sacramento?
Yes. We set up on-site across the Sacramento area, including Roseville, Folsom, Elk Grove and Davis, and remotely anywhere in the US.
What happens after setup?
You use and manage your system with the training provided. If you want help maintaining it, adding workflows or learning more, ongoing support is available separately.
What would you like to hand off, or finally create?
Tell us about your work, the tools you use and what you want to accomplish. We'll help you plan the system, install it on your server and teach you how to put it to work.