Straight answers about agents on computers you own.

Start with the jobs you repeat: the portal you sign into every week, the numbers you copy into a report, the system you update after every sale. Tell me about a few in your own words. Part of checking the fit is working out which are worth handing to an agent and which aren't. What to hand off

No. Describe the work and what gets in the way. I look at your apps, logins and computers and turn that into a plan.

Yes, and it's often the best way to start. Get one job working, then add more.

Someone who can sign in to the accounts the agent will use, for the steps that need a login or a code. We agree who and when before it starts. The setup itself usually happens remotely; it can also be on-site, or I can ship the computers ready to plug in. How it goes

It depends on the jobs, the apps and the computers. A call, or one job on a computer you already have, is the smallest project; more jobs, new computers or an on-site setup take longer. We agree the scope, the price and the schedule together before anything starts, so you know what to expect.

Jobs you repeat in a browser or a desktop app, with a result you can check. If existing software or a simple automation would do it better, I'll say so. What to hand off

Jobs a simple automation or the software you already have can do more reliably, decisions that need your judgment each time, anything that has to run entirely offline, and desktop programs that don't check out when I look at them. I'll tell you plainly, usually in the first conversation.

Websites and web apps are the most straightforward fit. Desktop programs can work too: I check each one, with its logins and two-factor steps, before we agree on anything, and I'll tell you plainly if something isn't a good fit. For the agent, anything that can use MCP tools works, such as Claude Code, Codex, opencode or omp. Your agent

Often. I check each program first, with its logins, two-factor steps and licences, before we agree on anything.

Yes, planned account by account. During setup, if a job needs two-factor codes, I ask you for access to them. If you'd rather type each code in yourself, you can, but then you'll need to be involved in every step I do. Once it's running, I plan each login and code with you, and you can always take over for a step.

Sometimes, and I'll tell you. If your apps already connect to each other, or a simple automation can do the job reliably, that's usually the better choice. Agents earn their place where there's no connection: supplier portals, older apps and work that only happens on screen.

Often, that's the place to start. I check whether your hardware can handle the work before you buy anything. The computers

ibara is built for Omarchy, on the computer your agent uses and the one you watch and take over from. Mac and Windows are in progress and a top priority. On a Mac or Windows? Tell me anyway, and you'll be first in line. For your agent, the old computer in your drawer may be enough. I set it up and show you how to watch an agent, take over and stop it. The computers

No. The agent's computer runs the browser, your apps and ibara, while the AI model runs with your agent's provider, so basic hardware can be enough. ibara is tested on five-year-old Celeron mini PCs. I check what your work needs before you buy anything.

So your own computer stays yours. The agent gets its own screen, its own browser and only the accounts its jobs need, and you can watch that screen any time without getting in its way.

Not on the agent's computer. It runs the browser and apps, and the model runs with your agent's provider. That's why basic hardware can be enough, and why your agent's provider still sees what it needs to do the job. If you want the AI to be yours too, your agent can run on a local model, on a computer that can handle it.

Wherever suits you: a shelf in the office or at home, on your own connection. Websites see an ordinary computer on a home or office connection, not a data center.

That's up to you. Give your agent as much access as you want, or as little. You choose what each one can use and what it has to ask you about first, and you can watch or take over any time. I'll also tell you what should never be connected. Staying in control

Yes. From your own computer, open any agent's screen and watch it work. Watching never takes control or interrupts it.

Take over, and the agent pauses before your input starts. You can also pause every agent at once, or remove an agent's access to a computer. I show you each of these at handover.

By default, steps that send, spend, delete or change access wait for your OK, and each OK covers one step. We agree together what else should ask first. It's a guard-rail for the steps the software can recognise, not a guarantee, which is why the agent's reach stays small too.

Anything you couldn't live with an agent getting wrong, such as payment accounts with no second check, or admin access to systems a job doesn't need. I go through your accounts with you and say plainly which to leave out.

Agents do make mistakes. That's why each one works on its own computer, reaches only the accounts its jobs need, and asks before the steps that matter. Each job ends with checks that pass or fail, so you can see what it did. If something goes wrong, you can take over, and if I'm keeping it running for you, fixing it is part of that.

The work happens on your computers and your network. The AI model your agent uses still sees what it needs to do the job, such as screenshots and task details, unless your agent runs on a local model. If everything must stay offline, that needs its own assessment.

I do the setup myself, so I see the screens and accounts the jobs involve while setting up and testing them. We agree up front what access the setup needs.

Only if you want me to, for example so I can keep it running for you. Otherwise the access is yours alone.

Yes. I'm happy to sign one before looking at anything sensitive.

Your agent uses a real browser on an ordinary computer and your own connection, so websites see an ordinary computer, not a data center. Sites keep their own rules, and your agent should follow them.

It depends on the work, your computers and how much help you want. A call, a setup and keeping it running are each priced per project, and we agree on scope and price before starting. Hardware, your agent's AI plan and any ongoing help are explained separately, so you know what you're paying for. What it costs

Per project, with optional ongoing help to keep it running. Scope and price are agreed before anything starts.

Not for the computers: they're yours, so there's no computer rent. Keeping it running is optional, and your agent's AI plan is billed by its provider.

Usually. Your agent's AI plan is separate and billed by its provider, and many owners already pay for one. I'll help you choose one that suits the work. Your agent can also run on a local model, on a computer that can handle it.

Only if you want to. You'll know how to run it, or I can keep it running: updates, fixes and changes. What's covered is agreed up front. Support and what you keep

Agents work from what's on the screen, so a small change to a page often doesn't matter. When a change does break a job, it needs fixing: if I'm keeping it running, that's my job; without it, your written instructions show you what to check.

Everything: the computers, the accounts, the admin access, and the written setup and instructions. ibara is free and open source, and nothing is switched off when support ends.

Yes. Start with one job, and add more once it's working, as a new project or as part of keeping it running.

Because you'd be renting. Your agent would work on their computer, in their cloud. With ibara you own everything: the computers, the accounts, the setup and the software. Want the AI to be yours too? Your agent can run on a local model, on a computer that can handle it. Cirlet and the alternatives

Not on a bare computer. Before your agent can even reach it, the computer needs Omarchy and a way in, and then ibara and Tailscale. That first stretch is where people get stuck, and it's what I do. After that, your agent can handle a lot of the rest. Built for Omarchy

Yes. ibara is free and open source. Requirements and documentation are at ibara.app. The hard part is the first stretch: before an agent can reach a new computer, it needs Omarchy and a way in, at minimum SSH, then ibara and Tailscale. That's where people get stuck, and it's what I do.

ibara is the free, open-source software that lets AI agents use computers you own. Cirlet is how you hire me, its creator: a call, a setup, or keeping it running. About Cirlet

Me, Tyler Mayberry. I built ibara, and you work with me directly, from the first conversation to the handover.

Not the build-anything kind. I don't sell custom automations or a retainer. I set up AI agents on computers you own, hand everything over to you, and keep it running if you want. If a simple automation would do the job better, I'll tell you. Cirlet and the alternatives