The employee keeps the role
Name, instructions, skills, conversations and backlog live in Tidy, so they move when the machine changes.
04 / Virtual employees
Give it a role, choose where it works and decide how it uses AI. Watch the whole setup, from adding a workshop to Penny starting as project coordinator.
The real product, narrated. Waits are trimmed, credentials hidden, people fictional.
Here's how to set up a virtual employee in Tidy. We'll create a workshop, give Penny a role, and deploy her.
A workshop is the computer where an employee runs. Use a machine you already have, or choose a supported cloud host. For this demo, we're using an isolated Linux machine.
Give it a name and choose its operating system. Then run the setup command on that computer. The command contains an enrollment credential, so we've hidden it here.
The installer checks the machine and starts the workshop. When it reports online, we can continue. We've cut out the installation wait.
Now create Penny. Give her a name, a role and instructions. She coordinates our store opening, checks dates and owners, and asks before contacting people or changing deadlines.
An employee can run on Tidy, with no machine to manage. Choose a workshop when the work needs local files and tools. We're assigning Penny to the workshop we just set up.
Next, choose a harness. That's the software that runs the employee. We're using Pi. Other supported choices include Claude Code and Codex.
The model connection is a separate choice. With Tidy inference, Tidy supplies the models and manages access. You don't need to bring a provider key.
With provider sign-in, you authorize a supported account through its login flow. OAuth is different from an API key. What you can use depends on the provider, the harness and your plan. Your provider's limits still apply.
Use the sign-in controls for accounts and keys. Don't put passwords in an employee's instructions. These model choices don't change what the employee can access in your workspace.
Choose a project folder and deploy. Tidy asks us to approve the harness installation. We confirm our identity before it runs.
Tidy installs the harness, applies the employee's configuration and checks the result. We've shortened the waiting time here.
Penny is now deployed. Start with a small task and review what comes back. Her identity and work stay in Tidy if you move her to another workshop.
Recorded on a Linux test machine. Your own computer or server follows the same steps.
The two things you're creating
Penny is the employee. The workshop is the computer where Penny runs.
Name, instructions, skills, conversations and backlog live in Tidy, so they move when the machine changes.
Files, tools and somewhere to run. Several employees can share one.
"Run on Tidy" hosts it for you. Use your own workshop when the job needs your files, a browser or local tools.
The setup
You can revise the employee's setup later. Start with a narrow job and only the access it needs.
Name the machine and pick its operating system. Tidy gives you one setup command. Treat it like a password.
Run the command on that computer. The workshop shows up in Tidy when it reports in.
Name, role, instructions. Here Penny coordinates a store opening and asks before contacting anyone or moving a deadline.
The harness is the software that runs the employee: Pi, Claude Code, Codex and others. The model connection is a separate choice.
Put the employee on the workshop, finish any sign-in Tidy asks for, and check it reports ready.
How it gets its AI
Three ways to reach a model. None of them changes what the employee can see or do.
Options depend on the harness. The walkthrough uses Pi with Tidy inference.
Start with a useful task
Point it at the project and say what you want back. Review the draft, then sharpen the instructions.
Only the records and tools the job needs.
Anything sensitive or outward-facing waits for a person.
Read its conversations and check its work.
Virtual employees need Plus. Model and hosting costs depend on your setup.
See what you can work on together ↗