PYE Agent Curator

PYE Agent Curator

You describe it. Curator builds the robot.

A Windows application. You write what you need in plain language. The artificial intelligence builds the automation, opens the browser and starts working — on your company’s machine.

Build

Builds

It edits the robot’s code. You approve each change before it sticks.

Plan

Plans

It studies the project and proposes a path — without touching a file yet.

Chat

Explains

It answers questions about the robot. Nothing is changed.

The conversation becomes the robot.

In the Agent tab you write what you want. The conversation is created on its own. In the footer you pick Build, Plan or Chat, turn on AI voice if you want to listen, and switch the model without leaving the screen.

The queue is the central list of cases.

A spreadsheet, a system, a list. The Dispatcher (the program that puts items on the list) reads the source and creates one item per case. A Performer on each PC (the program that picks and works) reserves one item — two machines never take the same case.

  • The lock comes first: the item is reserved for that PC.
  • If it fails, only that item returns to the queue and retries.
  • Pending, in progress, successes and failures match what the real screen shows.

Hover to pause. Click to watch again.

Each case has a history.

Priority, deadline, attempts and the time it started and ended. If it fails, you reprocess only that item — not the whole spreadsheet.

The interpreter is the reading recipe.

The document goes in. Each step has its own instruction and model. The next step reads what the previous one found — Type, Authority, District — and at the end the result is assembled for the robot to write into the system.

  • Each card is a step: number, model and the field that came out.
  • One step’s value moves to the next — as on the Curator screen.
  • Switching one step’s model does not rebuild the rest of the recipe.

Hover to pause. Click to watch again.

The drawing is the process.

Each block is a step. The arrow shows what is passed on. Analyst and developer look at the same recipe.

The benchmark answers: which model to use?

You build a batch of real cases with the right answer. Curator runs the same batch on several models and shows three numbers: accuracy (how many were right), cost and speed. The choice stops being an opinion.

Several machines, several robots.

Each Windows PC installs Curator and can run a different robot at the same time: memos at the front desk, invoices in finance, records in legal. It is not the same queue on three screens — the processes do not mix.

  • If one PC drops, only that robot waits. The others keep going.
  • A spare PC takes over the stopped process — you do not rewrite the program.
  • More volume on one robot: plug in another machine.

Hover to pause. Click to watch again.

The model is your choice.

DeepSeek, OpenAI, Anthropic, Google Gemini, Kimi, CompactifAI, xAI and Z AI in the same menu. The key stays on your machine: Curator calls the provider on your account, not ours.

See the providers screen in Curator

PYE Agent Curator

Try it on your process, on your machine.

60 days free. One Windows installer. The run happens on your machine.