I built an AI bot that gets its own private Linux computer to do tasks for you

Hi everyone! Sharing a project our small team has been building: POSITROM AI (disclosure: I’m on the team).

Most assistants stop at text. Ours runs an agent loop on a private cloud Linux computer per user: plan → act on the desktop/terminal → verify the file → report back.

Examples:

  • “Read these 3 sites and write me a Word report” → it browses, writes and saves the .docx, then checks the file exists.
  • “Write a script to sort and rename my files” → it writes it, runs it, shows the output.
  • “Draft a 10-chapter ebook” → generated chapter by chapter into a document.

You watch everything live in the Computer tab (a real desktop in the browser), the Terminal tab or the Files tab. Each machine is isolated (5 GB own storage); you only ever see your own stuff. There’s also an Android app to give orders by chat, voice or attachments.

What we learned building it:

  1. Long outputs need chunked writing + retries, or the agent truncates.
  2. Showing the live desktop builds far more trust than a log.
  3. Keep the input box pinned and the header collapsible — the chat needs the screen.

Feedback I’m looking for: what task would you hand it first? What would stop you from trusting it?

Pricing, in one line: Chat €12/mo (free trial), Bot with Linux computer €39/mo.