Video summary

Компьютерный Рой

Main summary

Key takeaways

Technology

Technological concept

  • The video demonstrates a “computer swarm”: multiple machines (Windows + Linux) coordinated through AI (Codex CLI) and voice commands.
  • The swarm includes an orchestrator-like router setup that decides where tasks should run (e.g., sends tasks to Linux if it handles them better, otherwise Windows).

System components and interactions

Devices in the swarm

  • Windows computer
  • Linux computer
  • Router
  • Two phones
  • TV used to display results

Communication flow

  • User tasks are listed via a chatbot.
  • The system prepares outputs on the computer (e.g., generates HTML and packages interactions).
  • Requests are sent to GPT/Cloud for AI understanding, while the computer swarm handles generating and routing the actual outputs to other devices (like the TV).

Product features shown (prototype behavior)

Task dispatch / orchestration

  • The user asks to display tasks (e.g., from Jira).
  • The orchestrator assigns tasks to the best-suited environment (Linux vs Windows).

Live result presentation on TV

  • The TV shows an up-to-date list of unfinished Jira tasks.
  • The system highlights that the TV displays not just plain text, but rendered HTML produced by the system.

Windows screen broadcasting to TV

  • The system can execute a prebuilt command to start screen broadcasting on Windows and then switch the TV to show the broadcast.
  • Supports multiple “tabs” / contexts (demonstrated as conceptually more than one).

Voice control pipeline

  • Voice is captured via a phone + microphone.
  • Xiaomi glasses are mentioned as a possible alternative, but with the note that glasses pick up voice worse.
  • Commands are “packed” and sent to the swarm for execution.

“Conveyor” (CI/admin panel) integration explained

The video describes the end-to-end control loop between the computer swarm and a pipeline/admin panel:

  1. Input: “Roy” takes voice and converts ordinary-language task requests into commands.
  2. Commands are sent via Linux to a Windows executable, which runs scripts / automation.

Admin / pipeline panel workflow (high-level)

  • Take a Jira ticket
  • Analyze it
  • Create a branch
  • Apply changes
  • Run checks
  • Generate a PR
  • Build and post to a testing wall
  • Wait for human approval

Additional capabilities

  • Show current tasks
  • Start/stop processes
  • Display results on TV
  • Provide spoken feedback

Review / analysis angle (prototype limitations and roadmap)

  • The speaker notes the current system is between a prototype and an MVP.

Limitations mentioned

  • No fine-tuning / specialized skills yet, so it doesn’t fully “know” domain-specific details (e.g., specific Jira/technical player information).

Planned improvements

  • Improve through training and better specification over time.

UX/behavior goals

  • Move from earlier “send text then silence” behavior to continuous/meaningful feedback during execution.

Security / controls planned for production-like operation

  • Need a direct secure IP adapter plus human confirmation for important actions.

Mentioned capabilities not fully realized yet

  • A broader “Jarvis-like” assistant experience for controlling tools from anywhere (walking/sitting/driving scenarios are joked about).
  • Broad control is claimed as possible, but emphasized as requiring describing commands and stabilizing the system.

Main speakers / sources

  • Single main speaker: the demonstrator (speaking throughout; refers to the system as “Roy” and mentions messages from “Kiryon/Kieron/Kira” displayed on the TV).

Original video