Skip to content

AI Delegation Grid: scalable multi-agent orchestration for Copilot and LLMs inside VS Code. Parallel tasks, adapters, automation.

License

Notifications You must be signed in to change notification settings

adamerso/adg-parallels

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

43 Commits
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

๐Ÿฅš ADG-Parallels v0.5.0

AI Delegation Grid โ€” Because one AI is never enough.


๐ŸŽ‰ POC COMPLETE!

Version 0.5.0 marks the completion of the Proof of Concept phase.

The extension is now fully functional in pre-production and successfully demonstrates its core vision: coordinating multiple AI agents (ejajki) working in parallel on distributed tasks.

โš ๏ธ End of v0.x Line: This version concludes the rapid prototyping phase. The next major version (v1.0.0) will bring a complete codebase refactoring with improved architecture, better error handling, and production-ready quality โ€” while preserving all the proven techniques and patterns developed here.


๐Ÿš€ What is ADG-Parallels?

ADG-Parallels is a fully functional VS Code extension that transforms your editor into a distributed AI workforce with hierarchy, automation, self-healing, and file-based communication.

Think of it as:

  • Kubernetes, but for Copilot sessions
  • A tiny AI startup inside your VS Code windows
  • A company where you are the CEO and all employees are polite little AI agents
    (internally known as Ejajki ๐Ÿฅš)

The extension:

  • Coordinates multiple VS Code windows running AI sessions
  • Delegates work using XML-based task queues
  • Monitors workers via heartbeat system
  • Auto-restarts frozen or unresponsive workers
  • Supports hierarchical delegation (Manager โ†’ TeamLeader โ†’ Worker)

โœจ Key Features (v0.5.0)

โœ… Working & Tested

  • Project Wizard โ€” 4-step interactive project setup with layer configuration
  • Multi-Ejajka Spawning โ€” Automatically opens N VS Code windows as workers
  • XML Task Queue โ€” All tasks managed in tasks.xml with atomic updates
  • Direct Task Mapping โ€” Form fields map directly to task XML (no abstraction layers)
  • Heartbeat System โ€” Workers send "I'm alive" pings every 30 seconds
  • Health Monitoring โ€” Detects frozen workers and restarts them
  • Hierarchical Roles โ€” CEO (human), Manager, TeamLeader, Worker (up to 99 layers)
  • Sidebar UI โ€” Real-time monitoring panel with controls
  • Continuation Prompts โ€” "Poganiacz" to nudge stuck ejajki
  • Start Processing Button โ€” Clean separation between project creation and worker launch
  • Upward Reporting โ€” Workers report status to managers via XML files

๐Ÿ“‹ Field Mapping (Form โ†’ XML)

Form Field XML Tag
Project Name <project_name>
Layers Count <layers_count>
Task Description <your_assigned_task>
Output Directory <move_completed_task_artifact_to>
Input Description <resources_description>
Input Files <list_of_additional_resources>
Continuation Prompt <continuation_prompt>
Reporting <reporting_instructions>

๐Ÿง  Roles in This AI Corporation

Role Emoji Model Description
CEO ๐Ÿง‘ Human You. Supreme overlord. Defines goals.
Manager ๐Ÿ‘” Claude Opus High-level AI. Delegates tasks downward.
Team Leader ๐Ÿ‘จโ€๐Ÿ’ผ Hybrid Works AND delegates. Middle management.
Worker ๐Ÿ‘ท GPT-4o Workhorse doing individual tasks.

๐Ÿ—๏ธ Architecture

CEO (You)
    โ”‚
    โ–ผ
Manager AI (Claude Opus)
    โ”‚
    โ”œโ”€โ”€ Team Leader AI(s)
    โ”‚       โ”‚
    โ”‚       โ””โ”€โ”€ Worker 1โ€ฆN (VS Code windows)
    โ”‚
    โ””โ”€โ”€ Worker Pool (GPT-4o, 4o-mini)

All communication happens through XML files in the project folder:

  • tasks.xml โ€” Task queue with statuses
  • worker.xml โ€” Worker configuration
  • heartbeat.xml โ€” Health status pings
  • finished.flag.xml โ€” Completion signals

๐Ÿ“ฆ Project Structure

root_of_project_{name}/
โ”œโ”€โ”€ project-spec.xml           # Project configuration (layers, resources)
โ”œโ”€โ”€ tasks.xml                  # Task list with statuses (one per worker!)
โ”œโ”€โ”€ hierarchy-config.xml       # Delegation limits
โ”œโ”€โ”€ .gitignore
โ”œโ”€โ”€ input/                     # Input files for processing
โ”œโ”€โ”€ workdir/                   # Working directory (per layer)
โ”œโ”€โ”€ output/                    # Generated outputs
โ”œโ”€โ”€ logs/                      # Execution logs
โ”œโ”€โ”€ prompts/
โ”‚   โ””โ”€โ”€ layer_*_prompt.md      # Auto-generated prompts per layer
โ””โ”€โ”€ workers/
    โ”œโ”€โ”€ worker-L1-1/           # Worker folder (pre-provisioned)
    โ”‚   โ”œโ”€โ”€ worker.xml         # Worker config & paths
    โ”‚   โ”œโ”€โ”€ heartbeat.xml      # Health status
    โ”‚   โ””โ”€โ”€ instructions.md    # Task instructions
    โ”œโ”€โ”€ worker-L1-2/
    โ””โ”€โ”€ worker-L1-N/

๐ŸŽฎ How to Use

1. Open ADG-Parallels Sidebar

Click the ๐Ÿฅš icon in VS Code Activity Bar

2. Create a New Project

  • Click "Create New Project" to open the 4-step wizard:
    1. Name โ€” Enter project codename
    2. Layers โ€” Choose how many workforce layers (1-99)
    3. Resources โ€” Add input files/folders, describe them, set output directory
    4. Configure โ€” For each layer: set type, workforce size, task description, continuation prompt

3. Start Processing

  • After wizard completes, click "Start Processing" in sidebar
  • Workers spawn automatically (one VS Code window per ejajka)
  • Each worker has its own task from tasks.xml

4. Watch the Magic

  • Workers execute tasks in parallel
  • Monitor progress via sidebar
  • Results appear in output directory

๐Ÿ’“ Heartbeat & Self-Healing

Every worker writes a heartbeat file every 60 seconds:

<heartbeat>
  <worker-id>worker-L1-1</worker-id>
  <timestamp>2026-01-16T22:30:00.000Z</timestamp>
  <status>working</status>
</heartbeat>

If a worker stops responding:

  1. Manager detects timeout (no heartbeat > 120s)
  2. Marks worker as unresponsive
  3. Task can be reassigned

๐Ÿ”ฎ What's Next: v1.0.0

The upcoming major version will bring:

  • ๐Ÿ—๏ธ Complete Code Refactoring โ€” Clean architecture, better separation of concerns
  • ๐Ÿ“ Improved Documentation โ€” Full API docs, tutorials, examples
  • ๐Ÿงช Test Coverage โ€” Unit and integration tests
  • ๐ŸŽจ Better UI/UX โ€” Dashboard, progress visualization
  • ๐Ÿ”Œ Plugin System โ€” Custom adapters and extensions
  • ๐Ÿš€ Performance โ€” Optimized spawning and communication

The refactoring will preserve all working patterns and techniques from v0.x while elevating the codebase to production quality.


๐Ÿงช Project Status

๐ŸŸข POC Complete โ€” Pre-production Ready

Version 0.5.0 (January 2026):

  • โœ… Full wizard-based project creation
  • โœ… Multi-worker spawning (tested with 6+ ejajki)
  • โœ… Task execution via Language Model API
  • โœ… XML-based communication
  • โœ… Heartbeat monitoring
  • โœ… Clean Start Processing flow

๐Ÿ“š Documentation


๐Ÿค Contributing

Contributions, ideas, and feedback are welcome!

  • Open an issue to discuss ideas
  • Star the repo to show support โญ
  • PRs welcome for bug fixes and features

๐Ÿ“œ License

AGPL-3.0-or-later โ€” See LICENSE


โญ If you like the project, consider starring the repo.

Ejajki get very excited when they see stars.
Many Ejajkas, One Goal! ๐Ÿฅš

About

AI Delegation Grid: scalable multi-agent orchestration for Copilot and LLMs inside VS Code. Parallel tasks, adapters, automation.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors