Skip to content

Releases: adamerso/adg-parallels

v0.5.0 - POC Complete! 🎉🥚

16 Jan 22:32
@adamerso adamerso
v0.5.0
d2804eb

Choose a tag to compare

View all tags
v0.5.0 - POC Complete! 🎉🥚 Pre-release
Pre-release

🎉 POC Complete!

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

✅ What Works

  • Multi-Ejajka Spawning - All workers spawn correctly via Start Processing button
  • Project Wizard - 4-step project creation with layer configuration
  • Direct XML Mapping - Form fields map directly to task XML
  • Task Execution - Workers execute tasks via Language Model API
  • Heartbeat Monitoring - Health status tracking

📋 New XML Fields

  • your_assigned_task - Main task description
  • move_completed_task_artifact_to - Output directory
  • resources_description - Input interpretation
  • continuation_prompt - Poganiacz for stuck workers
  • reporting_instructions - Progress reporting

🔮 What's Next

This is the final v0.x release. Version 1.0.0 will bring a complete codebase refactoring with clean architecture, tests, and production-ready quality.

Many Ejajkas, One Goal! 🥚

Loading

ADG-Parallels v0.1.0 — The First Real Copilot AI Orchestration Engine

07 Dec 08:37
@adamerso adamerso
ADG-parallels
939c959

Choose a tag to compare

View all tags
ADG-Parallels v0.1.0 — The First Real Copilot AI Orchestration Engine Pre-release
Pre-release

🥚 ADG-Parallels — v0.1.0 Early Orchestration Release

“Parallel AI workers with real queueing, adapters, and autonomous execution.”

This release marks the first fully functional orchestration build of
ADG-Parallels. The system is no longer a concept — it is now a working,
parallel, multi-agent execution engine inside VS Code.

Using the new LM Client (vscode.lm API), Workers can autonomously:

  • claim tasks,
  • execute them via models,
  • detect completion criteria,
  • save outputs,
  • update statuses,
  • and automatically continue with the next task.

Multiple workers run simultaneously, safely coordinated through the
shared queue with file locking.

This is the first version where ADG-Parallels behaves like a real
AI “workforce”.

🚀 Core Features Implemented in v0.1.0

✅ 1. Task Queue Engine (COMPLETE)

  • Full CRUD system for tasks
  • Atomic locking (no race conditions)
  • Status transitions (pending → processing → completed)
  • Timestamped updates
  • Continue-flow support (workers pick next task automatically)

✅ 2. Parallel Worker Execution

  • Multiple VS Code windows acting as Workers
  • Each Worker processes tasks independently
  • Validated POC: 2 workers × 4 tasks in parallel
  • Full compatibility with LM models (OpenAI, Copilot, etc.)

✅ 3. Role Detection (COMPLETE)

  • Automatic inference of CEO / Manager / Worker / TeamLeader
  • Based on .adg-parallels/ folder structure
  • Workers run fully autonomously once launched

✅ 4. Adapter System (COMPLETE for v1)

  • Adapter loader with schema validation
  • Mustache-based prompt rendering
  • Helpers: slugify, date formatting, truncation
  • Built-in adapters:
    • generic.adapter.json
    • article-generation.adapter.json
    • task-splitter.adapter.json (meta-adapter)
  • Completion criteria:
    • regex patterns
    • length constraints
    • "TASK COMPLETED" signal parsing

✅ 5. LM Execution Pipeline

  • Wrapper for vscode.lm API
  • Model selection
  • Request streaming
  • Token counting
  • Error handling (NotFound, Blocked, NoPermissions)
  • Unified interface for Workers

✅ 6. Worker Lifecycle & Health

  • Heartbeat file updates (every 30s)
  • Health monitor (every 15s)
  • Faulty worker detection
  • Logs in VS Code output window
  • Early support for future auto-restart

✅ 7. Project Provisioning

  • ADG: Provision New Project
  • Creates entire folder structure
  • Generates tasks, hierarchy-config, adapter setup
  • Supports selecting task type at provisioning time

🟨 8. Task Splitting Foundations

  • task-splitter.adapter.json implemented
  • Meta-adapter mechanism working
  • Split logic strategies planned, not yet added

✨ What This Release Demonstrates

This is the first version where ADG-Parallels can:

  • spawn autonomous agents
  • coordinate them
  • run parallel workloads
  • maintain isolation via worker folders
  • process a real multi-task job list
  • use a pluggable adapter system
  • monitor worker health
  • detect completion & continue work

This is now a true early orchestration framework, not just a prototype.

🧪 Partial / In-Progress

Dashboard / control center

  • Webview Dashboard — TODO
  • Live task + worker view — TODO
  • Pause/Resume global control — TODO

Adapters

  • Translation adapter — TODO
  • Code audit adapter — TODO

Audit flow

  • Audit rules — TODO
  • Audit review assignments — TODO

Task splitting

  • per-line strategy — TODO
  • per-chunk — TODO
  • AI-driven splitting — TODO
  • merging outputs — TODO

🔥 Why this version matters

This release proves the architecture is valid:

  • Workers can act without user interaction
  • Queue engine scales across windows
  • LM models can be orchestrated automatically
  • Parallel execution is stable
  • Output handling is deterministic
  • Adapter system is extensible
  • Role detection unlocks hierarchy features

From here, the system moves into Phase 2 (Dashboard) and
Phase 4–6 (Audit, Splitting, Self-Healing).

📦 Included Assets

  • adg-parallels-0.1.0.vsix — fully functional extension
  • All source code, adapters, schema types
  • Updated documentation:
    • Vision (POC validated)
    • Roadmap (Phase 1 & 3 complete)
    • Corporate Statute

⭐ Final Note

Your AI employees are no longer “conceptual”.
They work, they queue, they coordinate, they detect roles,
they use adapters, they continue tasks, and they don’t complain.

A new phase of ADG-Parallels begins.

Star the repo ⭐ and follow the rise of the AI workforce.
adg-parallels-main.zip

Loading