Available to download from this repository now - The first smoke-test to produce working code you can open in Visual Studio!
Update #2 Smoke 293 got SUCCESS so I am now moving to port the smoke test in to the UI as the full pipeline.
Here's the full, long, GitHub-ready README for Neon Sovereign:
The Autonomous Engineering Workstation
Define the outcome. Walk away. Return to working software.
Neon Sovereign is a high-performance, native C++20 AI-powered IDE engineered from first principles to function as a complete autonomous software house.
It is not a chat assistant.
It is not a copilot plugin.
It is not another Electron wrapper.
It is a deterministic execution pipeline that plans, coordinates, builds, verifies, repairs, and delivers real software with minimal human intervention.
Today's AI development tools suffer from deep architectural limitations:
| Issue | Current Tools | Neon Sovereign Approach |
|---|---|---|
| Performance & Bloat | Electron / Web-based | Native C++20 + Vulkan (120 FPS) |
| Constant Supervision | Frequent user prompts & approvals | Fully autonomous DAG execution |
| Context Amnesia | Short, stateless conversations | Persistent SQLite Ledger + BM25 + Semantic Search |
| Hallucinations | No grounded verification | Compiler Shadow Buffers + Sentinel QA + Self-Repair Loops |
| Blind Output | Cannot see the UI it creates | Silicon Retina (VLM visual validation) |
| Sequential Execution | One agent at a time | Parallel Specialist Swarm |
| Cloud Dependency | API quotas, privacy risks | Fully offline / air-gapped capable |
| Unreliable Delivery | Often fails to produce working code | Integration Gates + Crucible Repair System |
Neon Sovereign was built specifically to solve these systemic failures.
Neon Sovereign operates like a professional engineering organization:
- Demand Phase — High-level brief from the user
- Executive Alignment — CEO, CPO, and CTO agents define scope, constraints, and technical direction
- Architecture Phase — Architect agent produces a formal Task DAG (Directed Acyclic Graph)
- Swarm Execution — Parallel specialist agents (Developers, ForgeMaster, etc.) coordinate via gossip protocol
- Verification Gauntlet — Multiple layers of automated checks
- Self-Repair Loops — Automatic iteration on failures using compiler feedback and historical memory
- Delivery — Verified artifacts, forensic logs, and visual validation
Smoke Test Example:
From brief to complete, compiling Win32 C++20 application → 4 minutes 43 seconds
- Task DAG System — Mathematically optimal dependency planning using Kahn’s Algorithm
- Self-Repair Loops — Persistent iteration until integration gates pass
- Crucible System — Multi-stage peer review, Sentinel audit, and shadow compilation
- Silicon Retina — Vision model validation of running application UIs
- Persistent Memory Ledger — SQLite with BM25 + semantic retrieval
- Universal Target Support — Native, Web, Mobile, Embedded, Legacy, and Virtualized systems
- Local-First Design — Runs entirely on your hardware with Ollama weights
- Hyper-V + SPICE Labs — Built-in safe testing environments
The control plane is deliberately deterministic. LLMs are treated as powerful executors, not the source of truth.
- Sovereign Semantic Spine (Tree-sitter backed stable semantic graph)
- Contract Epochs + deterministic arbitration
- Causal Ledger for historical build incidents
- Resource Governor ("The Silicon Leash") for intelligent VRAM management
- Architect
- Specialist Developers
- ForgeMaster (build integration)
- Sentinel (QA & security)
- Silicon Retina (visual validation)
- Chaos Agent (fuzzing & stress testing)
- Hyper-V Laboratory — Modern Windows/Linux guest testing
- SPICE Viewport — Retro and exotic OS targets
- Binary Lab — Ghidra integration for low-level analysis
Proven Strengths:
- Excellent native C++/Win32 delivery
- Strong autonomous execution and self-repair
- Fast smoke test cycles
- Clean, maintainable generated code
Active Development Areas:
- UI/UX refinement (MDI workspace)
- Expanded language and framework support
- Multi-GPU and cluster scaling
- Enterprise features (secrets management, audit trails, permissions)
- Windows 11 (23H2 or newer)
- Visual Studio 2022 with Desktop C++ workload
- CMake 3.28+
- GPU with ≥ 8GB VRAM (strongly recommended)
cmake -S . -B build/windows -G "Visual Studio 17 2022" -A x64
cmake --build build/windows --config Release.\Neon-Sovereign.exe --headless "Create a minimal Win32 C++20 click counter app..."Neon Sovereign is designed to run fully offline.
Recommended Model Pairing:
- Logic / Architect Tier: Gemma-4 or Qwen 3.x class (planning & code generation)
- Speed / Sentinel Tier: Qwen Coder family (different family from Logic tier)
Models are loaded via Ollama. VRAM tiering is handled automatically.
This is just the beginning.
Future versions will push toward:
- Multi-node GPU farm scaling
- Full software product lifecycle automation
- Advanced air-gapped enterprise deployments
- Even more sophisticated self-improvement capabilities
Neon Sovereign represents a new paradigm: AI that doesn't just help you code — it builds software like a professional team.
Define the outcome.
Neon Sovereign handles the rest.
Business & Collaboration Inquiries: chronic9000@gmail.com