I’m excited to share that QonQrete v0.5.0 beta is now available for testing and feedback.
QonQrete is a local-first, agentic AI orchestration system designed for secure, observable, and human-in-the-loop software construction. It coordinates autonomous AI agents to plan, execute, and review code generation — all within an isolated sandbox environment on your own infrastructure. Think of it like a local-first, agentic AI “construction yard” that plans, writes, reviews, and version-controls your code inside a safe sandbox on your own machine.
Core Architecture
🤖 Three-Agent Pipeline
- InstruQtor - Analyzes tasks and generates detailed execution plans
- ConstruQtor - Executes the build process and generates code artifacts
- InspeQtor - Reviews output quality and provides actionable feedback
🔐 Security-First Design
All agent execution occurs within containerized environments (Docker/Microsandbox). The host system remains isolated from AI-generated code, ensuring a robust security boundary between orchestration and execution.
⚡ Flexible Execution Modes
Run fully autonomous pipelines for rapid iteration, or enable user-gated checkpoints for manual approval at each cycle. The control model adapts to your workflow requirements.
🧠 Multi-Provider Support
Supports OpenAI, Google Gemini, Anthropic Claude, and DeepSeek. Configure different providers per agent to optimize for cost, capability, or preference.
🏠 Local-First Architecture
Runs entirely on your infrastructure with no cloud dependencies — a self-hosted alternative to cloud-based AI development platforms. Your API keys, your compute, your data.
Current Status
Core pipeline functionality is operational. The Text-based User Interface (TUI) and Microsandbox runtime are currently in active development.
I welcome feedback, contributions, and discussions from the community.
🔗 Repository: github.com/illdynamics/qonqrete