Show HN: QonQrete – Local-first multi-agent system for sandboxed code generation (github.com)

QonQrete has launched its v0.5.0 beta, introducing a Secure AI Construction Loop System that employs a Multi-Agent Pipeline Orchestrator within a sandbox environment using YAML configuration. The system facilitates collaboration among three AI agents, allowing for both user-controlled and fully autonomous operation. This design cultivates secure, observable software development, employing a robust architecture that isolates AI processes from the host system, ensuring that any code generated is contained within a Docker-secured environment. The significance of QonQrete lies in its human-in-the-loop mechanism, where users review outputs after each cycle, thus offering control over the development process. The agents operate in distinct phases: planning by the instruQtor, execution by the construQtor, and review by the inspeQtor, collectively creating an auditable trail of communication through markdown files. With extensive testing underway for its Text-based User Interface (TUI) and opportunities for community contributions, QonQrete presents a novel approach to AI-driven software construction that prioritizes security, user oversight, and collaborative workflows, likely influencing future multi-agent systems in the AI/ML landscape.