Universal Agent HP Presents a 10-Layer Autonomous AI Agent OS for Software Development

Summary

Universal Agent HP is an open-source command-line system that presents autonomous software development as a 10-layer hierarchy. Its architecture includes a CEO-level meta-orchestrator, four departmental leads, and 27 specialist workers that divide coding, research, operations, and quality tasks. The project uses Kahn-style DAG planning to execute independent subtasks in parallel waves, with selective replanning when a branch fails. It also combines a Reflexion self-critique loop with a multi-agent debate process involving an advocate, skeptic, and arbitrator. For code changes, the repository describes AST-based caller and dependency analysis to estimate a change's blast radius, plus a sandbox that creates SHA-256 filesystem snapshots and can restore modified or deleted files. Additional components include dynamic tool discovery, Bayesian EWMA reliability ratings, capability-based model routing, and token or cost budgeting. The system supports cloud providers and local runtimes such as Ollama, Laya MLX, and LM Studio, and can be configured with no API keys when using the project's listed keyless or local options. Its interfaces include an interactive CLI, web panel, terminal TUI, and Telegram bot. The README lists Windows, macOS, and Linux support, Python 3.11 or newer, and a reported test result of 600 passed and one skipped, while these figures are claims made by the project documentation rather than an independent evaluation.