VibeCodeHPC-jp

VibeCodeHPC is a multi-agent framework that automatically optimizes high-performance computing code by coordinating multiple AI coding assistants (Claude Code, Cline, Gemini CLI, and others) through tmux inter-process communication. It uses a hierarchical agent structure—Project Manager, Software Engineer, Programmer agents, and Code Delivery—to explore parallel optimization strategies across different compilers (GCC, NVIDIA, Intel) and parallelization approaches (OpenMP, CUDA, OpenACC). The framework provides pluggable strategies for HPC parallelization and GPU optimization, with evolutionary exploration across a flat directory structure. It includes built-in monitoring of agent context usage and performance timelines, eliminating the need for external orchestration frameworks while enabling parallel search across optimization variants.

Prepare three items: a requirement_definition.md file (editable from template), _remote_info/ directory with site-specific hardware configuration, and BaseCode/ containing the code to optimize. Run the framework to spawn coordinated AI agents via tmux that automatically generate optimization variants, compile, benchmark, and track results. See docs/quickstart.md for detailed setup instructions.