Hi HN

I built AIRUNCODE because I was frustrated by two patterns in current AI coding environments: cloud vendor lock-in that wipes project memory when switching providers, and platforms adding heavy markups on top of user API keys.AIRUNCODE is a local-first desktop runtime (v1.4.7) that executes parallel multi-agent swarms directly against your local workspace. It's fully BYOK (Bring Your Own Key) or compatible with local inference via Ollama/GGUF, so you pay the providers directly with $0 markup.The Architecture:Desktop Shell: Electron 43 with a React 19 / Tailwind 4 interface.Local State & Memory: Powered by SQLite WAL and sqlite-vec to maintain a local, persistent knowledge graph and project index on your machine.Verification Loop: It autodetects local testing frameworks (Vitest, Cargo, pytest, etc.). Agents run tasks in parallel using isolated worktrees and must pass test assertions before completion. If they fail, the system triggers a localized context rollback to protect your main git branch.Graphics Pipeline: For graphics and simulation devs, it features V-CORE Studio, a native runtime built in C++, Vulkan, and SDL3 that allows agents to drive live 3D development and Lua scripting.The packaged binaries are ready for Windows (x64), macOS (Apple Silicon), and Linux.I'm the solo creator and would love to answer any questions about the multi-agent debate coordination, the local context compaction mechanics, or the Vulkan integration. Looking forward to your brutal feedback!