Sandstorm Desktop is an AI agent orchestrator I built to manage multiple AI agents working in parallel through Docker-based isolation. It started as a CLI tool and evolved into a full cross-platform Electron app after I kept hitting the limits of what a terminal interface could provide for real-time observability and control.
The core idea is simple: when you need multiple AI agents working on different tasks, Sandstorm spins up separate Docker stacks for each one, giving every agent its own isolated codebase and development environment. Each stack runs independently, so agents can operate in dangerous mode without risking your actual repository or stepping on each other's work.
I've been using it in production for weeks now — at my day job and on personal projects. It includes intelligent model selection that picks the right AI model based on task complexity to keep token costs sustainable, real-time token usage monitoring, comprehensive stack history, and the ability to group stacks by ticket for full visibility into agent activity per issue. I used it to complete an entire two-week sprint in three days.