The Original Multi-Agent Swarm.
Swarmussy is the Gen 1 experiment that started the Ussyverse agent lineage. It proved that multiple cheap AI agents, coordinated by a lead Architect agent, could collaborate on complex software development tasks. Experimental and chaotic, but it worked — and every Ussy agent that followed learned from it.
Each agent has a distinct role and tool access. Orchestrators get planning tools; workers get coding tools. True role-based separation.
A lead Architect agent delegates work to specialized workers rather than doing everything itself. Real orchestration, not just parallel prompting.
Real-time terminal UI for tracking task progress, blockers, and token usage across all agents in the swarm.
Support for managing multiple projects with persistent settings between sessions. Workspace state survives restarts.
Swarmussy was the first experiment in the Ussyverse. The core hypothesis: can you get multiple cheap AI models to coordinate on a software project if you give them clear roles and structured communication?
The answer was yes — with caveats. Agent coordination worked, but the Python-based architecture hit scaling limits, and the lack of structured planning meant agents sometimes worked at cross-purposes.
These lessons directly informed Ralphussy (Gen 2), which added structured DevPlan workflows, and then Geoffrussy (Gen 3), which rewrote everything in Go with the DevUssy planning engine built in.