Key results: - 100-agent swarm: 100/100 synchronized via pure audio - Pathfinding: 5 agents found optimal path in 0.35s - Resource allocation: 0.92 fairness score through audio negotiation - Cost: $2 vs $53 per 1K queries (96% reduction)

How it works: - 40 core concepts (exists, perceives, good, bad, future...) - Each concept = unique ultrasonic frequency - Agents decode with FFT, no LLM calls needed - Agent-to-agent communication is nearly free

Limitations: - Only 40 concepts (limiting for complex tasks) - Crude number encoding - Still needs LLM for human translation

Looking for feedback on: 1. Better encoding schemes for limited concept space 2. Real-world use cases beyond swarm coordination

GitHub: https://github.com/Nil4s/swl-agent Try it: python swl_swarm_sync_test.py --mode audio_fm --agents 50