I built a 17-agent AI swarm on my phone — here's how

Description and use

The author built a 17‑agent AI swarm running entirely on a smartphone using TinyML and Flutter. The system performs perception, analysis, and control tasks, communicating via a lightweight message bus, showcasing on‑device, multi‑task AI without cloud reliance.

Problem

Running a complex, multi‑model AI swarm locally on a smartphone without cloud dependence, while maintaining acceptable performance and battery life.

Outcome

The author successfully ran 17 concurrent AI agents on a smartphone, achieving real‑time perception and analysis locally, demonstrating the feasibility of on‑device swarm intelligence.

Workflow

• 1. Define 17 agents grouped into Perception, Analysis, and Communication & Control.
• 2. Select lightweight models (YOLOv5‑Lite, MobileFaceNet, FERNet, etc.) and convert to TensorFlow Lite with quantization and pruning.
• 3. Implement each agent as a Dart isolate in Flutter for asynchronous execution.
• 4. Build a simple message bus (Message, MessageBus classes) for event‑driven inter‑agent communication.
• 5. Load models using TFLite API and run inference on camera frames.
• 6. Agents publish relevant data to topics; subscribers process and act (e.g., anomaly detection, alerts).
• 7. Monitor resource usage (CPU, memory, battery) via Resource Monitor agent.
• 8. Iterate on optimization (quantization, pruning, async processing) to balance performance and power consumption.

Evidence

I’ve built a 17-agent AI swarm that runs, surprisingly well, on my phone. Not in the cloud. On my phone.

Source

I heavily relied on model quantization. Converting models to 8-bit integer format dramatically reduced their size and inference time.

Source

Each agent runs in its own isolate (a lightweight thread in Dart), preventing the UI from freezing.

Source