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

Description and use

A 17‑agent AI swarm runs entirely on an Android phone, performing collaborative object recognition and scene description without cloud dependency.

Problem

Run a multi‑agent AI system for collaborative object recognition and scene description entirely on an Android phone without cloud dependency.

Outcome

A fully local, 17‑agent AI swarm capable of collaborative scene understanding runs on a single Android phone, demonstrating edge AI feasibility and emergent consensus behavior.

Workflow

• 1. Quantize TinyLLM to 1.1B 4‑bit model with llama.cpp.
• 2. Deploy MobileNetV2 TFLite for object detection.
• 3. Define 17 agents with distinct roles and prompts.
• 4. Use Tasky to queue tasks and distribute to agents.
• 5. Agents process image data and generate partial outputs.
• 6. Consensus agents aggregate results into final description.
• 7. Run entire pipeline on Pydroid 3 on Android.
• 8. Monitor memory usage and optimize resource allocation.

Evidence

I’m using a 1.1B parameter version of TinyLLM, quantized down to 4‑bit precision using llama.cpp.

Source

Leveraging a pre‑trained MobileNetV2 model (using TensorFlow Lite), these agents identify objects within the image

Source

Tasky is my custom system for orchestrating the agents. Think of it as a task queue and communication bus.

Source