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

Description and use

A 17‑agent AI swarm was built on an Android phone to perform productivity tasks using llama.cpp and a lightweight LangChain setup.

Problem

Running a functional, multi‑agent AI system entirely on a mobile device is resource‑intensive and impractical.

Outcome

A fully functional, on‑device AI swarm capable of handling diverse productivity tasks, demonstrating that multi‑agent systems can run on mobile hardware with careful optimization.

Workflow

• Set up Termux and install Python
• Install llama.cpp with Q4_K_M quantization
• Create a simplified OpenAI‑style API wrapper around llama.cpp
• Select essential LangChain components for agent orchestration
• Implement a message‑broker for agent communication
• Define 17 specialized agents (voice input, transcription, summary, research, code generation, email drafting, to‑do list, context extraction, prompt refinement, memory, routing, action execution, formatting, sentiment analysis, translation, filtering, error handling)
• Use asyncio to run agents concurrently
• Cache frequent results and reuse the loaded model
• Deploy custom DuckDuckGo instant‑answer search tool
• Test and iterate on performance optimizations

Evidence

Running Llama 2 (even a quantized version) on a phone is challenging.

Source

I specifically used the Q4_K_M quantization which provides a good balance between speed and accuracy.

Source