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

Description and use

A 17‑agent swarm running locally on a Pixel 7 Pro, built with TinyLlama, BabyAGI, SQLite, and Kivy, capable of summarizing, fact‑checking, and content creation without cloud reliance.

Problem

Running a multi‑agent system locally on a smartphone to avoid cloud dependence and preserve privacy.

Outcome

The swarm can summarize articles, answer questions, and generate content locally, though it remains sluggish compared to cloud solutions.

Workflow

• 1. Quantize TinyLlama to 4‑bit with llama.cpp and load in Python.
• 2. Set up BabyAGI loop for task creation, prioritization, and execution.
• 3. Store agent memories, tasks, and results in SQLite for persistence.
• 4. Build a basic UI with Kivy and test via Termux on the phone.
• 5. Deploy 17 agents with specific roles (Task Master, Summarizer, Fact Checker, etc.).
• 6. Optimize performance by limiting context window and batching.
• 7. Implement garbage collection for SQLite to manage memory.
• 8. Refine prompts for each agent to ensure cooperation.

Evidence

I opted for TinyLlama-1.1B-Chat-v1.0, a surprisingly capable 1.1 billion parameter LLM. It’s small enough to run reasonably well on a modern smartphone.

Source

I chose BabyAGI. While simple, it provides a solid foundation for managing agent tasks and memory.

Source

Instead of relying on a vector database like ChromaDB, I leveraged SQLite to store agent memories, tasks, and results.

Source

I utilized termux on my phone for initial testing and debugging, accessing the Python environment via SSH.

Source

The current system is functional, albeit a bit sluggish.

Source