AI Chain

Chained AI inference across distributed edge nodes. Routes prompts through a priority-ordered mesh of Ollama instances on Raspberry Pi 5 hardware.

Architecture

Client → /chain → Probe all nodes → Route to fastest alive node
                                   ↓ (mode=full)
                              Refine via second node

Nodes: Octavia (deepseek-r1), Aria (qwen2.5-coder), Lucidia (tinyllama), Alice (tinyllama fallback)

API

Endpoint	Method	Description
/health	GET	Node health check with model inventory
/chain	POST	Run chained inference ({"prompt": "...", "mode": "fast|full"})

Run

pip install -r requirements.txt
python server.py  # http://localhost:8100

Test

pip install pytest httpx
pytest tests/

Deploy

docker build -t ai-chain .
docker run -p 8100:8100 ai-chain