OpenSquilla routes each turn to the cheapest model that can handle it

TL;DR AI
2 min readKey summary
OpenSquilla introduced SquillaRouter, an on-device classifier that routes each turn to the cheapest model that can still handle the task.
The router works across many providers, including OpenRouter, OpenAI, Anthropic, Ollama, DeepSeek, Gemini, and Qwen.
If native dependencies like ONNX Runtime or LightGBM are missing, it falls back to direct routing instead of breaking the workflow.
The shift from fixed model selection to per-turn routing could cut costs, improve agent performance, and turn usage data into a training flywheel.
