Dify supports integrating locally deployed models via Ollama, but does not recommend it as the first choice, because local models that can run on ordinary machines have fewer parameters and suboptimal performance.