unity-sentis

Use this skill when a trained model already exists and the job is to make it run reliably inside Unity. The core problems are model import, backend selection, warmup, inference cadence, and platform-safe deployment.

When to use this skill

• Importing an ONNX model into Unity and running inference in play mode
• Migrating from older Sentis wording to the current Unity AI Inference package path
• Choosing between BackendType.GPUCompute, BackendType.CPU, and fallback behavior
• Running NPC or gameplay inference without a Python runtime
• Profiling model cost and reducing frame spikes
• Connecting ML-Agents training output to runtime deployment

Instructions

Step 1: Install the package and note the rename

Current Sentis docs state that Sentis is now called Inference Engine and the latest guidance is moving to com.unity.ai.inference@latest. Existing Sentis 2.1.x projects still use the Unity.Sentis API surface.