Unity Sentis vs Neocortex for AI NPCs
If you search for "AI in Unity", you will quickly find Sentis, Unity's own library for running neural networks inside your game. (It was briefly renamed Inference Engine in 2025, then the display name went back to Sentis.)
It is natural to wonder whether Sentis is all you need for AI NPCs. The short answer: it is a powerful building block, but it is not a character system.
What Sentis does
From Unity's documentation (checked October 2026):
- Runs models on-device on the CPU or GPU, on all Unity runtime platforms.
- Imports ONNX, LiteRT and PyTorch models.
- Tokenizer support for language models, including Hugging Face tokenizers in recent versions (Unity 6 and later).
Sentis is excellent for custom on-device models: vision, pose estimation, classifiers, and small generative models.
What Sentis does not do
- No AI NPC feature. Unity has no official conversational character feature or LLM chat sample for Sentis. Unity AI, the successor to Muse, is an Editor assistant for building your game, not a runtime character system.
- No character design. Personality, knowledge, memory and guardrails are all up to you.
- No voice pipeline. Speech-to-text and text-to-speech are separate models you find, convert and integrate.
- You bring the model. Finding a language model small enough to run well, converting it, and tuning it for your target devices is your job.
Sentis vs Neocortex at a glance
| Unity Sentis | Neocortex | |
|---|---|---|
| What it is | On-device neural network runtime | Complete AI character platform |
| Character design | Code | Visual node editor |
| Language model | Bring, convert and tune your own | Managed for you, plus local LLMs via Ollama |
| Voice | Build your own pipeline | Streaming TTS and STT, voice cloning and design |
| Gameplay hooks | Build your own | Actions, interactable objects, per-line emotions |
| Engines | Unity | Unity, Unreal, Godot (beta), Web API |
| Time to a talking NPC | A research project | Test in minutes, integrate in hours |
Why teams choose Neocortex
Skip the ML work. You do not need to find, convert or tune models. Design the character, install the SDK and run a sample scene.
Designers own the character. Personality, emotion, knowledge, intents and actions live in a visual graph that anyone on the team can edit and test in the browser.
Voice included. Players speak, characters answer in a cloned or designed voice, including in Unity WebGL builds.
Still local when you want it. Prototype with open models such as Llama, Gemma, Qwen and Mistral running on your own machine through Ollama, and self-host the platform when a project needs it.
Use both
Sentis and Neocortex are not either-or. Use Sentis for on-device perception, such as gesture or pose recognition, and feed what it detects to a Neocortex character as game events.
Unity details were last checked in October 2026.
Written by
Sercan Altundas
Date
Oct 17, 2026
Read time
4 min
