LLMUnity vs Neocortex - Local LLM NPCs in Unity

If you have searched for "run an LLM locally in Unity", you have found LLMUnity. It is an open-source (Apache-2.0) package from undream.ai that runs models through llama.cpp inside your game, with no server and no data leaving the player's device.

It is excellent at what it does. This post covers what LLMUnity gives you, what you still have to build, and when Neocortex is the faster path.

What LLMUnity does well

From its GitHub repository (checked October 2026):

• Fully local inference on PC, Android, iOS and VR, with GPU support for NVIDIA, AMD and Apple Metal.
• Model management for downloading GGUF models, including downloading them at build time.
• RAG and embeddings for searching your own documents.
• Structured output with grammars or JSON schema, and function calling.
• Free and open source.

If you need a fully offline model inside a Unity build and you are comfortable with C#, LLMUnity is a solid choice.

What you still build yourself

LLMUnity is an inference library, not a character platform. Its README covers no:

• Voice. No text-to-speech or speech-to-text pipeline.
• Character editor. Characters are configured in the Inspector and in code.
• Hosted option. If a player's device cannot run a model, you need your own server.
• Other engines. Unity only.

You also choose, ship and tune the models, write the prompts and memory logic, and manage build size and performance on mobile, where small 1 to 2B models are recommended.

LLMUnity vs Neocortex at a glance

Why teams choose Neocortex

Use local and cloud models together. Prototype with open models such as Llama, Gemma, Qwen, Mistral and DeepSeek running on your own machine through Ollama, then switch to the cloud for players whose devices cannot run a model.

Characters that designers can own. Personality, emotion, knowledge, intents and actions live in a visual graph that anyone on the team can edit and test in the browser.

Voice out of the box. Players speak, characters answer in a cloned or designed voice, streamed so replies start fast, including in Unity WebGL builds.

One character, every engine. The same character works in Unity, Unreal, Godot and anything that can call a web API.

Which should you pick?

• Strictly offline, text only, Unity only, and you enjoy building the stack yourself: LLMUnity.
• You want voiced, designer-friendly characters that run locally while you prototype and scale when you ship: Neocortex.

Details about LLMUnity were last checked in October 2026.