Snapdragon Game AI SDK vs Neocortex for AI NPCs

Quick answer: The Snapdragon Game AI SDK is Qualcomm's set of Unreal Engine plugins for running speech recognition, a language model and text-to-speech on the NPU of Snapdragon phones and PCs. It is a toolkit for one chip family, not a character platform. Neocortex gives you finished AI characters that run in our cloud on any hardware and, with our team, on your own servers, fully offline, or inside the game build.

If your game ships only to Snapdragon devices, the SDK is one path to on-device NPCs. Most games ship to more than one kind of hardware, and that is where the decision gets interesting.

Snapdragon Game AI SDK vs Neocortex at a glance

What does the Snapdragon Game AI SDK include?

Qualcomm introduced the SDK at GDC as a way to build on-device NPCs, AI teammates, companions and coaches for mobile and PC games. It is published as Unreal Engine plugins in the Snapdragon Game Studios repository on GitHub, in three parts.

• Speech recognizer. Real-time speech-to-text through Qualcomm's Voice AI ASR SDK, with Whisper small (FP16 or quantized) and Whisper large v3 turbo. It supports Android and Windows on Arm.
• LLM pipelines. Conversations, streaming replies, tool calls, embeddings and vector retrieval, with C++ and Blueprint interfaces. Inference runs through Qualcomm's Genie SDK, and you prepare the model files for each target device through Qualcomm AI Hub.
• Text to speech. Speech synthesis through the Voice AI TTS SDK with MeloTTS, in English, Chinese and Spanish. It supports Android only.

NPU execution needs supported Snapdragon hardware and model files compiled for that target. The runtime SDKs come from Qualcomm Package Manager. The plugin source is BSD 3-Clause, and the runtimes and models keep their own license terms. No price is listed on its pages, and they list no shipped games built with it.

What would you still build yourself?

The SDK supplies inference. A character the player can talk to needs a lot around it. Plan for:

• The character. Persona, backstory, tone and guardrails, written as prompts and tuned for a model small enough to share a phone with your game.
• Knowledge. The retrieval module queries a prepared vector index. It does not build one from your design documents, so indexing your lore is your job.
• Memory. Tracking what the player said sessions ago, beyond the model's context window.
• Gameplay hooks. Tool calls give you a channel, but deciding which actions exist, validating them and running them is your code.
• Story structure. Branching beats and checkpoints the game can react to.
• Voice on PC. Text-to-speech is Android only, so Windows builds need another voice path.
• A fallback. Players on other chips need something: a cloud character, scripted dialogue, or no feature at all.
• Per-device QA. Model files are compiled per target, so every supported chip is its own test pass.

In Neocortex most of that list is a node in the editor. Personality, Emotion, Knowledge (upload a PDF or TXT), World, Objects, Intents and Actions are set up visually and tested in the browser without spending credits. Conversation Flows return flowState checkpoint keywords, and replies carry an emotion tag and actions such as { "name": "OPEN_DOOR", "targetId": "door_02" } for your scripts to run. Up to six characters can share one conversation over the API on Pro and Team plans.

What happens on devices without a Snapdragon chip?

An NPU SDK runs where its NPU is. An Android game reaches phones with chips from several vendors. On PC, most players run x86 machines rather than Windows on Arm. iOS and consoles are not on the SDK's platform list. Each of those players needs a different path to the same character, which means a second inference stack or a second design.

A cloud character avoids that split. The model runs on our servers, so a Neocortex character behaves the same on a flagship phone, an older laptop or a browser build, and Unity WebGL builds can take microphone input. The cloud product is built for real-time conversation, with streaming speech-to-text and text-to-speech, voice cloning and voice design.

Hardware-bound runtimes from other vendors raise the same question. We cover them in our comparisons of NVIDIA ACE, Unity Sentis and Meta Quest AI Building Blocks, and in the roundup of AI NPC tools with local AI support.

Can Neocortex run on-premise or offline?

Yes. Some projects cannot depend on a public cloud: a training simulator on a closed network, a museum kiosk, a booth at an event with unreliable Wi-Fi, a client with strict data rules. For those, Neocortex offers Custom (enterprise) arrangements, set up through a conversation with our team rather than self-serve:

• The full platform on your own servers or private cloud, including the character editor, the API, your characters and Conversation Flows.
• Local voice: speech-to-text and text-to-speech run on-premise too, not only the language model.
• Air-gapped operation with no internet connection at all.
• On-device inference that ships inside the game build and runs on the player's machine.
• Our team alongside yours on deployment, model choice and integration.

On-premise deployments are sized with the studio, so response times follow the hardware you plan together.

Measured speed. On a laptop RTX 3080 (8 GB), our on-device runtime returns its first token in 0.06 seconds and starts speaking in about one second, because it voices the first sentence while the rest is still being generated. Speech recognition and text-to-speech run on the CPU, so a 3B model needs about 3 GB of VRAM, and a 2B model generates over 300 tokens per second in under 1 GB.

If your project needs any of this, contact us.

Which one fits your project?

For how the plans and usage costs compare across vendors, see our AI NPC pricing models comparison and the comparison hub.

The bottom line

The Snapdragon Game AI SDK gives Unreal developers on-device speech and language building blocks for one family of chips. Neocortex gives you the character itself, designed visually and running wherever the project needs it: in our cloud for every player on every device, or on your own servers, offline, or inside the game build with our team. Build your first character on the free plan, and if your game needs it to run locally, talk to us about a Custom deployment.

Snapdragon Game AI SDK details checked October 2026.

Frequently Asked Questions