English | 简体中文

Intel® AI Gaming Assistant Library

Intel® AI Gaming Assistant Library is an AI-powered local backend service that enables in-game intelligent assistance features, including screenshot understanding, multimodal retrieval, knowledge base Q&A, and memory management. It runs entirely on-device, leveraging Intel GPU and NPU for accelerated inference.

What's Inside?

• Vision — Screenshot-based scene recognition and image retrieval
• Knowledge Base — RAG pipeline with embedding, reranking, and LLM-based Q&A
• Memory — Persistent conversation memory management
• MMR — Multimodal retrieval combining vision and semantic search

The server exposes a local HTTP API (default port 9190) for client integration.

System Requirements

Component	Requirement
OS	Windows 10/11 64-bit
CPU	Intel® Core™ Ultra processor (recommended)
GPU	Intel® Arc™ dGPU or Intel® Core™ Ultra integrated GPU
NPU	Intel® NPU (for vision models)
RAM	16 GB minimum, 32 GB recommended
Disk	20 GB free space

Note: The LLM, embedding and rerank models are not bundled in this release. See Getting Started for model download instructions.

Getting Started

1. Download Models

The following model directories must be placed under the models\ folder alongside the executable:

Path	Description
models\llm	Chat LLM — OpenVINO IR format
models\emb	Text embedding model — OpenVINO IR format
models\rerank	Rerank model — OpenVINO IR format
models\splitter	Text splitter model
models\mmr	Multimodal embedding model — OpenVINO IR format

The LLM, embedding, and rerank models are interchangeable. Any compatible model in OpenVINO IR format can be used. Pre-converted models are available on the OpenVINO community on Hugging Face. You can also convert models yourself using Optimum-Intel.

For the splitter model, you can download the language-specific package from the spaCy model catalog.

For MMR, the default configuration loads from models\mmr\gme (configured by config\mmr.json, field mmr.gme.model_path). You can download the model from Alibaba-NLP on Hugging Face and convert the model using Optimum-Intel.

In addition, we provide scripts and instructions under demo/dowload_models to download and arrange models from ModelScope for local personal evaluation and validation. These downloaded models are not part of the Game Assistant service release itself, and their use is subject to the applicable model licenses. At present, they are for personal use only and may not be used for commercial purposes.

2. Run the Server

Extract GameAssistantToolServer.7z, then double-click GameAssistantToolServer.exe in the extracted folder to start the service.

The server starts on 127.0.0.1:9190 by default. See config\gameassistanttoolserver.json to change the host or port.

3. Verify

Once started, the server log will confirm each component's initialization status. You can send requests to:

http://127.0.0.1:9190

Configuration

All configuration files are located in the config\ folder:

File	Description
gameassistanttoolserver.json	Main server config (host, port, model paths, feature toggles)
memory.json	Memory service configuration
mmr.json	MMR service configuration
runtime.json	Runtime behavior settings
logger.json	Logging configuration

See the docs\ folder for detailed configuration and API reference documentation.

API Reference

Full API documentation is available in the docs\ folder. Refer to the following guides:

• Interface Reference — HTTP API endpoints for all features
• Configuration Guide — Detailed explanation of all config fields

License

This software is licensed under the Intel OBL Tools License Agreement.

See third-party-programs.txt for open source components and their licenses.

Security

Please report security vulnerabilities following Intel's guidelines. See SECURITY.md for details.

---