Intel® AI Gaming Assistant Library
An on-device AI backend library for building in-game assistance features - screenshot understanding, retrieval-augmented Q&A, memory and speech recognition - accelerated on Intel GPU and NPU hardware, exposed to a game client over a local HTTP API.
Intel AI Gaming Assistant Library is a local backend service that game developers integrate to add AI-assisted features inside a game: screenshot-based scene recognition and image retrieval, a hybrid dense-and-BM25 retrieval-augmented knowledge base with reranking and optional query rewriting, persistent conversation memory, multimodal retrieval that combines vision and semantic search, speech-to-text with hotword support, optical character recognition on screenshots, and a boss-action recognition and tracking module for game-specific event annotation.
Everything runs on-device, using Intel GPU and NPU acceleration, and the server exposes its features to a game client over a local HTTP API on port 9190 rather than a cloud service. Models are not bundled; chat, embedding, rerank, splitter, multimodal, ASR and OCR models in OpenVINO IR format must be downloaded separately and placed under a models directory, with pre-converted options available from the OpenVINO community on Hugging Face or convertible with Optimum-Intel.
It requires Windows 10 or 11 and an Intel Core Ultra processor or Arc GPU, with 16 GB of RAM as a minimum.