# QSR Store Agent
The QSR Store Agent is an AI operations assistant for quick-service restaurants (QSR). It runs entirely on an Intel® edge system in the store. Restaurant staff ask questions in plain language and get answers built from live store data. The agent also watches store events, such as a weather change, a long queue, or an order mistake, and proposes actions that a manager approves before anything changes. Inference runs locally with the OpenVINO™ Model Server on an Intel® GPU, so store data, conversations, and decisions never leave the premises. ## What It Does | Capability | Description | | --- | --- | | Operator chat | Ask about queues, wait times, menu availability, order accuracy, kitchen safety, tables, footfall, signage, and weather. Answers come from the connected store services, not from the model's memory. | | Event-driven recommendations | Store systems post events to the agent. The agent gathers the relevant live context and either proposes an action or explains why no action is needed. | | Human approval | Every proposed action waits for an operator to approve or reject it in the UI. Nothing changes in the store without approval. | | Learning from operators | The agent records how operators respond to events and, after review, turns repeated responses into new guidance for future events. | | Pluggable store services | Each store system (kiosk, order accuracy, signage, and more) connects through the open Model Context Protocol (MCP). Systems can run on the same machine or elsewhere on the store network. | | Scope guardrail | Questions unrelated to the store are declined before they reach the language model. | ## Why Run It on Intel Hardware - **Local, private inference.** A Qwen3 INT4 model (set by `MODEL_ID`, default Qwen3-30B-A3B) is served by the OpenVINO™ Model Server on the store's Intel® GPU. No cloud dependency or per-request cost. - **Optimized with OpenVINO™.** The model is pre-converted and INT4-compressed for Intel GPUs. Prefix caching reuses the shared prompt between requests to reduce response latency. - **One edge system.** The model server, agent, operator UI, and demo services run as containers on a single Intel-based machine. ## What Is Included - The operator web UI and the agent runtime, packaged as one container image. - The OpenVINO™ Model Server configuration for the local language model. - A weather service that can simulate conditions for demos or report live weather. - Connection settings for store services, such as kiosk, order accuracy, kitchen safety, and digital signage, that run on their own hosts. ## Documentation - [Get Started](./get-started.md): install and run the agent. - [System Requirements](./get-started/system-requirements.md) - [Configuration](./get-started/configuration.md): set environment variables and connect services. - [Build from Source](./get-started/build-from-source.md): compile the agent locally. - [How It Works](./how-it-works.md): architecture and data flow. - [Operator Guide](./operator-guide.md): use the chat, approve decisions, and teach the agent. - [API Reference](./api-reference.md): HTTP endpoints of the operator UI. - [Troubleshooting](./troubleshooting.md) - [Release Notes](./release-notes.md) ## Supported Use Cases For details of the supported use cases, see these pages: - [Digital Signage](https://docs.openedgeplatform.intel.com/dev/edge-ai-suites/ai-suite-retail/digital-signage/index.html) - [Loss Prevention](https://docs.openedgeplatform.intel.com/dev/edge-ai-suites/ai-suite-retail/loss-prevention/index.html) - [Order Accuracy](https://docs.openedgeplatform.intel.com/dev/edge-ai-suites/ai-suite-retail/order-accuracy/index.html) - [Smart Kiosk Assistant](https://docs.openedgeplatform.intel.com/dev/edge-ai-suites/ai-suite-retail/voice-enabled-interactions/smart-kiosk-assistant/index.html) :::{toctree} :hidden: ./get-started.md ./how-it-works.md ./operator-guide.md ./api-reference.md ./troubleshooting.md Release Notes <./release-notes.md> :::