Get Started#
This guide takes you from a fresh Intel® edge system to a running QSR Store Agent.
Prerequisites#
A Linux system with an Intel® GPU that meets the System Requirements.
Docker Engine with the Compose plugin, Git,
make, and Python 3.Network access to GitHub, Docker Hub, and Hugging Face. If you are behind a proxy, see Configuration.
Step 1: Get the Code#
git clone https://github.com/intel-retail/qsr-agent.git
cd qsr-agent
Step 2: Download the Model#
make download-models
This downloads the model set by MODEL_ID in .env.example (default
OpenVINO/Qwen3-30B-A3B-Instruct-2507-int4-ov, about 16 GB) into ./models. The download is skipped if the model is already present.
Step 3: Connect Store Services#
Store data comes from the store services, which run on their own hosts. Set
their MCP URLs in .env.example before you start the agent, for example:
QSR_KIOSK_MCP_URL=http://<kiosk-host>:<port>/mcp
QSR_EXTRA_NO_PROXY=<kiosk-host>
See Connect Store Services. Without service URLs the agent runs with weather only.
Step 4: Start the Agent#
make up
make up checks Docker, the Intel GPU device, and the model. It then pulls the
agent image and starts three containers (to run the code you cloned instead of
the published image, use make up REGISTRY=false; see
Build from Source):
Container |
Role |
|---|---|
|
OpenVINO™ Model Server running the language model on the Intel® GPU |
|
Operator UI and agent runtime |
|
Weather service (simulated by default) |
The first start loads the model onto the GPU and warms it up. This can take a few minutes. Check progress with:
make status
The agent is ready when qsr-agent reports healthy.
Step 5: Open the Operator UI#
Open http://<host>:8600 in a browser, or http://localhost:8600 on the same
machine.
Try a few questions in Operator Chat:
What is the current queue count?
What is the current order accuracy rate?
What’s the weather at the store right now?
Step 6: Trigger an Event#
Change the simulated weather. This also sends a weather event to the agent:
make weather CONDITION=rain TEMP=12
The event appears in the Autonomy Decisions panel, and the agent evaluates it. If the agent proposes a change, such as taking cold drinks off the menu, select Approve or Reject. See the Operator Guide for more.
Stop the Agent#
make down
Settings, decisions, and learned behaviour are kept in Docker volumes and are
restored on the next make up.
Next Steps#
Connect your store services: Configuration.
Understand the system: How It Works.
Familiarise yourself with the Operator Guide.