AI agent that interacts with the online grocery store Picnic.

Picnic AI Agent

This project demonstrates an agentic use case for interacting with the online grocery store Picnic. Picnic is a Dutch online supermarket that delivers groceries to your door. I am a big fan of their service and simply wanted to create a fun project that combines their API with AI, without using any big agent framework such as LangGraph or Swarm. As an ML model, I have chosen the latest version of Googles Gemini model series called Gemini 2.0 Flash which is currently in experimental state. However, it is multimodal and accepts/generates not only written text but also voice out of the box.

Setup

This project requires a Google Claude API key and a Picnic account.

1. Create a .env file in the root directory and add the necessary environment variables. You can use the .env.example file as a template.
2. Create a virtual environment using uv venv and activate it using source .venv/bin/activate.
3. Run uv run python picnic_agent.py to start the agent.

How to run

In order to run the agent, simply execute uv run python picnic_agent.py. The agent will start and wait for your commands. At this moment in time, it is necessary to use a headset, while talking to the agent, as it uses the microphone to listen to your commands. Otherwise, the agent would talk to himself. picnic_agent.py is running the Gemini API in "TEXT" mode. That means that the model response modality is text and not voice. In a second step, I am converting the text response to speech using the Text-to-Speech API from Google. This gives me a lot more flexibility in terms of language and voice selection. The downside of this approach is that I loose the speech interruption feature. I always have to wait until the AI has finished speaking. I also tested the AUDIO response mode. However, that mode still seams to be a bit