Add Streamlit-based Weather Agent Example

[rmaceissoft]

Solves: #826

Includes:

• Added a new Streamlit example script: examples/pydantic_ai_examples/weather_agent_streamlit.py
• Updated the pyproject.toml in the pydantic-ai-examples workspace to include the Streamlit dependency.
• Modified weather_agent.py to postpone the model instance creation, allowing the model's api_key to be passed when running the agent instead of relying on the environment variable. (This change is necessary because there is currently no way to pass the model API key when running an agent.)

How to test:

• Launch the streamlit app by running the following command (make sure you install streamlit before)

uv run streamlit run examples/pydantic_ai_examples/weather_agent_streamlit.py

• You can also set default values via env variables:

MODEL_API_KEY=<KEY> WEATHER_API_KEY=<KEY> GEO_API_KEY=<KEY>uv run streamlit run examples/pydantic_ai_examples/weather_agent_streamlit.py

Screengrab:
https://www.loom.com/share/eb96048762a74d7dafd0ef3192c8ecb5

[sydney-runkle]

Cool! Excited to review this later today - thanks for the contribution!

[asaf]

btw you should look at that https://github.com/ishswar/pydantic-ai-weather-agent it generates a widget out of the agent result which demonstrates how content could be visualized as well, maybe you can adapt (unless it makes the sample too complex) :)

[rmaceissoft]

btw you should look at that https://github.com/ishswar/pydantic-ai-weather-agent it generates a widget out of the agent result which demonstrates how content could be visualized as well, maybe you can adapt (unless it makes the sample too complex) :)

@asaf Thanks for pointing that out! I'll take a look and see what ideas I can incorporate.

Here are a few things I'm planning to add/improve in this Streamlit-based example:

• Enhance the rendering of model responses:
  • Add a loading indicator before the model response is returned and disable the send button.
  • Allow users to re-send a prompt from any previous message (modifying the chat history from that point).
  • Render tool calls as well.
  • Add a toggle to switch between streamed and non-streamed results.
  • Properly catch model errors and display a corresponding error message in the chat.
• Allow users to clear chat history via a button in the sidebar.
• Enable tuning of model settings (at least temperature) from the UI (sidebar).

@sydney-runkle I'd love to hear your thoughts on this.

[sydney-runkle]

Alrighty, I've chatted with the team - we've decided that we don't want to include third party tools like this in our pydantic_ai_examples library because of the significant deps bloat introduced. That being said, we'll add an "external examples" page to the docs that highlights great use cases like this one!

[rmaceissoft]

@sydney-runkle Thanks for the update! I completely understand the concern about dependency bloat. I'll keep track of the "external examples" page and work on wrapping up my example so it can be included there.