Pydantic AI visual location scout

Cross-check visual references with maps and web evidence.

Agent pattern: Typed cross-verification
Search surfaces: Images, maps, and web
Saved artifact: pydantic-ai-location-scout.md

What you’ll build

A Pydantic AI location scout that finds visually distinct campaign locations, verifies the actual places, and checks practical access or restriction details before it makes a recommendation.

Default brief: Scout three Portland, Oregon locations for an outdoor lifestyle campaign. Return visual motifs, logistics, risks, and source URLs.

How the agent works

Find visual directions. Image search supplies candidate locations and the visual motifs that make them distinct.
Verify the place. Maps search confirms location details, access context, and nearby landmarks.
Check practical claims. Web search validates restrictions or logistics so the agent never infers permission from an image.

Core agent setup

This excerpt shows Pydantic AI’s typed, cross-verification tool set. The full script also configures the OpenAI model, validates keys, handles the result, and saves the scouting brief.

agent = Agent(
    model,
    instructions=(
        "Use image search for visual evidence, maps for place facts, "
        "and web search for practical verification."
    ),
    tools=[
        images_search(provider="pydantic-ai"),
        maps_search(provider="pydantic-ai"),
        web_search(
            provider="pydantic-ai",
            allowed_engines=["google_light", "bing"],
        ),
    ],
)
result = agent.run_sync(PROMPT)

View the complete main.py →

Run the recipe

Set SERPAPI_API_KEY and OPENAI_API_KEY in the repository-root .env, then run:

uv run --isolated --no-project --with 'serpapi-search-tools[pydantic-ai]' --with python-dotenv cookbook/pydantic-ai/main.py

Optional controls: OPENAI_MODEL, COOKBOOK_PROMPT, and COOKBOOK_OUTPUT_DIR.

Typed tools do different jobs: The instructions prevent visual search from becoming a proxy for place facts or access permission; every important claim must cross the appropriate tool boundary.

Inspect the result

The recipe writes cookbook-output/pydantic-ai-location-scout.md. Expect three candidate locations with visual motifs, verified place details, nearby landmarks, logistics, restrictions or risks, and source URLs.

Set COOKBOOK_PROMPT to scout a different city, campaign, or creative brief.

SerpApi tools used

images_search finds visual directions, maps_search verifies place details and nearby landmarks, and web_search checks restrictions or practical context. Using the tools together helps the agent avoid treating an image as proof of access, dimensions, or permission.

Framework pattern: The focused typed-tool setup follows Pydantic AI’s weather agent example and uses SerpApi image, maps, and web search.