Information retrieval

Research agent

Answers from what the search actually returned, and hands back the URLs.

web_search

The program, and what it printed

This ran against effGen with gemini:gemini-3.1-flash-lite. The pane under the code is that run’s output, pasted — so where the answer depends on a live API or on the model’s wording, yours will differ. The pane is the shape that run actually had, not a tidied one.

python

from effgen import Agent, AgentConfig
from effgen.tools.builtin import WebSearch

agent = Agent(AgentConfig(
    model="gemini:gemini-3.1-flash-lite",
    name="research-agent",
    system_prompt="Answer from what the search returned. Do not answer from memory.",
    tools=[WebSearch()],
    max_iterations=6,
))

response = agent.run("What is the Open-Meteo API, and does it need an API key?")

print(response.text)
print()
print("sources:", len(response.sources))
for url in response.sources[:3]:
    print(" ", url)

what it printed

The Open-Meteo API is a service that provides access to historical, current, and forecasted weather data [1, 4]. It is designed to be easy to use and is a popular choice for projects requiring weather information [3, 4].

Regarding authentication, the Open-Meteo API does not require an API key for development or low-volume production use [2, 5]. It allows users to access its services without the need for registration, signup, or credit card information [3, 4]. However, users should review the current terms for heavy commercial workloads [2].

sources: 5
  https://freeapihub.com/apis/open-meteo-historical
  https://apideposu.com/en/blog/build-weather-widget-open-meteo-nextjs
  https://freeapi.watch/open-meteo/

What it does

A search tool and an instruction not to answer from memory. What makes the result usable is not the prose: it is that response.sources carries the URLs the run retrieved, so you can check the answer against them. In 1.0.0 these are separate fields — .sources is everything the search returned, .citations is what the answer referenced.

What the run shows

The URLs are the ones the tool returned on this run. Run it tomorrow and they will be different, because the web is.
WebSearch defaults to a backend that needs no API key, so this example runs with nothing configured.
The system prompt is what keeps the model on the retrieved text. Without it, a model will happily answer from what it remembers.

The full script

The program above is the short version. The one in the repository at examples/web_retrieval/web_agent.py covers more cases and takes a --model flag. It ships with the package, so it runs from the command line without cloning anything:

effgen examples run web_retrieval/web_agent

Read it on GitHub

All 6 examples Next: Data analysis