How do I add adverse media screening to an AI agent workflow?

Steps

1. Install crewai-strale.
2. Create the toolkit with your Strale API key.
3. Give a CrewAI agent the adverse-media-check tool.
4. Run a task and read risk_level, the hit counts and top_articles.

The solution

# pip install crewai-strale
from crewai import Agent, Task, Crew
from crewai_strale import StraleToolkit

toolkit = StraleToolkit(api_key="sk_live_YOUR_KEY")
tools = [t for t in toolkit.get_tools() if t.name == "adverse-media-check"]

agent = Agent(
    role="Media Screening Analyst",
    goal="Screen entities for adverse media coverage",
    backstory="You report what the screening found, with the sources, and leave the decision to a person.",
    tools=tools,
    llm="gpt-4o",
)

task = Task(
    description="Screen 'Acme Corporation' for adverse media such as fraud, lawsuits and regulatory actions",
    expected_output="The findings with their sources",
    agent=agent,
)

crew = Crew(agents=[agent], tasks=[task])
result = crew.kickoff()
print(result)

How it works

The tool searches Dilisense's news index, which is organised by financial-crime category. If Dilisense is unavailable (quota exhausted, a server error or a missing key), it falls back to a web search. The answer says which path produced it in source, and results from the fallback are weaker evidence.

The answer has risk_level, total_hits, a categories breakdown and top_articles, each with a headline, category and source link. The risk level follows fixed rules, and the thresholds are in risk_level_thresholds in the answer. No language model makes the call.

The search window is the last 12 months, and the company needs to be searched by its registered name. For a plain-language summary of several checks, run risk-narrative-generate on the results. That tool uses a language model, and this one does not.

A hit means news coverage matched the name. It is not a finding about the person or company: a person should read the articles and decide.

Tools in this guide