CrewAI Integration

The signalstack-crewai package provides a SignalStackTool for CrewAI agents. Add verification as a crew task, and let your agents automatically check claims, businesses, documents, and media before acting on information.

Installation

pip install signalstack-crewai

# With all extras
pip install "signalstack-crewai[async]"

Basic usage

from signalstack_crewai import SignalStackTool
from crewai import Agent, Task, Crew

# Create the verification tool
verify_tool = SignalStackTool(
    api_key="ssk_live_YOUR_API_KEY",
    min_trust_score=0.7,
)

# Create a fact-checker agent
fact_checker = Agent(
    role="Fact Checker",
    goal="Verify all claims and data before the crew makes decisions",
    backstory="I specialize in using SignalStack to verify information across multiple sources",
    tools=[verify_tool],
    allow_delegation=False,
    verbose=True,
)

# Create a researcher agent that depends on verified info
researcher = Agent(
    role="Researcher",
    goal="Gather and analyze verified information",
    backstory="I only work with verified facts",
    tools=[verify_tool],
)

# Tasks
verify_task = Task(
    description=(
        "Verify the following claim using SignalStack: "
        "'The Great Wall of China is visible from space with the naked eye'"
    ),
    expected_output="A trust score and verdict with supporting evidence",
    agent=fact_checker,
)

research_task = Task(
    description="Summarize the verified facts about the Great Wall of China",
    expected_output="A bullet-point summary based on verified information",
    agent=researcher,
)

# Create and run the crew
crew = Crew(
    agents=[fact_checker, researcher],
    tasks=[verify_task, research_task],
    verbose=True,
)

result = crew.kickoff()
# The fact checker will call SignalStack to verify the claim
# Agent determines: Trust score 0.04 — Verdict: False
# Researcher then uses the verified result for their summary

Sequential verification pipeline

from signalstack_crewai import SignalStackTool
from crewai import Agent, Task, Crew, Process

verify = SignalStackTool(api_key="ssk_live_...", min_trust_score=0.8)

screener = Agent(
    role="Content Screener",
    goal="Screen all content before it reaches the writing team",
    tools=[verify],
)

writer = Agent(
    role="Writer",
    goal="Write accurate content based on verified facts",
)

screen_task = Task(
    description="Verify this document: https://example.com/article.pdf",
    expected_output="Authenticity analysis and trust score",
    agent=screener,
)

write_task = Task(
    description="Write a summary of the verified document",
    expected_output="A one-paragraph summary",
    agent=writer,
)

crew = Crew(
    agents=[screener, writer],
    tasks=[screen_task, write_task],
    process=Process.sequential,
)
crew.kickoff()

Configuration

Option	Type	Default	Description
`api_key`	str	`SIGNALSTACK_API_KEY` env	API key
`min_trust_score`	float	`0.8`	Minimum acceptable trust score
`on_threshold_fail`	str	`"raise"`	`"raise"` or `"return_evidence"`
`timeout`	int	`30`	Request timeout in seconds
`max_retries`	int	`3`	Max retries for transient errors
`cache_results`	bool	`True`	Cache results in memory to avoid duplicate calls

Tool input schema

The tool accepts a JSON string with the following fields:

Field	Type	Required	Description
`action`	string	Yes	Verification type: `verify_claim`, `verify_business`,`verify_document`, `verify_media`
`params`	object	Yes	Endpoint-specific parameters (same as API)

Error handling

from signalstack_crewai import SignalStackTool, TrustThresholdError

tool = SignalStackTool(
    api_key="ssk_live_...",
    min_trust_score=0.8,
    on_threshold_fail="return_evidence",
)

{`# When trust score is below threshold, the tool returns:
# {
#   "error": "Trust threshold not met",
#   "trust_score": 0.04,
#   "threshold": 0.8,
#   "evidence": [...],
#   "message": "The claim is not supported by authoritative sources"
# }`}