AI governance for CrewAI — task guardrails, step callbacks, and audit trails for every crew run.

CrewAI Integration

Palveron integrates natively with CrewAI's guardrail system. Every task output is checked against your governance policies before being passed to the next task. Blocked outputs trigger automatic retries with feedback.

Installation

pip install palveron-crewai

Quickstart: Task Guardrail

from crewai import Agent, Task, Crew
from palveron_crewai import palveron_guardrail

guardrail = palveron_guardrail(api_key="pv_live_xxx")

report_task = Task(
    description="Produce a customer analytics report",
    expected_output="A report with customer insights",
    agent=analyst_agent,
    guardrail=guardrail,           # Palveron governance
    guardrail_max_retries=2,       # Retry on block
)

If the agent's output contains PII or violates your policies, CrewAI automatically initiates a retry with Palveron feedback.

Step Callback (Crew-wide)

For governance across all agent steps, not just individual tasks:

from palveron_crewai import PalveronStepCallback

callback = PalveronStepCallback(api_key="pv_live_xxx")

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, write_task],
    step_callback=callback,
)

result = crew.kickoff(inputs={"topic": "AI safety"})
print(f"Checked: {callback.check_count}, Blocked: {callback.blocked_count}")

Combined Governance

Convenience class for both patterns together:

from palveron_crewai import PalveronCrewGovernance

gov = PalveronCrewGovernance(api_key="pv_live_xxx")

sensitive_task = Task(..., guardrail=gov.task_guardrail())
crew = Crew(..., step_callback=gov.step_callback())

result = crew.kickoff(inputs={...})
print(gov.summary())  # {"checks": 12, "blocked": 1, "trace_ids": [...]}

Decision Behavior

Decision	Guardrail returns	Effect
`ALLOWED`	`(True, output)`	Task continues normally
`MODIFIED`	`(True, sanitized)`	Continues with PII masked
`BLOCKED`	`(False, feedback)`	CrewAI retries with feedback

Configuration

guardrail = palveron_guardrail(
    api_key="pv_live_xxx",
    base_url="https://gateway.internal.company.com:8080",
    fail_open=False,
    metadata={"department": "finance"},
)

Source Code

Open source (MIT): github.com/palveron/adapter-crewai.

Next Steps

Create your own policies for your crew workflows
LangChain integration for LLM-level governance
OpenClaw integration for personal AI agents

CrewAI

On this page