Integration · CrewAI

CrewAI.

CrewAI exposes tools to its agents via the @tool or BaseTool patterns. Wrap the Agent Cert REST API as a tool and your crew can certify any decision before returning it.

Define the tool

from crewai_tools import BaseTool
from pydantic import Field
import os, requests
from typing import List, Optional

API_BASE = os.environ["AGENT_CERT_API_BASE"]
API_KEY = os.environ["AGENT_CERT_API_KEY"]

class CertifyDecision(BaseTool):
    name: str = "certify_decision"
    description: str = (
        "Certify an AI decision with Agent Cert. Returns a signed, anchored "
        "Decision Receipt with a public verify URL. Call this for any "
        "decision the crew makes that requires auditability."
    )

    def _run(
        self,
        agent_id: str,
        user_intent: str,
        agent_output: str,
        tools_used: List[str],
        retrieval_summary: Optional[str] = None,
    ) -> dict:
        r = requests.post(
            f"{API_BASE}/v1/receipts",
            headers={"X-API-Key": API_KEY},
            json={
                "agent_id": agent_id,
                "platform": "crewai",
                "user_intent": user_intent,
                "agent_output": agent_output,
                "tools_used": tools_used,
                "retrieval_summary": retrieval_summary,
            },
        )
        r.raise_for_status()
        return r.json()

Add the tool to an agent

from crewai import Agent

reviewer = Agent(
    role="Compliance Reviewer",
    goal="Approve invoices under $5,000 from approved vendors only.",
    backstory="You triple-check every approval against policy.",
    tools=[CertifyDecision()],
    verbose=True,
)

Wire the tool into the agent's task

In your task description, explicitly instruct the agent to certify the outcome:

from crewai import Task

task = Task(
    description=(
        "Review the invoice for vendor {vendor} at amount {amount}. "
        "Decide approve or deny. After deciding, call certify_decision "
        "with the user_intent, your decision as agent_output, and the "
        "list of tools you used. Include the resulting verify_url in "
        "your final response."
    ),
    expected_output="An approval decision plus a verify_url from Agent Cert.",
    agent=reviewer,
)

Multi-agent flow

For a crew with multiple agents (e.g. researcher → reviewer → writer), typically only the final agent calls certify_decision — that's the agent making the ultimate decision. Intermediate agents leave a trail you can reference in retrieval_summary.

Costs