Docs › CrewAI with RunVouch

CrewAI with RunVouch

A crew has max_iter per agent and max_rpm; neither is a budget. A nightly crew that keeps re-planning burns tokens until the iteration limit, then returns whatever it has.

How it runs on CrewAI

Around crew.kickoff() with the Python client, plus run.tool() from a step callback for loop detection.

Store the key

An environment variable on the host or container that runs the crew.

Wrap the job

import runvouch
from crewai import Crew

def on_step(step):
    run.tool(step.tool or "llm", {"input": step.tool_input or ""}, cost=getattr(step, "cost", 0))

crew = Crew(agents=[...], tasks=[...], step_callback=on_step)
with runvouch.vouch("weekly-competitor-brief", evidence=lambda: {"brief": len(out.raw) > 500}) as run:
    out = crew.kickoff()

rv fails open: if RunVouch is unreachable the job still runs and you get one warning line.

Register the cadence and caps

rv agent nightly-report --cadence 24h --grace 30m --max-runtime 1h --evidence --cap-run-cost 2

Register the agent once, from anywhere with the key. Cadence is what turns a schedule that stopped into an alert; --evidence makes a run without evidence a failure; the caps pause the agent when it overspends.

What goes silent on CrewAI

• Reaching max_iter is not an error; the crew returns a partial answer. Evidence (length, file, URL) is what fails it.
• Per-agent token usage is available in crew.usage_metrics; pass it as cost so the daily cap works.

What RunVouch detects

Need a key? Get a free key · Stuck? contact