Frameworks and agents · Published 2026-09-29 · Updated 2026-11-06 · AgentiSend
Send email from CrewAI
Give a CrewAI agent one tool that posts to POST /emails with its own budgeted key, and stop when the reply is agent_budget_exceeded.
To send email from CrewAI, give the crew one tool whose function posts to POST /emails with the agent's own key. The key is sending_access and carries a budget. The file below is that tool, called through run, which is the path a crew uses. It is run by the test suite against a local AgentiSend server: the first call sends, and the second message is refused with agent_budget_exceeded. The run does not call a hosted model.
Install
pip install crewaiThe agentisend package is standard library only, so pip install agentisend adds no other dependency.
The tool and the loop
"""CrewAI — one tool that posts to POST /emails, and a scripted agent loop.
`tool.run` is the path a crew uses. The first call sends. The second is a
different message, so a key whose budget is one comes back as
agent_budget_exceeded. No model is called.
Install both packages with `pip install agentisend crewai`. The suite runs
this file against the SDK checkout; a fresh project uses the released package.
AGENTISEND_API_KEY is the agent's key, with sending_access and a budget.
MAIL_FROM is an address on a domain you have verified.
AGENTISEND_BASE_URL is optional.
"""
from __future__ import annotations
import json
import os
from crewai.tools import tool
from agentisend import AgentiSend, AgentiSendError, idempotency_key
agentisend = AgentiSend()
TO = "crewai@example.com"
UPDATE = {
"to": TO,
"subject": "Ticket 4182 — we are looking into it",
"text": "Someone from support will reply within the hour.",
"purpose": "ticket-4182-update",
}
CLOSED = {
"to": TO,
"subject": "Ticket 4182 — closed",
"text": "The ticket is closed.",
"purpose": "ticket-4182-closed",
}
@tool("send email")
def send_email(to: str, subject: str, text: str, purpose: str) -> str:
"""Send one email to a person who asked for it.
Returns the message id, or a refusal with a code and a fix. When the fix
says a person decides, stop and report it: calling again will not change
the answer.
"""
try:
sent = agentisend.send_email(
{"from": os.environ["MAIL_FROM"], "to": to, "subject": subject, "text": text},
idempotency=idempotency_key(purpose, to),
)
except AgentiSendError as err:
return json.dumps({"sent": False, "code": err.code, "fix": err.fix})
return json.dumps({"sent": True, "id": sent["id"]})
def run() -> None:
"""Two tool calls, then stop when the second is refused."""
for call in (UPDATE, CLOSED):
raw = send_email.run(**call)
result = json.loads(raw)
print(json.dumps(result), flush=True)
if result.get("sent") is False:
return
if __name__ == "__main__":
run()The function is wrapped with CrewAI's @tool("send email") decorator and returns a JSON string, because a CrewAI tool's result goes back to the agent as text. A refusal is caught as AgentiSendError and returned as {"sent": false, "code", "fix"} rather than raised, so the crew reads agent_budget_exceeded and its fix as the tool's answer. The loop in run() stops on that answer instead of calling the tool again.
The key
The crew's key is minted with sending_access and given its budget before the crew starts; An AI agent with a budget and a loop guard runs those calls. One key per crew keeps one crew's spend from eating another's budget.