AI matches incoming questions to your knowledge base and sends personalized replies. Monitored mode keeps your team informed while responses go out in seconds.
Support agents spend a disproportionate amount of time answering the same questions repeatedly — questions that already have documented answers in your knowledge base. This creates response delays for customers and wastes skilled agent time that could be spent on complex, high-value issues. The disconnect between your documentation and your inbox is a staffing problem disguised as a knowledge problem.
MultiMail's AI agent reads incoming emails, extracts the core question, and matches it against your FAQ database or knowledge base using RAG. It generates a personalized, natural-sounding reply and sends it automatically under monitored oversight. Your team can review the response log for quality assurance without being a bottleneck on every message.
Connect your existing FAQ content, help docs, or knowledge base to your AI agent. The agent uses this as the source of truth for generating responses.
Your agent uses check_inbox and read_email to process new messages. Each email is parsed to extract the core question or questions being asked.
The AI searches your knowledge base for the most relevant answer, then composes a personalized reply that addresses the specific question with accurate, up-to-date information.
The reply is sent via reply_email under monitored oversight. Every response is logged for review, and the agent tags the email with the matched FAQ category for analytics.
Your team reviews the response log periodically. Unanswered or low-confidence matches are flagged for human follow-up, and new Q&A pairs are added to the knowledge base.
import requests
API = "https://api.multimail.dev/v1"
HEADERS = {"Authorization": "Bearer mm_live_xxx"}
"cm"># Poll for new emails
inbox = requests.get(
f"{API}/mailboxes/support-mailbox-id/inbox",
headers=HEADERS,
params={"status": "unread", "limit": 20}
).json()
for msg in inbox["emails"]:
email = requests.get(
f"{API}/emails/{msg[&"cm">#039;id']}", headers=HEADERS
).json()
# Extract the core question
question = extract_question(email["text_body"])
# Match against knowledge base
match = search_knowledge_base(question)
if match["confidence"] > 0.85:
send_faq_reply(email, match)
else:
flag_for_human_review(email)Check for new emails and extract questions for knowledge base matching.
import requests
API = "https://api.multimail.dev/v1"
HEADERS = {"Authorization": "Bearer mm_live_xxx"}
def send_faq_reply(email: dict, match: dict):
"cm"># Generate personalized reply from FAQ match
reply_text = generate_reply(
question=email["text_body"],
answer=match["answer"],
customer_name=email["from_name"]
)
"cm"># Send reply via MultiMail
response = requests.post(
f"{API}/reply",
headers=HEADERS,
json={
"in_reply_to": email["id"],
"text_body": reply_text,
"html_body": format_html(reply_text)
}
)
"cm"># Tag for tracking
requests.post(
f"{API}/emails/{email[&"cm">#039;id']}/tags",
headers=HEADERS,
json={"tags": ["faq-auto-reply", match["category"]]}
)
print(f"FAQ reply sent: {response.json()[&"cm">#039;id']}")Generate a natural reply from the matched FAQ entry and send it through MultiMail.
"cm">// Using MultiMail MCP tools for FAQ responses
"cm">// 1. Check for new questions
const inbox = await mcp.check_inbox({
mailbox_id: "support-mailbox-id",
status: "unread",
limit: 10
});
for (const msg of inbox.emails) {
"cm">// 2. Read the full email
const email = await mcp.read_email({ email_id: msg.id });
"cm">// 3. Match against your knowledge base (your AI logic)
const match = await matchKnowledgeBase(email.text_body);
if (match.confidence > 0.85) {
"cm">// 4. Tag and reply
await mcp.tag_email({
email_id: email.id,
tags: ["faq-auto-reply", match.category]
});
await mcp.reply_email({
email_id: email.id,
text_body: generateReply(match, email.from_name)
});
} else {
await mcp.tag_email({
email_id: email.id,
tags: ["needs-human-review"]
});
}
}Handle FAQ responses using MultiMail MCP tools in an AI agent.
Customers get accurate answers in seconds instead of waiting hours for an agent to type out the same response they've written a hundred times before.
Every response is generated from your canonical knowledge base, eliminating the inconsistency that comes from different agents paraphrasing documentation differently.
When the AI handles 70% of incoming questions automatically, your human agents can focus on the nuanced issues that require judgment, empathy, and creative problem-solving.
Low-confidence matches reveal questions your knowledge base doesn't cover yet. Every unanswered question is an opportunity to improve your documentation.
Every auto-reply is logged with the matched FAQ entry and confidence score. Review quality over time and identify categories where the AI needs better training data.
Email infrastructure built for AI agents. Verifiable identity, graduated oversight, and a 38-tool MCP server. Formally verified in Lean 4.