Milo, the no-code AI agent, reads your help articles, your website, and the notes you give it, and answers customer emails and live chats from that. When it doesn't know, or the customer wants a person, it says so and moves the conversation to your team. You pay per resolved conversation, with a free monthly allowance.
Test Milo, the AI agent, against the available sources before a single customer sees it. No complex flows and no developer needed, and nothing goes live until you activate it.
Choose the channels the AI agent may reply in: live chat, email, or both. Switch on live chat first if you'd rather watch it handle website visitors before it answers your support inbox. Turning it on for email means it works for every shared inbox you've connected.
Before anyone outside your team meets Milo, ask it the questions your customers ask most: delivery times, what's in which plan, the return policy. If an answer is missing or wrong, the fix is in your sources - add the right document or re-sync the page, then ask again.
With one click, the AI agent starts taking every new, unassigned conversation in the channels you pick. Each reply is labeled as Milo's and sits in the thread, so you can read along from day one, and the sidebar shows how many conversations it's holding right now. If something looks off, assign the conversation to yourself, and Milo steps aside.
about 260
hours saved monthly
99%
faster first response
up to 60%
conversations resolved
The no-code AI agent is a separate add-on to Inbox, billed monthly by resolutions. A resolution is a conversation Milo AI closed because the customer said it helped, or didn't come back. One per conversation per billing month, and never for a conversation a teammate took over or you marked as spam. Every workspace gets a free monthly allowance, and when it's used up, new conversations wait for your team, and the AI agent resumes next month or on a higher tier.
Answers come only from the files, pages, and notes you added, never from the open internet or whatever the model happens to remember. If the answer isn't in your sources, the AI agent says it doesn't know and offers a person instead of guessing confidently.
Every reply is labeled as Milo's and sits in the thread like any teammate's, so you can read the whole exchange afterward and see exactly what was said. The customer knows too: in live chat, Milo's first message is signed as Milo AI, and a person can take over.
"Talk to a person" sits under every reply in live chat. Milo AI also offers a person when it has no answer or the conversation goes in circles. The moment a teammate takes over, the AI agent stops replying in that conversation for good, even if the customer writes again first.
Switch Milo AI off for live chat, for email, or entirely, at any time. Switch both channels off to pause it. Two permission levels decide who can see its settings and who can change its sources and activation, so a new hire can read along without being able to retrain it.
It doesn't answer from the open internet or make things up. If it's not in your sources, Milo AI says it doesn't know and hands over.
It doesn't pretend to be a person. Replies are labeled as Milo's.
It doesn't keep a customer from reaching you. “Can I talk to someone?” is a handover, every time.
It doesn't act in other systems - no refunds, no account changes. It answers questions.
It doesn't learn from your customers' messages to answer other companies' customers.
Before Milo AI answered a customer of yours, it answered ours - in Elastic Email's 24/7 support, where it works today. So far, we're saving on average 260 hours on customer support monthly, so our team can focus on more advanced inquiries.
Try for Free
Product
Resources
© Elastic Email Inc. 2026
