> ## Documentation Index > Fetch the complete documentation index at: https://feddy.app/docs/llms.txt > Use this file to discover all available pages before exploring further. # Milo > Milo, the AI that answers users from your answers and your docs, and the Copilot that drafts replies for you in the inbox. Milo is the AI in Feddy. It does two jobs. It answers users on its own, within limits you set, and it helps you write replies in the inbox. It never makes up facts about your product: everything it says to a user comes from your answers, which you write or it learns from your team's replies, or from the docs and text you added as sources. You'll use it to: * Answer the questions you get over and over, without touching them. * Let users know they've been heard while they wait for you. * Draft, shorten, translate, or soften a reply before you send it. Its settings live under **Settings → Milo AI**. ## Answers An answer is the reply you'd give to a common question, saved once so it can be reused. Each one has a title for your own reference, and optionally keywords, a topic, and a platform that narrow when it applies. Write them under **Settings → Milo AI → Answers**, or import a batch from the same page. ## Sources A source is material Milo reads to answer questions your answers don't cover. Add one under **Settings → Milo AI → Sources**, where each kind has its own add button: * **Website**: the address of your docs or help center. Milo reads that page and the pages under it on the same site, and reads them again every week. Click **Refresh** on its row to read them again now. * **App Store**: the link to your app on the App Store. Milo reads the description, the release notes of the latest version, and which devices, system versions, and languages the app supports. Prices are left out, and lines in the description that state a price are skipped. The listing is read from the country in the link and again every week, or now with **Refresh**. * **Text**: anything you paste, or a `.md` or `.txt` file you drop in. Only the text is kept, not the file. Milo writes replies from your answers and sources in its own words and never sends a passage as is. When a source and one of your answers disagree, the answer wins. If nothing covers a question, Milo hands the conversation to you. Each source has a **Used by** setting: * **Milo and team**: Milo can use it in replies to users. Anything in it may end up in a reply, so add only what any user may read. * **Team only**: only the Copilot reads it, for drafts you review before sending. Use this for internal notes. Open a source to see every passage Milo reads from it. If a site draws its pages with JavaScript, there's no text to read and the source shows why. Add the text instead. ## About your app **Settings → Milo AI → Settings** has an **About your app** box: a short description Milo reads before every reply. Fill it in. It's how Milo knows what your product is and what it isn't. For a first draft, click **Draft from a link** and paste your app's App Store link or your website's address. Milo reads that one page and writes a short paragraph in English, without prices, into the box. Nothing is saved until you click **Save**, so edit it first: add what the page leaves out, and remove anything you don't want Milo to say. **Also read this page for answers**, ticked by default, adds the same link as a [source](#sources) too, so Milo can answer from it. If it's already a source, nothing changes. ## Modes Choose how Milo replies to users under **Settings → Milo AI → Settings**: | Mode | What users get | | - | - | | Keyword answers only | An answer whose keyword appears in their message. Anything else gets the fallback message. | | Answer with AI | A reply written from your answers and sources, using what's known about their device and app version and any screenshot they attached. A message too short to answer gets one question back. Follow-ups get replies too, up to three per conversation. If part of a message isn't covered, Milo answers the rest and leaves that part to you. | | Suggest replies to me | Nothing automatic beyond keyword answers. Instead, Milo drafts a reply for you in the inbox. | New projects on a paid plan start in **Answer with AI**; on the Free plan, in **Keyword answers only**. Turn **Enabled** off to send nothing automatically. Whatever the mode, Milo never changes a conversation's status and never marks it resolved. Only you, or the user, can close a conversation. Users see replies under the name you give Milo in **Name**. New projects start with "Milo". ## Style **Style** sets how replies sound: **Tone** (friendly, neutral, or playful), **Length** (short or detailed), **Formality** (casual or formal), and **Emoji** (never or sometimes). Each starts at **Auto**, which follows the usual way support writes in the user's language. Changes apply to the next reply. ## Rules **Rules** decide what Milo does in three common cases. Each is a switch: | Rule | On | Off | | - | - | - | | Complaints: acknowledge only | A complaint about price or content gets one sentence that names it, and nothing more. This is the default. | Milo also says what your answers and sources say about it, such as what's free, without prices. | | Bugs: ask one question first | When your answers don't explain the problem, Milo asks one question, such as what they tapped. This is the default. | The conversation goes straight to you. | | Refunds and account problems: always hand off | Refunds, disputed charges, and account problems always go to you. | Milo explains how refunds work from your answers. This is the default. | For anything else, use **Other instructions**. ## Other instructions Free text Milo follows on every reply: what to call your product, things it should never say, anything **Style** doesn't cover. Where they differ from **Style**, these win. Keep facts in your answers and use instructions for behavior. ## Handing over to you Milo steps back and leaves the conversation to you when the user asks for a person, when nothing in your answers or sources covers the question, after answering only part of a message, or after its third reply. If it hasn't replied yet, the user gets the fallback message. If it has, it tells them in its own words that you'll take it from here. After that it stays quiet in the conversation, even if a later message is one it could answer. When it hands over, Milo leaves you an internal note in the conversation: what the user wants, what's already known (their device, version, plan, what their screenshots show), what Milo told them, and what's still missing. The user never sees it. The note is written in the language your team replies in. ## Fallback message The fallback message goes out when a new conversation gets no answer, so the user knows they've been heard and roughly when to expect you. The default is a thank-you with a 24 to 48 hour expectation; change it to match your real response time. Leave it empty to send nothing. ## What Milo did Under each of Milo's replies in a conversation, a one-line card says what it did and why: which answer it used, or what was missing that made it ask a question or hand the conversation to you. Open the card to see the answers and pages it drew on (**Based on** links the page), how it read the message, and whether the user found it helpful. The **Milo AI** page lists the past week or month: how many times Milo replied, what users said about the replies, the conversations it handed to you grouped by reason with an example of each, any time it could not reach the AI service, what it learned from your replies, and new [requests](https://feddy.app/docs/concepts/requests.md) from users. Every Monday, everyone on the project also gets the same list by email. Turn it off under **Account → Notifications → Weekly Milo summary**. If the AI service doesn't respond, the user gets your fallback message, the conversation is left to you with a note saying why, and the week's list counts it. ## Was this helpful? After one of Milo's replies, the user sees a **Was this helpful?** prompt. Yes closes the conversation as resolved. No leaves it open and flags the reply in your inbox so you can step in. ## Learn from your replies When this is on, which it is by default, Milo reads what your team replies to users and keeps what the next user asking the same thing should hear. It saves it to your answers and uses it right away. There is nothing to review. * It learns only what holds for everyone. A refund you granted or "it works now" stays with that one user. * It keeps your steps and conditions as you wrote them, and doesn't turn them into a yes or a no. * When your team's answer changes, it updates the answer it already has instead of adding a second one. Your newest reply wins, also over an answer you wrote by hand, unless you edited that answer after the reply. * Something temporary, like an outage or a workaround until your next release, stops being used after two weeks, or from the app version you said fixes it. Learned answers are marked **Learned** in your answers list. Each answer page has a history of every change, and any earlier version can be restored. In a conversation, Milo's card shows which answers a reply used; next to one it learned, **Undo** takes back its latest change. Editing a learned answer makes it yours. Questions Milo couldn't answer, and hasn't learned an answer for since, are listed under **Unanswered**, each with a button to write the answer. ## Copilot Next to every conversation, the **Copilot** tab has already read the thread, its screenshots, your notes, your answers, and your sources, including the ones for your team only. Ask it to draft a reply, summarize, translate into the user's language, or soften the tone. What it writes goes into your reply box, where you edit it and send it yourself. Nothing reaches the user until you press send. Copilot answers up to 30 times per conversation, counting everyone on your team. ## Try it **Settings → Milo AI → Settings** has a phone on the right where you can talk to Milo the way a user of your app would: several messages in a row, in any language, with a screenshot if you like. It uses your settings as saved, so a change on the left applies to your next message. Under each reply, a line says what Milo did; click it to see the answers and pages it used and its note for you. Once Milo hands the conversation over, click **Clear** to start again. Nothing is sent to anyone, and it doesn't count toward your plan's AI replies. ## Availability AI answers, sources, and the Copilot are available on paid plans. Pro includes 5 sources and 500 pages per project, Business 20 sources and 3,000 pages; pasted text and an App Store listing each count as one page. Keyword answers and the fallback message work on every plan. ## Related guides > [card] [Conversations](https://feddy.app/docs/concepts/conversations.md) > > What "Was this helpful?" does to a conversation's status. > [card] [Projects](https://feddy.app/docs/concepts/projects.md) > > Where Milo's settings live inside a project. > [card] [Team](https://feddy.app/docs/concepts/team.md) > > Saved replies for the answers you'd rather send by hand. > [card] [Troubleshooting](https://feddy.app/docs/troubleshooting.md) > > What to do when Milo answered something it shouldn't have. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.