Privacy Policy

• You can chat without an account. A guest’s conversations are kept in their own browser, apart from a copy of each reply that our server holds until the browser collects it.
• With an account, your conversations, memories and personalization are kept on our servers, so they follow you from browser to browser.
• What you send in a chat goes to the AI providers that write the reply: OpenRouter, Google and Anthropic. See How AI replies are made.
• We don’t sell your data, show ads, or train AI models on what you send.
• You can delete a conversation, your memories or your whole account in the app.

Your account. To create an account you give a name, an email address and a password. We keep the password only as a one-way hash, never as you typed it. We don’t check that the email address is yours. The only email we send is a link to reset the account’s password, to its email address, when someone asks for one on the sign-in page. If you sign in with Google or GitHub instead, where we offer it, we receive your profile from them, and from GitHub the list of your email addresses. We keep your name, one email address, the address of your profile picture, whether they verified that email, your account ID there and the sign-in tokens they issue.

What you write and add. Your messages, including earlier versions of ones you edit, the replies, conversation titles, files you attach and pictures the app makes for you. Also what you choose to tell the app about yourself in Settings (what it should call you, what it should know about you and how it should respond), the memories you ask it to save, the stocks you follow, and any topic you ask the encyclopedia to write about.

How much you use. For each reply and title, and each Discover article, stock analysis or part of the market summary your visit causes to be written, we record when it ran, what it cost and how many tokens it used, so we can apply the daily limits and show your usage. These records hold no words from your conversations.

Your device and network. When you sign in, we keep your IP address (shortened, if it is an IPv6 address) and your browser’s user agent with the session, and we count attempts to sign in, sign up or reset a password by IP address to stop abuse. A guest’s usage records hold a keyed, one-way hash of their IP address, never the address itself, which is how the guest daily limit is applied. To limit how often anyone can call the app, our servers also hold IP addresses briefly in memory, never in the database; for the encyclopedia, that includes which articles each address read that day, so each reading is counted once. Our host tells us the approximate city, coordinates and country your IP address is in. We use them for the weather on the home page and Discover, to choose °F or °C, and to tailor search suggestions to your country.

Location, only if you turn it on. If you turn on Location in Settings, each message you send carries your device’s position, rounded to about 1 km, and your time zone, so replies can use them. If you choose “Use precise location” on a weather card, or turn on Precise weather, your browser sends your position, rounded to about 11 km, to get your local forecast.

What you type before you send it. In a new, empty chat, what you type is sent to our server as you type, to offer search and stock suggestions, even before you press Enter. This doesn’t happen in an incognito chat, once a conversation has started, or with a file attached.

We use what we collect to run the app: to write replies, keep your conversations in sync, apply daily limits, keep accounts secure, let you reset a forgotten password, find and fix errors, count page views, and see how quickly pages load. Discover feeds are put together on our server from the interests you pick, which your browser sends with each request; we don’t keep them.

We don’t sell or rent your personal information, we don’t show ads, and we don’t use what you send to train AI models.

Replies are written by AI models we reach through OpenRouter. To write a reply, our server sends OpenRouter your message and as much of the conversation as fits, with its attachments. For an account, it also sends what you told the app about yourself in Settings, your saved memories (unless memory is off, the chat is incognito, or the conversation is private), the names of files in the conversation’s workspace, and your location if you turned it on. It doesn’t send your name from your account, your email address, your IP address or your account ID, but anything you write in a message is sent as written.

The models behind the app, which may change, are:

• Google’s Gemini models, for replies, web searches, pictures and encyclopedia articles. They run on Google Vertex AI, and on Google AI Studio only when Vertex AI fails.
• Anthropic’s Claude models, for a conversation’s title, made from its first message, the names of any files attached to it and the start of the reply; and for an advisor the reply’s model may consult, which doesn’t see the conversation, only the question that model writes for it, though that question can repeat what you wrote.

According to the provider details OpenRouter publishes, Google Vertex AI neither trains on prompts nor keeps them, and Google AI Studio doesn’t train on them but keeps them for up to 55 days. OpenRouter handles requests under its own privacy policy.

When a reply searches the web, the search the model writes is run on Google Search through Gemini. When it reads a web page, OpenRouter passes the page’s address to Exa, a service it uses to read pages, or to the model’s own provider where that provider reads pages itself. To show when a cited page was published, our server fetches the page itself, sending nothing from your conversation.

Files you attach reach the model through OpenRouter, which fetches most of them from their link. An account’s conversation also has a workspace at OpenRouter where the model can run code, and files you attach can be copied there. OpenRouter keeps that workspace’s files for 30 days after the conversation was last used. We can’t delete them sooner, even when you delete the conversation or your account.

A topic you ask the encyclopedia for is sent to Google’s Gemini, which searches Google to write the article. Until it is written, anyone can see the topic as you typed it on its page, without your name.

GitHub. If you connect GitHub in Customize, the app can read the repositories its GitHub App is installed on that your GitHub account can see, including an organization’s where you or someone else installed it, and only when a chat calls for it. It asks GitHub for read access to contents and issues and can’t change anything on GitHub. We store the tokens GitHub issues encrypted, with your GitHub account ID, and renew the tokens while you keep using the connection. Once a chat reads GitHub it becomes private for good: it gets no web search, pictures, code running, memories or attachments, and what it reads from GitHub is sent through OpenRouter to Google’s models as described above. Nothing a chat reads from GitHub goes into its title.

Disconnect asks GitHub to revoke the app’s access, and we delete the tokens and your GitHub account ID once GitHub confirms. If GitHub can’t confirm it, you can remove the app in GitHub’s settings yourself and ask us to clear the connection by hand; we then keep a note of that, with your GitHub account ID, until you delete your account. The app stays installed on GitHub until it is uninstalled there. Private chats, whose replies can quote your repositories, stay in your history until you delete them. Patches the app proposes in a chat are saved; you can delete them in that chat, and deleting the chat doesn’t delete them, so otherwise they stay until you delete your account.

Notion. Where we offer it, you can connect one Notion workspace. If you do:

• the app reads only the pages and databases you share with it in Notion, only when a chat calls for it, and it can’t change anything in Notion; to find pages, it sends Notion search words the AI model writes from your chat;
• what it reads is sent through OpenRouter to the AI model that answers you, as described above;
• we don’t use content from Notion to train AI models;
• we store the tokens Notion issues encrypted, and delete them once Notion confirms they are revoked, when you disconnect or delete your account;
• a chat that reads Notion becomes private and stays private after you disconnect, and its replies, which can quote your pages, stay in your history until you delete that chat or your account;
• Disconnect revokes the app’s access, and you can also remove the app in Notion, under Settings, Connections.

We share data only with the services that run the app for us, only for that, and each under its own terms:

• Vercel hosts the app, stores the files you attach and an account’s pictures, keeps a copy of each finished reply until your browser has it, and counts page views and measures loading speed (Vercel Web Analytics and Speed Insights). When something fails on our server, the error written to Vercel’s logs can quote what the server was working on, such as a conversation’s text, your settings, your account ID or your session token.
• Neon hosts our database: accounts, sessions, conversations, memories, settings, followed stocks, encyclopedia requests, usage records, counts of attempts to sign in, sign up or reset a password by IP address, links to reset a password, kept only as hashes, and connector tokens and saved patches.
• OpenRouter, Google and Anthropic write replies, as described in How AI replies are made.
• Better Auth provides our sign-in system’s security service, and the dashboard we manage accounts from. It checks sign-ups, sign-ins and password resets for abuse, such as throwaway email addresses and passwords known from data breaches, so it receives the email address typed into each, whether or not an account has it, with the IP address it came from. It receives an event each time you sign up, sign in or out, change your name or password, link a sign-in method or delete your account, and each time a reset of your password is asked for or made, with your account ID, name, email address, and the IP address and country the request came from, and for signing in and out your browser. For a failed sign-in it also receives a keyed hash of the password tried, and for the breach check only the first few characters of a hash of your password. When we open your account in its dashboard, it receives your account’s records, including your sessions’ IP addresses and browsers, your password’s hash and any tokens Google or GitHub issued at sign-in. It never receives the password itself, or your conversations.
• Resend sends the email with a link to reset your password, so it receives the account’s email address and the email itself, link included, and keeps a copy of each email it sends for a period it sets.
• Sentry receives reports of errors, a record of each page load, and timings of one in ten requests to our server. They carry what went wrong in our code, your browser’s user agent, and the app’s pages and requests that led to it, such as an article or stock you had open, never your conversations, your IP address, cookies or who you are.
• Brave Search, Yahoo Finance and Nasdaq receive what you type in a new chat, to suggest searches and stocks, and what you type in the stock search. Our server sends it, adding for Brave only the country your IP address is in, so they don’t see the address itself.
• Open-Meteo and OpenStreetMap’s Nominatim receive rounded coordinates, or a place a reply names, to give forecasts and place names. Our server sends them.
• Google and GitHub handle your sign-in, if you use them, and GitHub serves the repositories you connect.

Your browser also loads some things straight from other sites, which see your IP address: icons for the sites a reply cites, from DuckDuckGo; news pictures, from the newsrooms that published them; company logos, from Financial Modeling Prep; maps, from OpenFreeMap; pictures and summaries from Wikipedia; pictures beside some search suggestions, from Brave; your profile picture, from Google or GitHub if you sign in with them; and any image a reply shows, from the site it is on. Those sites may also read or set their own cookies, under their own policies, and may recognize you from other sites; we don’t control that.

We may also disclose information when the law requires it, or to protect the app and the people who use it.

Files you attach, and an account’s pictures, are stored at long, unguessable links, so that you and the AI models can open the files and your other browsers can show the pictures. Anyone who has a link can open the file, and a file’s link includes its name. A guest’s copy of a reply, kept for their browser to collect, is stored the same way, without encryption.

Deleting a conversation asks for its files to be deleted. If that request doesn’t reach us, as can happen when you are offline, the files stay at their links, even after you delete your account. Copies that browsers or AI providers already fetched may also stay in their caches for a while.

We use no advertising or tracking cookies. The app sets only these:

• a session cookie that keeps you signed in, for 7 days from when it was last renewed;
• a short-lived cookie while you sign in with Google or GitHub, or connect GitHub, for up to 10 minutes;
• a cookie that remembers whether the sidebar is open, for 7 days.

We don’t track you across other sites, and we don’t sell or share your information for advertising, so the app works the same whether or not your browser sends a Do Not Track or Global Privacy Control signal.

The app also keeps things in your browser’s own storage: your conversations, the pictures made for you, your theme, Discover and home preferences, which Settings permissions you turned on, and the encyclopedia articles you have read on that browser. Signing out doesn’t remove them. Deleting your account on that browser removes its conversations, pictures, permissions and Discover and home preferences, but leaves your theme, the list of encyclopedia articles read there, a note of the deleted account’s ID and a note of when someone last signed in or out there, though not who, which stay until you clear the site’s data.

On a shared computer, a guest’s conversations and Discover and home preferences pass to the next account that signs in there for the first time, and the conversations and followed stocks are then saved to that account on our servers.

• Conversations are kept until you delete them or your account. An account keeps up to 500 conversations or 20 MB; past that, the ones changed longest ago are removed first, pinned ones last. A deleted conversation leaves only a marker of its ID and when it was deleted, so your other browsers delete it too. Markers are removed the next time your account saves a change after they are 90 days old, and the oldest go sooner once there are 1,000.
• Memories and settings are kept until you delete them or your account.
• Followed stocks are kept until you delete your account: a stock you stop following stays recorded as unfollowed, so your other browsers don’t add it back.
• Encyclopedia topics you ask for are kept with your account until you delete it. The topic and its article stay in the encyclopedia after that.
• Usage records are kept until you delete your account. Guests’ are kept without a time limit, under the network hash alone.
• Sessions end 7 days after they were last renewed. A session’s record, with its IP address and user agent, can stay in the database until you sign out, it is next used, or you delete your account.
• Counts of attempts to sign in, sign up or reset a password by IP address are kept for at least an hour after the last attempt, until later sign-in activity clears them.
• Links to reset a password work once, within an hour. We keep each only as a hash, with your account ID, until your password is reset or, once it has expired, until later sign-in activity clears it.
• The copy of a reply kept so it can be recovered is deleted once your browser has the reply. One your browser never collects is removed by a clean-up that runs from time to time and deletes copies more than a week old, so it can stay longer. An account’s copies are encrypted.
• Files in a conversation’s workspace at OpenRouter are kept by OpenRouter for 30 days after the conversation was last used.
• Connector tokens are kept until you disconnect and the service confirms it has revoked the app’s access, or until you delete your account.
• Error reports, server logs and database backups are kept for the periods our providers set.

• Delete a conversation from the sidebar, and your browser asks for its files to be deleted (see Files and links).
• See, edit and delete your memories in Settings, Memory, or turn memory off.
• Start an incognito chat. We don’t keep it, give it a title or remember anything from it, but it is still sent to the AI providers, its usage is still recorded, and what you told the app about yourself, and your location if you turned it on, still go with it. Files you attach to it are deleted as you leave it, unless the tab crashes or the browser keeps the page to come back to, and with an account its workspace at OpenRouter is kept like any other.
• With an account, turn off personalization, location and precise weather in Settings. As a guest, you can stop precise weather by taking back the site’s location permission in your browser.
• Disconnect GitHub in Customize, or Notion where we offer it.
• Delete your account in Settings, Account. This deletes your account, sessions, sign-in methods, kept conversations and the files they name, memories, settings, usage records and connector data. A connected GitHub or Notion has its access revoked first, and the deletion waits until that is confirmed, or, where the service can’t confirm it, until you have removed the app there yourself and we have cleared the connection by hand.

Some things outlast a deleted account: copies in other browsers, until their site data is cleared; copies of replies your browser hadn’t collected, until the clean-up removes them; files no conversation kept on our servers names, such as those of a conversation too large to keep there or whose deletion request failed, which stay at their links; workspace files at OpenRouter, for up to 30 days; the events Better Auth holds, such as your sign-ins and the deletion itself, with your name, email address and IP address; the emails Resend sent you, for as long as it keeps them; encyclopedia articles written from topics you asked for, which are public and never name you; the record of any connection we cleared by hand at your request, with your account ID, the GitHub account or Notion workspace it named and what you showed us, which we keep with our notes of that request; and what is already in our providers’ logs and backups, until they expire.

The app has no export. To ask for a copy of your data, to correct it, to delete it, or with any other question about your privacy, write to privacy@emalise.com. We may need to confirm that the account is yours first.

Connections to the app are encrypted, passwords and links to reset them are stored only as hashes, and connector tokens and an account’s copies of replies are encrypted. Your conversations, memories and settings are not end-to-end encrypted: we can reach them, and do so only to run and fix the app. No system is perfectly secure, so use a password you don’t use anywhere else.

You must be at least 18 to use Emalise, and it is not meant for anyone younger. If you believe someone under 18 has given us personal information, write to privacy@emalise.com and we will delete what we hold of it.

We store and process data in the United States. Some of our providers may process it in other countries.

When we change this policy, we update it here and change the date at the top. If you keep using Emalise after a change, the new policy applies. This policy works together with our Terms of Use.

Emalise is run by Andre Tran, an individual developer, who is responsible for the data this policy describes. For anything about this policy or your data, write to privacy@emalise.com.