Privacy at Senseline

If you're not signed in

Senseline sends us nothing except a check that our service is up. No page text, no addresses, nothing that identifies you. (The one exception is signing in itself, which exchanges a one-time code with us.)

When you're signed in, we receive

The terms found on the page. When Senseline spots an acronym or term, it sends us the term itself, such as VCU, so your organization can see which of its terms people actually run into. That includes capitalised words your glossary doesn't know yet. That's how your organization finds out what's missing.

The site, never the page. We receive the site where terms were found, like https://wiki.example.com. We never receive the path, the query or the page title. A ticket link reaches us as the site alone; the ticket number never leaves your browser. Full addresses and page titles stay in your browser and are cleared when you close the tab.

To be clear about what that means: because it's tied to your account, we can see which sites you came across terms on. We can't see which pages.

Up to 1,200 characters around a term, only when its card opens. When a term's card opens (because you rested your pointer on the term or asked what it means) and your organization has AI suggestions on, Senseline sends up to 1,200 characters of the surrounding text (about 1,000 before the term and 100 after) so it can tell which meaning fits, or suggest one for a term your glossary doesn't have. The text is used for that lookup and nothing else.

What you do with terms: which ones you open, pin, hide or follow. This keeps your settings in sync across your devices and shows your organization which terms are useful.

The tags you follow, so their terms come first for you on every device. Only you can see which tags you follow. People in your organization can see how many people follow a tag, but not who. In a small organization, that number can make it possible to guess.

What you submit to the glossary. When you define a term or propose a change to one, we receive the term, its meaning, its description, any tags you pick for it, and when you sent it. We don't keep the page or the site you were on.

What you dictate, as text. If you use the microphone in the web app, we receive the text of what you said, never the audio. “When you dictate” below has the details.

What you type into Senseline AI. If you use Senseline AI in the web app, we receive what you type or paste into it, and we may keep it. “When you use Senseline AI” below has the details.

Your email address, so we know who you are and which organization you belong to. If you're an admin, we also receive the email addresses you type in to invite people, so we can send the invitations.

A sign-in token, kept on your device, so you don't have to sign in every time.

We never receive

• The full text of the pages you read. Only the terms found, and the short snippets above.
• Page addresses, queries or titles, beyond the site.
• Your location.
• Health information.
• Your card details. Payment happens on Stripe's own page. We learn which plan you chose; we never see a card number.
• Your messages, email or files.

Anything you type or paste into Senseline AI is the exception: it reaches us because you sent it.

When AI suggests a meaning

When a term's card opens, we send our AI provider, Anthropic (the maker of Claude), the term, the possible meanings from your glossary and that snippet of up to 1,200 characters, so it can choose between the meanings. If you type your own guess at a meaning, we send that guess too, up to 200 characters.

A term your glossary doesn't have. When you rest your pointer on a term Senseline has marked as unknown, we send Anthropic the term and that same snippet, so it can suggest a meaning. The card shows the suggestion and says it was written by AI. We don't keep the snippet or the suggestion: it reaches your glossary only if someone adds it. This happens once your pointer has rested on the term for a little over half a second, and at most three times on a page. After that, the card offers a Guess with AI button instead. Your organization's admins can turn these suggestions off on their own, in Organization settings, under Suggestions (“Guesses for unknown terms”).

We send nothing that identifies you or your organization: not your email, not your organization's identity, not your sign-in token and not the site you're on. The request carries the text and nothing else. A test in our code fixes the exact set of fields that can be sent, so an identifier can't be added by accident.

When you dictate

In the web app, the New definition and New tag forms each have a microphone. It's on only while the dictation box is open, and that opens when you press the microphone.

Your browser turns your speech into text. That step is your browser's own, not ours. In Chrome, the audio goes to Google's speech service under Google's terms. It never reaches us. We receive the text of what you said, up to 1,000 characters at a time.

We send that text to Anthropic, together with what the form holds at that moment, so it can work out which fields you were filling in. For a definition, that's the term, its meaning and its description. For a tag, it's the tag's name, its description and the name of its color. If you say the name of one of your organization's tags while dictating a definition, that tag's name goes too. A tag you didn't name isn't sent. Nothing else goes with it: not your email, not your organization's identity, not your list of tags and not the rest of your glossary.

We don't keep what you said. We don't store it and we don't log it. What lands in the form is saved only if you press Submit or Create tag.

Nothing is sent while your organization has AI suggestions turned off.

When you use Senseline AI

In the web app, Ask Senseline at the foot of the left rail opens Senseline AI. It's a chat that looks things up in your organization's glossary and adds to it when you ask.

We send what you type to Anthropic. That's what you type or paste, the earlier messages of the same conversation, and the glossary entries and tag names it looks up to answer you. We add nothing that says who you are: not your email, not your name and not your organization's identity. What you type goes as you wrote it, so a name in it goes too.

We may keep the conversation. When we do, it's stored with the rest of what we hold for your account, and it's deleted when your account is. Whether we keep conversations, and for how long, is ours to decide and can change.

It works with your access and no more. A term it adds is saved as if you had used the form, with your name on it, and the term's history says it was added through Senseline AI. Text you paste isn't saved as it stands: Senseline AI lists the terms it found, and only the ones you tick and confirm are added.

We may limit how many messages you send. To apply a limit, we count your messages. The count holds nothing of what you typed. Whether there's a limit, and what it is, is ours to decide and can change.

Nothing is sent while your organization has AI suggestions turned off.

When you use Read aloud

Senseline can read a term aloud: on its page in the web app, and in the extension's hover card and the term's details in the side panel. It's off unless your organization's admins turn it on, in Organization settings, under Suggestions (“Read aloud”). The first time you press Read aloud, Senseline asks you before anything is sent.

We send the term's text to Cartesia, our voice provider, to be spoken: the term, its meaning and its description as your glossary has them, up to 600 characters. Nothing from the page you're reading and nothing you type goes with it, and nothing that identifies you or your organization. Your voice isn't recorded. Cartesia may use the text it receives to improve its models. That's why Read aloud is off until an admin chooses it.

We keep the audio. Cartesia sends back a short clip, and we store it with your organization's glossary, so the term plays again without going back to Cartesia. A clip is deleted when the term's text changes or the term is deleted. Once an organization has 2,000 clips, the oldest are dropped.

Nothing is sent while Read aloud or AI suggestions are turned off for your organization.

What we never do with any of it

• We don't sell your information to anyone.
• We don't use it for advertising, or for anything other than making Senseline work.
• We don't use it to judge creditworthiness or for lending.
• We don't use your organization's glossary or page data to train AI models. The one exception is Read aloud: if your organization turns it on, Cartesia may use the text it receives to improve its models.

Your organization's admins can see usage

Senseline is built for organizations. Your organization's admins can see which terms people come across and how often, including the sites where they were found. They can't see which pages you visited or what those pages said. If you use Senseline through your employer, assume your usage is visible to them.

What you submit. Your organization's admins see what you submit along with your name and email address. Anyone your organization lets approve terms for a tag sees it with your name, but not your email address. Other members see only their own submissions. You can withdraw a submission before it's reviewed, and that deletes it. An approved submission becomes a term in the glossary. A rejected one is kept, with the reviewer's note.

People in your organization can see what you add or change

When you add or change a term in your organization's glossary, the people in your organization can see your name on it, and when you did it. If you haven't set a name, they see “A member of your organization” there, not your email address.

Links to terms

People in your organization can copy a link to a term. An admin can also publish a list of terms at one link, with a title if they want one. Only people in your organization can open these links, and they have to be signed in.

An admin can also allow public links. This is off unless an admin turns it on. Anyone who has a public link can see the terms on it, their meanings and their descriptions, and a list's title, without signing in. The page doesn't name your organization, and it asks search engines not to list it. The person who made a link, or an admin, can turn it off, and an admin can turn off public links for the whole organization. Either way the link stops working. Your organization's admins can see the links that still work and who made each one.

An admin can also have these pages show your organization's own icons in place of Senseline's name and mark. The page still doesn't name your organization, but people who know the icons can tell whose page it is.

Keeping Senseline out of a page

Any website can keep Senseline out of a page, or part of one, by adding the data-glossary-ignore attribute. Nothing inside it is highlighted or sent anywhere. Senseline never scans or highlights its own sites (senseline.io and senseline.app). On senseline.app it only tells the page that it’s installed and whether you’re signed in, so setup can continue. Nothing is sent anywhere. Local files stay off unless you turn them on.

Where your information is kept

On Google Cloud in the United States.

Our server logs record your account's id, never your email address. Before you have an account (for example while you're asking for a sign-in code), they record only your email's domain, like company.com, and never a sign-in code.

Companies that help us run Senseline

We use a few companies to run Senseline. Each handles only what its job needs:

• Google Cloud runs our service and database in the United States. Everything we store is kept there.
• Cloudflare runs senseline.io and senseline.app, including the connection the extension uses. It passes every request on to us and sees its technical details, such as your IP address and the address requested.
• Resend sends our sign-in and invitation emails, so it receives the email address, the code and your organization's name. Resend lists the companies it uses on its sub-processors page.
• Cartesia turns a term's text into speech for Read aloud. It receives what “When you use Read aloud” above describes, and nothing that identifies you or your organization. It may use that text to improve its models, which is why Read aloud is off unless your organization's admins turn it on.
• Anthropic is used for AI suggestions, dictation and Senseline AI only. It receives what “When AI suggests a meaning”, “When you dictate” and “When you use Senseline AI” above describe, and never the page address. AI suggestions are on by default, and your organization's admins can turn them off in Organization settings. If you're not an admin, you can also email privacy@senseline.io.

How long we keep it

Sign-in codes expire after ten minutes and work once. Sign-in tokens are removed when revoked or after a long period of disuse.

Server logs are kept for 30 days and then deleted.

Your activity records (the terms you came across or hovered, and the site they were on) are kept until your account is deleted. They don't expire on a timer.

Senseline AI conversations, where we keep them, are kept until your account is deleted at the latest. We may remove them sooner.

Read aloud clips are kept until the term's text changes or the term is deleted, and an organization's oldest clips are dropped once it has 2,000.

Glossary terms and meanings your organization writes are kept until your organization deletes them, or until the account or organization is deleted. We don't expire your organization's content.

When anything is deleted, it also drops out of our backups within 30 days.

Deleting your account

To delete your account and data, email privacy@senseline.io from your account address. We delete your account and associated data within 30 days, and aim to do it within seven. The request has to come from your account's own address because that's how we know it's really yours: we won't delete an account on someone else's word.

That covers:

• Your account: your email address, sign-in tokens and sessions, any unused sign-in codes, your settings and your synced preferences.
• Your activity records, in every organization you belong to, shared organizations included: the terms you viewed and hovered, your suggested-meaning history and the sites where terms were found. These are deleted, not kept.
• Your Senseline AI conversations, where we kept them, and the count of your messages.
• Your organization memberships, and any invitations waiting for you.
• An organization where you're the only member: the whole organization, its glossary included.

In a shared organization, the glossary belongs to the organization, so it stays. Terms, meanings and shares you added stay with it, and your name comes off them. If you were that organization's only admin, its longest-standing member becomes admin, so the organization isn't left without one. Organization totals that were already counted without your name attached may stay; anything tied to you goes.

Why 30 days: we keep nightly backups for 30 days and don't edit them. Your data is removed from our live systems when we process your request, and the last backup holding it expires within 30 days.

If you are the only member of a paid organization, deleting your account cancels its subscription. Otherwise the organization and its subscription continue without you.

We keep billing records (invoices and payment history) for 7 years after your account is deleted, for tax purposes. They aren't used for anything else.

Changes and contact

If we make an important change to this policy, we'll email everyone with an account 30 days before it takes effect, and update the date at the top. For small edits, like fixing wording, we just update the date.

Questions about privacy: privacy@senseline.io. Senseline is run by RAG Consulting LLC.