Privacy notice

Who runs this service

Claravo is a product of Twishiv Platforms Private Limited, B-203, Rosedale Heights, Vasna Bhayli Road, Bhayli, Vadodara, Gujarat 391410, India. Contact: info@claravo.ai (more on the contact page).

Claravo is for business use by adults: you must be 18 or older, and using it for work, to make an account. It is not meant for children, and we do not knowingly hold a child's personal information.

What we store

• Your account: name, email address, and your password as a one-way hash, if you set one (we cannot read it). For each way to sign in with Google or Microsoft: that service's id for your account (for Microsoft, also your organisation's id), the email address it showed, and when you added and last used it; no Google or Microsoft token.
• Your presentations, their earlier versions, files you upload, images you add, and exports you make. An image made with AI is kept with the description it was made from. A picture you paste from one deck into another is kept as a copy that belongs to the deck you pasted it into; a copy that nothing names any more (a paste you undid, once no saved version names it either) can be deleted, after at least a day, when the storage clean-up runs.
• A record of each generation: your prompt and settings (with your answers to the brief interview, if you chose it), its progress, the plan it produced (brief and outline), the web excerpts it cited, and a preview copy of each slide it wrote. It is deleted with the deck it made. Records of runs that made no deck (failed, cancelled, or an outline never built) are deleted 30 days after they last changed.
• The text read from files you upload and web pages you add (with the page's address), so it can be used in later decks. Uploaded files stay until you delete them; the create page lists your 12 most recent for that, and support@claravo.ai can delete older ones.
• Scanned pages in a PDF (pages that are pictures, with no text) are read by text recognition (OCR) on Claravo's own servers, with open-source software (Tesseract) and its English and Hindi language files, which are part of the app. No outside service sees the file or its pages. We keep the text read, which pages were read this way and how sure the reading was (and, for a page not used, why, such as its picture's size), with the file's other text; the page images are not kept apart from the file you uploaded.
• PowerPoint and Word files you export carry your account name as the file's author, in the file's properties. Anyone you send such a file to can see it there. When a piece was generated by Claravo's AI, or it shows an image made with AI, its PowerPoint, Word and PDF files also say “Made with AI assistance (Claravo)” in their properties. Edits made with Ask AI do not, by themselves, mark a piece.
• The size of each file you store (uploads, pictures, PowerPoint templates), to apply the limit on how much each account can store. Each account has room for its own files: as much as its plan gives. What counts: files and web pages you add, pictures (uploaded, stock and AI images, logos, the copies made when you copy a deck or paste a picture, and pictures from a PowerPoint file you open as a deck), PowerPoint templates, and the design kept from an imported PowerPoint file. Exports and data downloads do not count. In a team, each person's own files count to them. Something that would go over the limit is refused, and nothing of it is kept; nothing you already stored is ever deleted because of the limit. Settings → Files shows your room, what you use, and lets you delete files.
• Your brand kit, if you set one up, and any client brands you keep beside it: each one's name, colours, fonts, the smallest text size you set and the logo image (an SVG logo is kept only as the picture drawn from it). A team admin can make their kit, or one of their client brands, the team's brand: a copy of it (with the logo) is stored with the team and shown to its members, until an admin removes or replaces it.
• An invoice for each payment and a credit note for each refund: its number and date, your billing details as they were then (legal name, address, country, and in India the state and GSTIN if you gave one), the plan and period, the amounts, the tax, and the Razorpay payment or refund id. A team's invoices are shown to the admin who pays for its plan. They are kept for 8 years, also after the account is deleted (see below).
• PowerPoint templates you upload: the file as you uploaded it, and what was read from it (colours, fonts, slide size, layout names and positions, where its logo is). A template is seen only by you, and by the members of a team you share it with. It is not sent to the AI or anywhere else; a PowerPoint export of a deck that uses it is written into it.
• Your slide library: the slides you save to it, with their pictures (copies kept as yours, counted to your storage), a title, tags, who saved them, when, the deck and version they came from, and who approved that version if it was approved. Your personal library is seen only by you. Slides saved to a team's library are seen by everyone in the team, as the team's decks are, and stay with the team when the person who saved them leaves. Nothing in the library is sent to the AI or anywhere else.
• Your reference library: which of your uploaded files and your own decks or documents you marked as past work to draft from, their tags, the date each was written and the name to show it by (if you give them), and the team each is shared with. No copy is made: a draft reads the file or the deck's shown slides when it runs. A reference is seen only by you until you share it with a team, one file at a time and only by your choice, confirming each time. While it is shared, the team's members see its name, tags, dates and your name, and the team's editors and admins can draft from it into the team's decks: their AI runs then read its text and send it to the AI, as they would their own files. Past proposals can hold a previous client's confidential details, so Claravo asks you to confirm each time you share one; you are responsible for what you share (see the terms). You or any admin of that team can stop sharing it at any time, and it stops when you leave the team or become a commenter there. Claravo leaves out sentences it recognises as stating a price or a large amount, and shows the file's owner how many, with up to three examples (only them, when they draft from it themselves). It cannot catch every way of writing a price (for example, prices written in words), so check a draft before sending it. Fee tables never take amounts from past work.
• House rules: the words to avoid, spelling, tone, firm name and other rules you set for your personal decks, and those a team's admins set for the team's decks (every member can read the team's). They are checked against the deck's text in Claravo, and sent to the AI with each generation they apply to.
• Company numbers: the key figures you keep for your personal decks, and those a team's admins and editors keep for the team's decks (every member can read the team's), each with its other names, source, date, its last 10 values and who changed them. They are compared with your decks' text in Claravo only, never sent to the AI, and a deck is changed only when you choose Update.
• A PowerPoint file you open as an editable deck: the upload stays as you uploaded it; the deck holds its text, tables, charts and speaker notes, its PNG and JPEG pictures are kept as your images, and a second copy of the file is kept as that deck's own design for PowerPoint exports, with a report of what came across. Nothing is sent anywhere to do this. Improving its writing sends the deck to the AI, as Ask AI does.
• A copy made with “Adapt” (a translated copy included): a new deck of yours, with a report of what changed from the original and which of its numbers the original does not hold (for a translation, also which numbers it changed or left out), and when you chose to keep one. It is never shared or exported.
• Teams you belong to: the team's name, its members and their roles, and unused invitation links (only a one-way code made from each). Who saved each version of a deck.
• API keys you make: their name, access (read, or read and write), when they were made and last used, and a one-way code made from each key (the key itself is not kept). A key acts as your account until you revoke it. If you connect a key to an AI assistant (through MCP), what the assistant reads or exports goes to that assistant and its provider, under their terms; we answer only the calls made with your key.
• Comments on slides: the text, who wrote it and when, and who resolved a thread. Everyone who can open the deck sees them. They are not sent to the AI, not exported and not shown on share links (a comment that mentions someone is emailed to them when email is on; see below). They are deleted with the deck, or when deleted by their author (while the author can still open the deck), a team admin, or the owner of a personal deck. A member who leaves a team leaves their comments there. If a comment is edited, we keep when (it shows as “edited”), not the earlier text. A mention (@name) is stored as the mentioned person's account id and shown with their current name; only people who can open the deck can be mentioned.
• Review and approval of a team deck: who asked whom to approve it, their note, the version asked about, and each answer (approved or changes asked for), with its comment, who gave it, when and for which version. Everyone who can open the deck sees this record. It is not sent to the AI, not exported and not shown on share links or published pages. It is kept with the deck and deleted with it; someone who leaves the team stays in it, shown as “a former member” once their account is deleted.
• Who has a deck open: while you have a deck open in the editor (and its tab is visible and in use), we store that you are there and which slide you are on, so others who can open the deck see it. It is updated about every 10 seconds, deleted when you leave the deck, hide the tab or stop using it for 10 minutes, and any left behind is deleted after an hour.
• Notifications (the bell): for each new comment on a deck you can open, each mention of you, each request for you to approve a deck and each answer to your own request, we store who wrote it, which comment or request, when, whether you have read it and whether we emailed you about it. Read ones are deleted after 90 days, every one after 180 days, and they go with the comment, the deck or your account. Whether you want mentions emailed is stored with your account (Settings, Notifications).
• Share links: who made each one, when, and when it expires if an expiry was chosen. The deck's editors see this in the Share panel. When someone leaves a team or is removed from it, the share links they made for its decks are revoked.
• Views of your share links: when each link was opened and, for each slide, how long it was on screen (on a phone, where the slides are shown as text down the page, how long it took up most of the screen). Nothing that names the viewer (see “People who view a shared deck” below).
• For each signed-in session we store a short description of the browser and operating system (for example “Chrome on Windows”), never the full browser string, and when the session was last used (updated at most every 5 minutes). Sessions end after 7 days without use, and 14 days after signing in in any case.
• If you turn on two-step sign-in, we store your authenticator secret encrypted, and your 10 recovery codes only as one-way hashes. We also store the time step of the last code you used, so a code cannot be used twice, and, after a right password, how many wrong codes were entered and when (to lock the code step for a while and to email you that someone may know your password). Each signed-in session records whether it passed the second step.
• Team invitations store the invited email address; when email is set up and the admin leaves “Email the invitation” on, we send the invitation to that address.
• When email is set up and someone mentions you in a comment, we email you the deck's title, who mentioned you and the comment's text (with a link to it), unless you turned this off in Settings.
• When email is set up and a teammate asks you to approve a deck, we email you the deck's title, who asked, their note and a link to the deck.
• With approval mode on, when your address is confirmed, platform administrators receive an email that a sign-up is waiting (without your address).
• When email is set up, these account emails. A welcome, once, when your account can first be used (your address confirmed and, where sign-ups are approved by hand, approved): three steps to a first piece, with links to Create and to the templates. It is a service email, sent once whatever your email choices. A reminder, once, three days later, only if you have not made anything yet and “Tips and reminders” is on: it points to the templates. An export reminder, once, a day or two after your first deck, only if nothing has been exported yet and “Tips and reminders” is on: it names that deck and links to it. Neither reminder is sent twice. To stop them, use the link at the foot of either (it works without signing in) or turn off “Tips and reminders” in Settings, under Email choices. “Tips and reminders” starts on for new accounts, and was left off for accounts made before these emails were introduced. A yearly reminder of the Terms, the acceptable-use rules and this notice, once a year from when your account was made: a service email, sent whatever your email choices, because the law in India asks that users are told of them at least once a year. For each of these we store its kind, when it was queued and sent, and, if a send failed, only the kind of failure (the provider refused it, a timeout, a network fault, or other), never the address, the text or the provider's own message. We keep this for as long as your account exists, so none is sent twice.
• Your email choices (whether “Tips and reminders” is on) and whether you closed the first-run guide on the dashboard, so it stays closed on every device.
• Product numbers: we record a few moments that nothing else keeps (closing the first-run guide, turning tips and reminders off, the reminder being sent, a plan given to you by hand by the site owner, opening Create or the editor (at most once a day each), approving an outline, and an export that failed), each as your account id, its kind and the time. Never content. Together with the records above (when you signed up, confirmed your address, were approved, made your first deck, exported, paid), they are counted for platform administrators on a “Growth” page that shows only totals per week, never a name or an address, and any count from 1 to 4 only as “under 5”. There is no outside analytics service, browser script or cookie. They are kept for as long as your account exists and deleted with it, and they are in your data download.
• Usage counts: the number of tokens each language-model call used, each AI image requested, each copy made by filling a template's fields, and, if prices are set, their cost. Not the text itself.
• Your plan, and whether the site owner gave it to you by hand. If you buy a plan, or a team admin buys one for your team: the payment provider's subscription id, and for a team which admin bought it and how many seats. Never your card or bank details.
• Billing details, if you add them before buying a plan (or a team admin adds them for the team): the legal name and address that invoices are made out to, the country and, for an address in India, the state and a GSTIN if you give one. We need them to make tax invoices, and the country decides the currency you pay in (rupees in India, US dollars elsewhere). A GSTIN is checked only for its form; it is not looked up anywhere. Your own details are seen only by you; a team's by its admins, and only the admin who pays for its plan can change them. They are not sent to the AI, to Razorpay or anywhere else. They are kept until you change them or your account (or the team) is deleted. Each invoice keeps a copy of them as they were when it was made (see “How long we keep things”).
• Billing emails: a record of each email about your billing (or, if you are a team's paying admin, the team's): the address it goes to, its kind (payment received, payment failed, payments stopped, cancelled, yearly renewal reminder), what it says (plan, amounts, dates, the period paid for, the invoice number and links to Settings), and whether and when it was sent. Never card or bank details. Settings lists your last 20. They are recorded even while email is off, so you can see them there; one not sent within 7 days is never sent.
• For each creation counted against a team's shared plan, which team it was counted in. It stays counted in that team's month if you later leave the team.

What is sent to other services, and why

Only what each feature needs, and only when that feature is switched on by the operator and used by you.

• Addresses you subscribe to events (webhooks, for example a Zapier or Make workflow), only when you subscribe one with an API key: when a generation you started ends, or an export you made is ready, that address is sent the ids, status (with a fixed sentence for a failed or cancelled run), the deck title, an export's format, file name and size, and links to the deck and the download (the download itself still needs your key). Never the deck's content. What happens next is under that service's terms.
• A Slack or Microsoft Teams channel you connect (Settings → Slack and Teams), only if you paste its webhook address: Claravo then posts the titles of your decks, the names of the people who commented on them, mentioned you or took part in an approval, the number of figures a Living deck update changed, and links to the decks, to the channel you chose, for the kinds of update you tick. Never a comment's text or a deck's content. The address is stored encrypted, and deleted when you disconnect or delete your account. Anyone in that channel can read what is posted there.
• Anthropic (language model), when AI writing is on: your prompt and settings (and your brief interview answers; turning the interview on also sends your prompt once to word its questions), the outline if you edit it before the slides are written, and text taken from the files and web pages you attach to that generation (with their file names, and a page's address), the past work you choose from your reference library or one a teammate shared with your team (the name and date it is shown with, without the sentences Claravo recognises as stating a price or a large amount), and the house rules that apply. When you use “Ask AI”, your request, the text and speaker notes of the whole deck, and relevant passages from the deck's files and cited web pages. When “Adapt” rewrites a copy's words, the audience and purpose you chose and the text and speaker notes of the whole original deck, with each slide's layout name and the titles of the sources it cites (file names and page titles), once for each slide it rewrites (not the files or web pages themselves). When Adapt's “Translate” makes a copy in another language, the language you chose, the text and speaker notes of the whole original deck as context, and, once for each slide, that slide's words: its title, text, table cells, labels, chart titles, categories and series names, takeaway, picture descriptions and speaker notes (not chart values, pictures, files or web pages). When you ask for “Suggest a title with AI” on “Turn into a chart”, that slide's title and text (or the chosen text box's), and the chart's labels and values, read from that text or from the passage the slide cites (not other slides, speaker notes, pictures or files). When you use “Prepare for questions” with AI on, the text of every shown slide (titles, text, table cells, labels, diagram text, takeaways, chart titles, categories and values), the deck's audience, the numbers the numbers check marked with a warning, and a few passages from the deck's files and cited web pages (with their file names and the pages' addresses), plus the speaker notes only if you tick the box to send them; without AI it sends nothing. The other slide actions and “What changed” comparisons send nothing. With “Research the web” on, the model also searches the web through Anthropic, using search terms it writes from your prompt. Per Anthropic's terms, Anthropic keeps what it receives through its API for up to 30 days.
• Anthropic, for Ask your files, when you ask a question of your files on the Files page: your question (at most 300 characters) and up to 8 passages of the files you chose (or, if you chose none, of up to 50 of your newest files), each of at most 1,500 characters, with the file's name and the place in it (such as a page or a sheet). Claravo picks the passages by the words they share with your question; if none shares a word, nothing is sent. The question and the answer are not kept unless you save the answer as a brief, which is then one of your documents. “Summarise” sends, for at most 8 files you choose, 12 passages of the same size spread through them, with their file names and places, and a fixed request in place of a question.
• Anthropic, for the Follow-up pack, when you make one from a deck: the text of the deck's shown slides (cut to about 24,000 characters; its speaker notes only if you tick the box), the attendees you type (at most 500 characters) and, if you choose one, a meeting transcript, which is one of your uploads: its text in order (cut to 60,000 characters), with its file name. A transcript can hold the names of the people who spoke and what they said. The pack (what was agreed, the actions with who and by when, and a follow-up email to copy) is saved as a new document of yours, which names the deck and the transcript's file name; the transcript stays an ordinary upload.
• Anthropic, for the Stakeholder update, when you write one from a Living deck's “What changed” slide: the text of the deck's other shown slides (cut to about 16,000 characters), the changes (up to 30 rows, each with its value before and now), the note you type (at most 1,000 characters) and the audience you chose. The email is shown for you to copy; Claravo does not send it, and keeps nothing of it.
• Analyst works out a spreadsheet's data checks, movements and outliers in Claravo, from your file, and shows them without keeping them. It sends nothing to the AI or to any other service, except when you ask for the summary document: then the findings Claravo has worked out (each a short sentence naming a measure, a group and the figures) are sent to Anthropic to be put in order and worded, and the document is saved as one of yours. The file and its table are not sent.
• Compare two files works out in Claravo the figures that differ and what only one of them has. It sends nothing to the AI or to any other service, and shows the result without keeping it.
• Anthropic, for Ask your numbers, when you ask a question about a table in one of your files: your question (at most 300 characters) and a description of that table: its column names, whether each holds numbers or text, up to 15 different example values of each text column, and its first 3 rows (each cell cut to 40 characters). Claravo does the arithmetic itself. Neither the question nor the answer is kept; a chart slide made from the answer is added to a deck of yours only when you choose.
• Anthropic, for Checklist review, when you review a deck, a document or a file against a checklist: the checklist items that need reading (those Claravo can check by counting, such as the number of slides or a section's heading, are checked in Claravo and sent nowhere) and up to 40 passages of the thing reviewed, each of at most 1,500 characters. The results are shown and not kept; the checklist itself is kept in your account, or your team's.
• Pexels (stock photos), when photos are on: short search terms (at most 100 characters), typed by you or written by the model from a slide's subject, so they can name a topic or company from your deck. Your browser loads photo previews directly from Pexels, which therefore sees your IP address. Not your slides or files.
• Your browser's speech to text, only when you press “Speak” on the create page, or tick “Listen to me” when you rehearse a deck in present mode: your browser turns what you say into text, and most browsers send the audio to their maker to do it (Google for Chrome, Microsoft for Edge, Apple for Safari), under that company's terms. Claravo never receives or keeps the audio; it receives only the text, as if you had typed it.
• OpenAI (AI images), when AI images are on and you make one: the description you type for that image (at most 1,000 characters). Not your slides or files.
• Microsoft Azure AI Speech (narrated video), only when you export a deck as a narrated video: the text of its speaker notes, slide by slide, which Microsoft turns into an AI voice. Not the slides themselves. Claravo keeps the finished video with your other exports.
• Microsoft Azure AI Speech (transcription), only when you add a recording as a source: the recording, which Microsoft writes down as text with each voice told apart. Claravo keeps the recording with your other files and the text as what was read from it. A video you add as a source is not sent as it is: Claravo takes its sound out on its own server, keeps only that sound (as an .mp3, listed under the video's name) and sends only the sound to Microsoft. The picture is never stored or sent.
• Google (Drive and Slides), only when you connect your Google account in Settings: Google learns that you use Claravo and which account you signed in with. When you choose “Open in Google Slides”, the deck's PowerPoint file (its slides, text, pictures and speaker notes) goes to your own Google Drive. When you add a file from Google Drive, Google's own file picker runs in your browser (loaded from apis.google.com, with a short-lived access token for your account), and Claravo fetches only the files you pick. We keep your Google email address and an encrypted refresh token (lasting access to those files) until you disconnect, which also asks Google to withdraw it; short-lived access tokens are held in the server's memory for up to an hour. Google keeps the files under its own terms.
• Microsoft (OneDrive and SharePoint), only when you connect your Microsoft account in Settings: Microsoft learns that you use Claravo and which account you signed in with. While you browse your files on the create page, Claravo asks Microsoft for the names, sizes and dates of the files in the folder you open or that match the words you search for (those words go to Microsoft), as you, and shows them to you without keeping them. When you add a file, Claravo downloads only that file and keeps it as one of your uploads. We keep your Microsoft email address and an encrypted refresh token (lasting permission to read the files you can open) until you disconnect; short-lived access tokens are held in the server's memory for at most an hour. Microsoft does not let an app withdraw its own access, so after disconnecting you can also remove Claravo at myapps.microsoft.com (work or school accounts) or account.live.com/consent/Manage (personal accounts).
• Decks kept up to date, only when you set a deck to update itself in Update data: on the day you choose, Claravo fetches again, with your connected Google or Microsoft account and while you are not there, the files you picked that the deck's charts came from, and keeps each new version as one of your uploads. Nothing else in your Drive or OneDrive is read. Stopping it, disconnecting the account or deleting the deck ends it.
• Signing in with Google or Microsoft, only when you choose “Continue with Google” or “Continue with Microsoft” or add it in Settings → Security: that service learns that you sign in to Claravo, and with which of its accounts. Claravo asks it only for your account id, email address and name. It stores the account id (for Microsoft, also your organisation's id), the address shown, and when you added and last used that way to sign in; the name is used only as your account's name when you sign up with it. Claravo receives no lasting access and keeps no Google or Microsoft token. You are matched by that account id, never by email address. When one is added, we email your account's address. These ways to sign in are deleted with your account, and removed by a password reset, by a password change unless you untick the box, or when you remove one in Settings.
• Resend (email), when email is on: your email address and the message we send you (a password-reset or address-confirmation link, a team invitation, a mention of you in a comment (the deck's title, who wrote it and the comment's text), a request to approve a deck (its title, who asked and their note), a notice that a Google or Microsoft sign-in was added to your account, a notice that a file you watch changed (its name, the figures that changed with their values before and after, and where in the file) or that watching it stopped after failed checks, a notice that two-step sign-in was turned on for your account, a notice that your account was approved, the welcome, the reminders and the yearly reminder described above, or a warning that wrong two-step sign-in codes were entered after your password). If you pay for a plan (or for a team's plan, as its paying admin): emails about it, which hold the plan, amounts, dates and your invoice (attached as a PDF when it is ready), never card or bank details.
• Razorpay (payments), when payments are on and you buy a plan, or a team admin buys the Team plan: the plan, the number of seats for a team, and a note naming your Claravo account id (or the team's id and the buying admin's account id) and the currency (rupees or US dollars, from the country in your billing details). We do not send Razorpay your name, email address or billing address. On Razorpay's own payment page you enter your contact details and your card, UPI or bank details; they go to Razorpay directly and never reach us. Afterwards Claravo asks Razorpay for the subscription's state (active, cancelled, its period and seats), and asks it to cancel the subscription or change its seats when you or your team admin do so in Claravo. To make an invoice or a credit note, Claravo reads back from Razorpay each payment's and refund's amount, currency, state and the period it paid for; nothing is sent to Razorpay for that.
• A website you name, when you use “Start from your website” in the brand settings: our server fetches that page, up to four of its stylesheets and up to three logo images, the way a browser would but without running its scripts. The site sees a request from our server, not from you: no cookies, nothing about your account. What was read is shown to you and kept only if you save the brand (the logo as your brand's logo). The log notes your account and the site's name, not the full address.
• A web page you add as source material (“Add a web page” on the create page): our server fetches that one page once, without running its scripts, and keeps the page and the text read from it with your uploads, with its address, until you remove it. The site sees a request from our server, not from you: no cookies, nothing about your account. The log notes your account and the site's name, not the full address.
• Video and calendar services on published pages (YouTube, Vimeo, Calendly, Cal.com): a publisher can put a video or a booking calendar on their page. These services are chosen by the publisher, and they are reached by the viewer's own browser, only after the viewer presses Play or Open on that page. Claravo sends them nothing: the page only builds the player's address from the video or booking page the publisher gave. That is why they are not in the table of outside services below, which lists only services Claravo sends personal information to.
• Microsoft Azure (hosting): the app, its database, its file storage, its backups and its logs run on Azure, in its Central India region. The other services above, except Razorpay, are run by companies based outside India and may process data in other countries.

Outside services (sub-processors)

Every outside service that can receive personal information, what it gets, and where. A service marked “off” is not set up, and receives nothing.

Changes to this list

We give at least 30 days' notice, by email and here, before adding or replacing a service (as the data processing agreement says).

• 29 October 2026: This record starts. The table above is the full list in use on this date; every later addition or replacement is listed here at least 30 days before it takes effect.

Training

Claravo does not use your content to train any model, on any plan: it trains no models at all. Anthropic states that, under its commercial terms, it does not train its models on data sent through its API by default; see Anthropic's current terms for the details. OpenAI, which receives image descriptions when AI images are on, states the same for its API; see OpenAI's current terms.

Sharing

If you turn on a share link's “Numbers checked” badge, anyone with that link can open its numbers report: every number on the slides, the names of the sources it was looked for in (file names and web page addresses), and the passage each number was found in (at most 300 characters from your file or page). The report is a copy taken when you turn the badge on or press “Check again”; it is kept on the link, and viewers never cause a new check. The badge is off unless you turn it on, link by link.

Your presentations are private to your account unless you move one into a team. Every member of that team can then open it, and, by their role, edit, share, export or archive it; its version history shows who saved each version. Members of a team see each other's names and email addresses. A member who leaves or is removed loses access to the team's decks, including the ones they made, which stay with the team. If you create a share link, anyone with that link can view that presentation until the link is revoked.

Published pages

If you publish a deck or document as a web page, anyone with its address can read it without signing in, and search engines may list it only if you allow that (never while it asks for a password or an email address). We keep a copy of it as it was when you published or last updated it; speaker notes are never included. A page can ask visitors for a password (we store only a one-way code made from it) or for an email address. The addresses visitors type are kept with the page and shown to you (and, for a team deck, the team's editors and admins); we never send anything to them. A visitor who gets past the gate gets a cookie for that page for 12 hours. The page can show the “Numbers checked” badge and report, as a share link can. Views are counted as for share links, below. The email gate names the page's publisher (the account that made the deck) and says why it asks, and links to this notice.

A page can also show a video (from YouTube or Vimeo), a booking calendar (from Calendly or Cal.com) and up to three buttons that lead to the publisher's own links (a web address or an email address). Nothing loads from YouTube, Vimeo, Calendly or Cal.com until the viewer presses Play or Open; the page says so next to each one. Then that service sees the viewer's IP address and may set its own cookies, under its own privacy policy (YouTube is loaded from its privacy-enhanced address, and Vimeo is asked not to track). When the viewer presses Play on a YouTube video, YouTube is also told the site's address (for example the published page's domain), never the page's own address, because YouTube does not play a video that is told nothing; Vimeo, Calendly and Cal.com are told neither. What a viewer types into a booking calendar goes to that calendar's service, not to Claravo. A video's picture, if the publisher adds one, is served by Claravo. Button presses are counted without names: for each press we keep only the page, the button, its text and the time, never the viewer's name, address or any number from their browser, and the publisher sees the count per button. A press goes through Claravo's own address first, which sends the browser on to the publisher's link without telling that site which page it came from.

If you add your own domain for a page, we keep the domain name with your account and read its public DNS records when you press Check. To give it a certificate, the domain name is sent to Microsoft Azure, which hosts the service and issues the certificate. Removing the domain deletes it here and asks Azure to let it go.

Asking a published deck

A publisher can let the readers of a published page ask it questions (“Ask this deck”). The question a reader types is sent, with the text of the page, to Anthropic, whose AI writes a short answer from the page alone (on training, the same applies as to anything else sent to Anthropic, above). We keep the question, the answer shown and the time, with no name, email address, IP address or cookie, and the publisher (the account that made the deck) and, for a team deck, its editors see them. Questions are counted per network address for an hour to stop abuse; the address is not kept with the question. It is not offered inside another site's frame, and only on plans that include it.

The assistant

“Ask Claravo” is a help panel on the pages you see when signed in. When you send it a message, Claravo sends Anthropic what you typed (or said, as text), the conversation so far in the panel, which page you are on, the titles of your recent decks, and the content of the deck open in the editor or one you ask about, so its AI can write the reply (on training, the same applies as to anything else sent to Anthropic, above). Claravo does not keep the conversation: it stays in the panel until you reload or close the page. We keep a count of your messages, with their time and the tokens used, to apply the limits and see costs. When you tick “Read replies aloud”, the text of each reply (never what you typed) goes to Microsoft Azure AI Speech, which reads it in an AI voice; Claravo does not keep the audio. The assistant changes nothing by itself: anything that would change your work waits until you press “Do it”.

Recorded presentations

A presenter can record their voice while moving through a deck (“Record and send”). We keep the sound, when each slide came on screen and a copy of the slides as they were, without speaker notes or hidden slides. Anyone with the recording's link can play it without signing in. We count how many times it was played, at most a few times an hour per network address, and keep nothing about who played it (no name, email address, IP address or cookie). The presenter can make a new link, which stops the old one, or delete the recording at any time.

Live audiences

A presenter can go live from present mode so the people in the room can join on their phones with a code, with no account, to vote in polls, ask the presenter questions and react. We keep each vote (one per browser, with a random number kept in a cookie for a day so a browser votes once), each question as typed, and the count of each reaction, with no name, email address or IP address. Only the presenter sees the questions; the audience sees a poll's totals after voting. Questions and votes are counted per browser and per network address for an hour to stop abuse; the address is not kept with them. Nothing is sent to any other service.

Verifying an exported file

When you export a PowerPoint, PDF or Word file, we keep a fingerprint of it: a hash of the whole file and of the text on each slide, page or section (not the text itself), the deck's title, your name as you set it, and the date. A PowerPoint or Word file also carries a signed code in its document properties. Anyone who has the file can bring it to the Verify page, with no account, and see whether it is the file you exported or which slides were changed, with that title, your name and the date. If you tick “Also show them the numbers check”, they also see its one-line summary as it stood at export. A file brought to the Verify page is read and then dropped: it is never stored, and nothing about it is sent to any other service.

Checking an email in Outlook

When you choose Check numbers in the Claravo add-in for Outlook, the text you wrote in that email (not the earlier messages quoted below it) and its attached files are sent to Claravo, read to compare the numbers, and then dropped: they are never stored, no AI reads them, and nothing about them is sent to any other service. We count that a check ran, with no content. While Check history is on, we also keep a short summary of the check (see “Check history”): when it ran, that it came from Outlook, and how many numbers were traced, needed a look, were not found or differed from your company numbers. It holds no title and nothing from the email or its files. The add-in signs in with a read-only key named “Outlook add-in”, which you can revoke in Settings.

Checking text in Chrome

The Claravo extension for Chrome reads a page only when you click its button or its right-click item, and then only the text you selected. That text, and any files you choose in the extension, are sent to Claravo, compared with your company numbers and those files, and then dropped: they are never stored, no AI reads them, and nothing about them is sent to any other service. We count that a check ran, with no content. While Check history is on, we also keep a short summary of the check (see “Check history”): when it ran, that it came from Chrome, and how many numbers were traced, needed a look, were not found or differed from your company numbers. It holds no title and nothing from the page or the files. The extension keeps its key in Chrome on your computer; it is a read-only key named “Chrome extension”, which you can revoke in Settings.

Add-ins for Excel, PowerPoint and Google Slides

Excel. When you choose Make the deck in the Claravo add-in for Excel, the cells you selected (or the sheet's used range, when one cell is selected), as Excel shows them, are sent to Claravo as a CSV file named after the sheet, with the prompt you typed and the number of slides. Claravo keeps the file with your other uploads, until you delete it, and counts it to your storage. It is the source of the deck it then makes, so its text is sent to the AI as any file you attach is (see Anthropic above), and the deck counts as a creation. The add-in reads nothing else in the workbook. It signs in with a key named “Excel add-in”, which can read and write, because it uploads the file and makes the deck; you can revoke it in Settings. The key is kept in the add-in pane's own storage and sent only to Claravo.

PowerPoint. In the Claravo add-in for PowerPoint, a new deck sends Claravo the prompt you typed, the number of slides and whether to use your brand; Claravo then makes the deck as it does in the app (see Anthropic above). Adding a deck to the open presentation asks Claravo to export it as a PowerPoint file (kept with your exports, as above), downloads that file and hands it to PowerPoint. The add-in sends Claravo nothing from the presentation you have open. It signs in with a key named “PowerPoint add-in”, which can read and write; connecting again replaces it, and you can revoke it in Settings.

Google Slides. The Claravo add-on works as the PowerPoint add-in does: it sends Claravo the prompt, the number of slides and whether to use your brand, or asks for an export of a deck of yours. To add that deck, the add-on puts its PowerPoint file in your own Google Drive as a temporary Google Slides file, copies its slides into the open presentation, and then deletes the temporary file (if deleting it fails, the file stays in your Drive, named as temporary, until you remove it). The add-on runs inside Google, under your Google account and Google's terms, and can reach only files it makes in Drive; Claravo sends Google nothing for it, and the add-on sends Claravo nothing from the presentation you have open. It signs in with a key named “Google Slides add-on”, which can read and write; connecting again replaces it, and you can revoke it in Settings. The key is kept in the add-on's storage for your Google account and sent only to Claravo.

Check history

Each time you check a deck, a document or a text (in Claravo, Outlook, Chrome, the API or an AI tool connected to Claravo), we keep a short summary of it: when, what kind of thing it was, where you checked it from, the title of the deck or document when you checked one in Claravo, and how many numbers were traced, needed a look, were not found, or differed from your company numbers. The text and the files you checked, and the numbers themselves, are not kept in it. The summary is used for your check history and your plan's monthly checks. A team's admins see how many checks each member ran this month and those counts, never what was checked. You can delete one summary or all of them in Check history: its title goes at once, and the rest is kept until the month ends only to count your monthly checks, then deleted. Summaries are deleted with your account, and included when you download your data.

Reports from Stripe and HubSpot

You can connect your own Stripe or HubSpot account to Claravo with a read-only key you make in that service, so a report can read its numbers there. We keep the key encrypted, the last four characters of it so you can tell which key it is, and the name the service gives the account. We use the key only to read the tables you ask for (such as revenue by month, or deals won and lost), when you add one and on the schedule you set for a report; Claravo only ever reads with it, and the steps ask you to give the key read access only. What we read is turned into a table in your files, kept and deleted like any file you upload; the people, customers and deals behind the numbers are not kept, only the totals by month or stage. Stripe and HubSpot see that your key was used, as they would for any app you give a key to. Disconnecting on the Reports page deletes the key at once; the key is also deleted with your account, and is never included when you download your data.

Watched files

When you watch a file you picked from Google Drive or OneDrive (or a table from a Stripe or HubSpot report), Claravo reads it again on the schedule you choose (every hour, every day or every week), as you and while you are not there, using the connection you made: the encrypted refresh token we keep for your Google or Microsoft account, or the key you gave for the report. Only that file is read, never anything else in your Drive or OneDrive. The new reading is compared with the last in Claravo; nothing is sent to the AI or to any service other than the one the file comes from. Each check stores a record of when it ran and whether the file had changed, had not changed or could not be read (with the reason); for a change it also holds the counts, the places that changed and the figures that changed, before and after, but not the text of the passages. These records are deleted after 180 days. When the file has changed, the new version is kept as one of your uploads, counted to your storage, and the version before stays beside it; when it has not, nothing new is stored. Unless you turn it off for that file, we email you when it has changed, with the file's name, the figures that changed (up to 20) and where, and once when three failed checks in a row pause the watch. You can watch up to 20 files. On the Watched files page you can change the schedule, turn the email off, pause, resume or stop watching: stopping deletes the watch and its records at once. A watch is also deleted with its file or your account.

Courses

When you make a course from a deck, the text of its shown slides (not its speaker notes or hidden slides), with the titles of its sections, is sent to Anthropic to write the questions, section by section (at most 6,000 characters of each), as with other AI features, and Claravo keeps a copy of the deck as it was then, with the questions and their answers. Someone who takes your course gives their name and email address; Claravo keeps them with their answers and score, for you to see, until you delete the course, the deck or your account. They are not used for anything else.

Data rooms

When you make a data room, Claravo keeps its name, introduction and terms, a copy of each deck you put in it as it was then, and which of your uploaded files it holds. Someone who enters your room gives an email address and, when the room has terms, agrees to them. The address is whatever they type: it is checked only for its form, nothing is sent to it, and Claravo does not know whether it is theirs. We keep the address, whether they agreed and when, and which items they opened, when, and how long a deck was on screen, and show these to you. Their browser gets a cookie that lets it back in for 24 hours (see “Cookies”). A visitor not seen for 365 days is deleted with that record, by a clean-up that runs every 6 hours; deleting the room, or taking a visitor out, deletes it at once. If you entered someone's data room, they decide what happens to your address: write to them, or to us.

People who view a shared deck or a published page

When someone opens a share link, the deck's owner (and, for a team deck, the team's editors and admins) can see that it was opened, how many different browsers opened it, and how long each slide was on screen (time counts only while the page is visible; on a phone, where the slides are shown as text down the page, a slide's time is the time it takes up most of the screen). To tell a return visit from a new viewer, the viewer's browser keeps a random number in its own storage; we store only a one-way code made from it and the link, which cannot be linked across decks. We do not record the viewer's name, email, IP address or any cookie. A browser that sends Global Privacy Control or Do Not Track, or cannot keep the number, is counted as one view and nothing more. Visits by the owner, or by a member of the deck's team, while signed in to Claravo's own address are not counted (on a customer's own domain, or inside an embed on another site, the browser does not send that sign-in, so those visits are). The shared page tells viewers this, and links to this notice. View records are kept while the deck exists (revoking a link stops new views but keeps the counts), and are deleted with the deck and after 13 months. A viewer can clear the random number by clearing the site's data in their browser.

Reports and administration

If you report a page, we keep the page address, the reason and details you give, and your email address if you choose to give it, so we can look into the report.

Actions our administrators take on accounts and content (such as approving, suspending, removing a page or deleting an account) are recorded with who did it, when and why. When an account is deleted, that record, including the account's email address, is kept as proof of what was done.

Cookies

One cookie, which keeps you signed in (up to 14 days). It is needed for the service to work. While you connect a Google or Microsoft account, or sign in with Google or Microsoft, a second cookie holds that sign-in's check for up to 10 minutes. A visitor who passes a published page's password or email gate gets one more, for that page only, for 12 hours. Someone who votes or asks in a live presentation gets one holding a random number, for a day, so a browser votes once. A visitor who enters a data room gets one for that room only (named “cl_room_” followed by the room's id; it is kept out of the page's scripts and holds a signed pass, not the address), for 24 hours. There are no analytics, advertising or tracking cookies from Claravo. A video or calendar the viewer chooses to open on a published page may set its own. (Share-link views use the random number described above, kept in the viewer's browser storage, not a cookie.) When you copy items in the editor, your browser also keeps the last copy in its own storage, so it can be pasted in another tab; it stays on your device, is read back only while you are the one signed in, is replaced by your next copy, and is removed when you sign in or out, or clear the site's data.

Logs

The server log is for finding faults. It records errors, the account and generation involved, and counts such as tokens used; the language-model log lines hold counts and field names, not text. It never records passwords or the links in emails. It is not meant to hold your content, but an error report can occasionally include a fragment of the data being handled when the error happened. It is kept in Azure Log Analytics, in the Central India region, for 180 days, as India's CERT-In directions require.

When a page fails in your browser, Claravo's own script sends us a report, which is written to the same log as an error. It holds the page's path (any part that looks like an id or a token is replaced with “:id”), the error's name, the browser's error message (cut to 300 characters, with email addresses and token-like strings replaced), up to 8 lines of the error's trail with our own address and any query string removed, the account id if you are signed in, and the reference of an error page if it showed one. It holds no page content, and not your IP address or your browser's description. Reports that come only from a browser extension are dropped. These reports are kept as long as the other logs: 180 days.

Monitoring is built into Claravo. One outside service, UptimeRobot, checks every so often that our health page answers; that page holds no personal information, and UptimeRobot receives none. Our domain's DNS is run by Squarespace, which sees only lookups of our address, as any DNS provider does. We record security events (sign-in results, two-step sign-in, password changes and resets, session sign-outs, sign-up approvals and team security settings, as well as API key and plan changes, administrator actions and account deletion) with account ids and outcomes only, never passwords, codes or content. A security event also holds the network you connected from (the first three parts of an IPv4 address, or the first half of an IPv6 address), never the full address. These, and warnings and errors, are kept for 180 days in our database in India (Central India region). Counts of requests, errors and jobs are kept per minute for 90 days and per hour for 13 months; they name no one.

To limit abuse (for example, too many sign-in attempts), the server counts requests by the visitor's IP address. These counts are kept in our database, so every copy of the server counts the same way: each count holds the address and the number of requests in a window of at most an hour, and is deleted by the hourly housekeeping once its window has ended. They are not written to the logs.

How long we keep things

• Presentations and files stay until you delete them. “Delete forever” removes a deck, its versions, its exports, its photos (unless another of your decks shows them), its share links and the generation records that made it; archived decks stay until then. A team admin can delete a team deck forever. Uploaded files are deleted separately.
• A deleted file is removed from storage at once. Database backups may hold deleted data for up to 35 days, and daily snapshots of the file storage for up to 30 days.
• Exported files: deleted 30 days after they are made (the deck stays; export it again to get a new file).
• Records of generation runs that made no deck: 30 days after they last changed.
• A data room, its terms and its copies of decks: until the owner deletes the room, or the deck or file it holds, or their account. A room's visitor, with what they opened: 365 days after they were last seen, or at once when the owner deletes the room or takes them out.
• A course, its copy of the deck and the learners' names, email addresses, answers and scores: until the owner deletes the course or the deck, or their account.
• A watch of a file: until the owner stops it or deletes the file or their account. The records of its checks: 180 days after each check.
• Password-reset links expire after one hour; address-confirmation links after two days. A link is deleted a day after it is used or expires, and a sign-in session once it expires.
• A password reset signs you out everywhere, revokes your API keys and disconnects any Google or Microsoft account. It also removes every way to sign in with Google or Microsoft.
• Changing your password in Settings signs out every other session. It revokes your API keys only if you tick the box (a password reset always does). Unless you untick its own box, it also removes every way to sign in with Google or Microsoft.
• Share-link view records: deleted when their deck is deleted forever, and after 13 months.
• A recording, its slides and its play count: until the presenter deletes it, or with the deck or the presenter's account.
• A question asked of a published page and its answer: until the page is unpublished or the deck deleted, and at most the newest 500 per page.
• Who has a deck open: when you leave it, and at most an hour after your last report.
• The fingerprint of a file you exported: for as long as your account exists, even after the deck is deleted (the file may still be in use), and with your account.
• A live presentation's polls, votes and questions: 30 days after it ends, and with the deck.
• The review and approval record of a team deck: for as long as the deck exists, and with it.
• Notifications: read ones after 90 days, every one after 180 days, and with their comment, deck or your account.
• Billing email records: 13 months after they were made, and with your account.
• Records of the welcome, reminder and yearly reminder emails, your email choices and product-number moments: for as long as your account exists, and with it.
• A share link's numbers report (when you turn on its “Numbers checked” badge): until you turn the badge off, revoke the link or delete the deck.
• A published page's copy of the deck, its numbers report and the email addresses its gate collected: until you unpublish it or delete the deck (turning the email gate off also deletes its addresses). Its view records are kept like a share link's. Its buttons, videos and calendars go with the page.
• Counts of button presses on a published page: kept like its view records (deleted when the deck is deleted forever, and after 13 months).
• PowerPoint templates: until you delete them; the file is removed from storage at once. A deck's own design (the copy kept when a PowerPoint file is opened as a deck): until that deck, and every copy you made of it, is deleted forever (it goes with the last of them, or within a few hours after that by the hourly housekeeping).
• Invoices and credit notes, with the billing details printed on them: 8 years from the year they were issued, also after you delete your account, because Indian GST law requires the supplier to keep them (Central Goods and Services Tax Act, section 36: until 72 months after the due date of the annual return for that year). They cannot be deleted on request before then. When the account is deleted they are no longer linked to it, and the billing profile itself is deleted.
• API keys: until you revoke them (a password reset revokes them all); a revoked key's name and dates are deleted 30 days later.
• Webhook subscriptions: the event and the address, until you end them or revoke the key that made them (a password reset ends them all). Each delivery (what was sent, when, the answer's status and its first 200 characters) is kept 30 days.
• “Download my data” (Settings → Your data): the zip is kept for 7 days after it is made, then deleted, and at once if you delete your account. You can ask for one a day.
• A change of email address: the new address is kept, waiting, until you open the link we send to it (it works for 24 hours) or cancel the change. We then tell your old address that it changed.
• You can delete your account yourself in Settings → Delete account: it shows what will happen first, then asks for your password (or, with no password, a fresh sign-in with Google or Microsoft), a code if two-step sign-in is on, and your email address. You can also write to support@claravo.ai; an account that has not confirmed its email address, is waiting for approval or was not approved cannot open Settings, so it writes in. We then delete your account and everything that is only yours: your personal decks (with their versions, exports, share links, published pages and the addresses their gates collected), uploaded files, images, PowerPoint templates, brand kit, generation and usage records, API keys, webhook subscriptions, a connected Slack or Teams channel, your ways to sign in with Google or Microsoft, custom domains, your personal slide library (with its pictures), your reference library marks (shared ones stop being shared), house rules and company numbers, and your billing profile. Invoices and credit notes are kept, unlinked from the account, for the 8 years GST law requires (above). A team you are the only member of is deleted with its decks. We cancel your own subscription, and the Team plan of any team deleted with your account; Razorpay keeps its own record of the payments. A Team plan you pay for, for a team that stays, is not cancelled automatically: we either cancel it in Razorpay or ask the team to buy it again under another admin. We ask Google to withdraw a connected Google account; Microsoft offers apps no way to, so remove Claravo there yourself (see above). Team decks you made stay with the team: they pass to another member (an admin where there is one), with the images they show, and if you were the team's only admin another member becomes one. Slides you saved to a team's library stay with the team, their pictures passing to another member (an admin where there is one); the same happens when you leave a team. Your comments on decks that stay are kept, shown as by “a former member”, and version history shows versions you saved the same way. While you pay for a Team plan of a team that stays, Settings does not delete your account: cancel the Team plan first, or ask another admin to buy it. Invoices are kept for 8 years with the billing details on them, without a link to your account, because GST law requires it (CGST Act, section 36).

Your rights

What you can do today: download a copy of all your data (Settings → Your data), read and download your decks and documents (every export), delete them, delete your uploaded files, change your name and email address, disconnect a Google or Microsoft account, revoke API keys, and delete your account (Settings → Delete account). You can also ask support@claravo.ai to correct your details or delete your account (see above).

We answer a request within one month (sooner where the law sets a shorter time), free of charge, and may ask you to confirm who you are first. Write to support@claravo.ai or the Grievance Officer (below).

Your content and the people in it

For your account, your billing and keeping Claravo secure, we decide how personal information is used (we are the “controller”, or “data fiduciary” in India). For the content you upload and make, and for the viewers of your share links and published pages, we handle personal information on your behalf and only to run the service for you (we are your “processor”). If someone asks us about their information in your content, we pass the request to you and help you answer it. The data processing agreement sets out our promises as your processor.

Why we use your information

• To run your account and the service you asked for, and to take payment: because it is needed for our contract with you.
• To keep Claravo secure, stop abuse, count use against plan limits and fix faults: our legitimate interest in a safe, working service.
• To keep invoices and answer lawful requests: because the law requires it.
• To connect a Google or Microsoft account, or anything else you switch on yourself: because you chose it, and you can switch it off at any time.

Where it goes

Your information is stored in India and is sent to the outside services listed above, several of them in the United States. Where the law of the European Union or the United Kingdom applies, such transfers need recognised safeguards, such as the European Commission's standard contractual clauses. The safeguard used with each service is in the table of outside services above. UptimeRobot, which checks that the site is up, is based in the European Union and receives no personal information.

Rights by place

• India: you can review and correct your information and withdraw consent you gave (Information Technology Act, 2000 and its 2011 rules). As the Digital Personal Data Protection Act, 2023 comes into force, you can also ask for a summary of your information and who it was shared with, its correction or erasure, and name someone to act for you; if we do not resolve a complaint, you can then go to the Data Protection Board of India.
• European Union and United Kingdom: you can ask for access to your information, its correction or erasure, to restrict or object to how it is used, and for a copy in a portable form (Settings → Your data); you can withdraw consent at any time. You can complain to your data protection authority (in the United Kingdom, the Information Commissioner's Office).
• United States: we do not sell or share personal information for advertising, and we honour Global Privacy Control. Where a state privacy law gives you rights (to know, correct or delete), you can use them here, and we will not treat you differently for doing so.

Complaints and the Grievance Officer

Grievance Officer: Nirav Soni, grievance@claravo.ai, Twishiv Platforms Private Limited, B-203, Rosedale Heights, Vasna Bhayli Road, Bhayli, Vadodara, Gujarat 391410, India.

Under the Information Technology Act, 2000 and the Information Technology (Reasonable Security Practices and Procedures and Sensitive Personal Data or Information) Rules, 2011, you can write to the Grievance Officer about how your personal information is handled, or about anything on Claravo you think is unlawful. Say who you are, how to reach you, what the complaint is about and, for a page or link, its address. We acknowledge a complaint within 24 hours of receiving it and resolve it within 7 days; a request to remove content that the law requires us to remove faster is handled within the time the law sets. The Grievance Officer also handles complaints about personal information from anywhere in the world.