Chrome Extension Privacy Policy

This policy covers the Chrome extension Call2Me, a voice assistant for the page you are on with live tab translation built in, published by Call2Me. It is separate from the platform privacy policy because the extension handles specific things: your microphone and the page you ask about (assistant), and the audio of a browser tab you choose (translator).

What the extension does

The extension has these features. Each runs only when you start it.

• Voice assistant. When you open it (the Alt+J shortcut, or “Start the assistant” in the extension's side panel, which the toolbar button opens), it listens through your microphone, reads the page you are on, and answers your questions about that page out loud. You can also type a question instead of speaking. On a Pro account it can also act on the page for you — see “What the assistant does on the page”.
• Live tab translator. It captures the audio of one browser tab that you explicitly choose, sends it to Call2Me's servers to be translated, and plays the translation back to you as speech. It runs only while you press Start and stops when you press Stop, when the tab closes, when the tab moves to a different website, or when your free minutes or balance run out. Moving to another page on the same website does not stop it.
• Phone calls — only if you use the phone call feature (included from version 0.7.0). From the side panel's Call tab you can phone a number yourself, from a phone number you bought in your Call2Me account. No AI takes part: you talk to the other person directly. A call starts only when you press the Call button in the side panel; the assistant can at most fill in the number for you, never dial it.
• AI chat — only if you pick a model and send a message. In the side panel's Assistant screen you can choose an AI model from the model picker (for example Claude, GPT, Gemini, Grok or DeepSeek), or open the full-page chat, and write to it. In the side panel you can also attach files and add the open page with the “Include page” button (only while you keep it on). Each answer is paid from balance you added with real money. Your chats are kept only in your browser. See “AI chat: what is sent”.
• Page button. A small button on the right edge of http/https pages that opens the side panel. It does not read the page. You can hide it on a site with its × or turn it off completely in the settings.
• Phone links — only in versions with the phone call feature. When you left-click a tel: link on a page, instead of Chrome's “open a phone app” prompt the side panel's Call tab opens with the number filled in. The call does not start by itself; you still press Call. You can turn this off in the settings.
• My account screen. When you open it, it reads your numbers, recent calls and spending from your Call2Me account. None of it is stored on the device.
• Number request — only if you send it. In countries where a phone number needs an address or documents, you can send a short request form: company or name, contact email and an optional note go to Call2Me only to handle that request. Our team is notified of the request (including the company name) through an internal Telegram alert.
• Interpreted phone calls — only if you turn on “Call with interpreter” in the Call tab. You speak in your language from the browser; the other person is phoned from your Call2Me number and hears an AI interpreter speak your words in their language, and you hear their reply translated. At the start of the call the other person is told that an AI interpreter is on the line. This is the only kind of phone call in which AI takes part, and only when you choose it for that call.

Your consent before the first use

Before you use the assistant for the first time, the extension opens a one-time consent screen. It explains that the assistant uses your microphone and reads the page, that it can click, type and take screenshots when you ask it to, that it asks for your confirmation before critical actions, what is sent and to which providers, and links to this policy. If this policy's scope grows (as it did with version 0.5.0, which added actions and screenshots), the consent screen is shown again once and you confirm again. Until you press Agree and start, the assistant does not turn on the microphone, read the page or start a session. If you cancel, nothing is sent.

What the assistant reads from the page

The assistant reads a page only while the assistant is open, and only the tab you are looking at in that window. If you switch to another tab in the same window while it is open, the assistant follows you and reads that tab instead; the side panel always shows which page it is looking at. Known email and messaging sites are not read without your permission: the assistant reads and acts there only after you press “Allow” in the side panel, for that tab only, until the tab moves to another site or the session ends. Known banking sites and sign-in, account and payment pages are never read and no action is taken there (only the site name is shared). Tabs in other windows are not read (except when you ask it to compare tabs, see below). Opening a page on its own never makes the extension read anything.

If the tab moves to a different page — a different site or a different path on the same site — the assistant session ends and the extension shows a message saying why, with a way to start again. Changes only to the part of the address after ? or # (for example an in-page link, or switching videos on YouTube) do not end the session; the assistant reads the page again so its answers match what you now see.

When it is open, it reads:

• the readable text of the page — the article or main content, without navigation menus, headers, footers, sidebars, scripts or embedded frames — limited to about 24,000 characters;
• the page title and address. The address is sent as the site and path only (for example https://example.com/recipes/bread) — never the query string or fragment (the part after ? or #), which can contain search terms or sign-in codes. The one exception is a YouTube video page, where the video ID (?v=…) is kept so the assistant knows which video you mean;
• on YouTube, the video's captions, when the video has them. They are fetched from youtube.com. If a video has no captions, the assistant tells you it cannot summarise it; it does not listen to the video's sound.

It never reads the values you enter in fields: text boxes, text areas, drop-downs and editable areas are skipped when it reads the page text. On a Pro account, so that it can act on the page, it reads the names and labels of fields and buttons (for example their accessible name, label or placeholder); password, payment-card and one-time-code fields are never listed.

On a page that shows a visible password field, payment-card field or one-time-code field — including one that appears while the assistant is open — the assistant sends no page text and no captions at all, only the page title and the address (site and path, with no parameters, not even a YouTube video ID).

What the assistant does on the page

On a Pro account (a single top-up of at least $10 in the last 12 months; the Call2Me server decides this, not the extension), the assistant can act on the tab it is looking at, only when you ask it to by voice or text:

• Click, scroll and type. It clicks buttons and links, scrolls, types into fields, opens new tabs and compares tabs. Page text cannot trigger these actions; only your own request does.
• Confirmation first. Before anything that spends money or cannot easily be undone — payments and purchases, sending, deleting, starting a subscription, changing account settings — and, after it has read other tabs, before typing text you did not say yourself, or before opening an address you did not give, the assistant asks you. The question appears in the extension's side panel with “Confirm” and “Cancel” buttons (if the side panel is closed, in a separate confirmation window of the extension). You confirm by saying “yes” or by pressing “Confirm”; nothing is confirmed automatically. A confirmation is good once, for about 60 seconds, for that tab and that page; switching to another tab cancels it. A “yes” that appears in a page's text is not a confirmation, and a spoken “yes” is not accepted while another tab is playing sound.
• Fields it never touches. Password fields, payment-card fields and one-time-code fields, and embedded payment frames. On a sign-in or payment page the assistant only reads: it does not click, type or take a screenshot there. It does not work on chrome:// pages or the Chrome Web Store.
• Screenshots (“look”). When you ask it to look at something, or to check that an action worked, and only while the assistant's tab is in front, it takes a screenshot of the visible area, shrinks it (JPEG, at most 1024 pixels wide) and sends it once, through OpenRouter to Google's Gemini model (with data_collection: deny), only to answer that question. The extension and Call2Me do not store the screenshot. It is not taken on a page with a visible password or card field.
• Comparing tabs. Only when you ask for it, the titles and addresses of the open tabs in the same window, and the readable text of the tabs you refer to (except sign-in and payment pages), are read and sent to compare them. Sign-in, account and payment pages, and tabs showing a visible password, card or one-time-code field, are left out: only their title and address are shared. For two minutes after tabs are read, the assistant asks before opening a new tab or following a link to another site.

Without a Pro account, the assistant only reads the page and answers; it does not click, type, scroll or take screenshots.

Full-page screenshot. Only when you ask for it (the “Screenshot of the whole page” button in the side panel), the extension scrolls the page you are looking at, captures it screen by screen and joins the pieces into one image on your device, for you to download. It is not sent to Call2Me or anyone else and is not stored by the extension. It does not work on sensitive pages (sign-in or payment pages, or pages with a visible password, card or one-time-code field). It needs no account.

Typed conversation as context. When you start the voice assistant from the side panel, the most recent typed messages you exchanged with the assistant in the side panel (at most 4,000 characters; 1,000 during the free taste) are sent once, with that session, to the assistant's language model (see “Who processes the data”), so it knows what you were talking about. They are treated as context, not as instructions. Chats with a model you picked in the model picker are not included.

AI chat: what is sent

Only when you send a message in AI chat (the model picker in the side panel's Assistant screen, or the full-page chat), the following is sent to Call2Me and, through our model router OpenRouter, to the provider of the model you picked (Anthropic, OpenAI, Google, xAI, DeepSeek, Alibaba/Qwen, Moonshot or Mistral), with the provider preference data_collection: deny:

• Your message and the earlier messages of that chat, so the model can follow the conversation.
• The open page — only while you keep the “Include page” button on. Its title, address (site and path, never the part after ? or #), readable text and the names of the visible items on it (headings, links, buttons), at most 12,000 characters of text. While the button is on (shown filled) the current page is added to each message; it is off by default and turns off when you start a new chat or reopen the panel. On sensitive sites (for example banking, email, payment, sign-in and account pages) no page text, title or path is sent.
• Files you attach (side panel only): images (PNG, JPEG, WebP, GIF), PDF documents and text files, up to 4 files per message. Our server checks them and, for a PDF, may extract its text to pass on instead of the file. Call2Me does not store or log the content of your files, your messages or the page text.

Your chats are saved only in this browser (the extension's local storage), and you can delete each one. The saved history keeps only the page title, never the page text; for images and PDFs it keeps only the file name, never the content; the content of an attached text file is part of your message and is saved with it. Call2Me keeps only the model used and the cost, for billing. AI chat needs a signed-in account and is paid only from balance you added with real money (the welcome credit does not count).

Local activity log

On a device that is signed in and linked to a Call2Me account, the extension keeps a short log of actions that otherwise never reach our servers, and sends it to Call2Me in batches together with the random device identifier: switching tabs in the side panel, changing a setting, clicking a tel: link (the site's name and the number, masked), signing out, choosing an AI model, deleting a chat (the number of messages), copying or sharing a conversation-room link, attaching files (how many), installing or updating the extension, and pressing the page button. It is used only for security and abuse prevention (for example to investigate misuse of an account). No content is written to it: no messages, page text, page addresses or files; phone numbers and email addresses are masked (for example +90555***4567). Each entry is stored with your account, the device identifier, the time, the extension version, and the IP address and browser user agent of the request, and is kept for one year (deleted after at most 400 days). If the device is not linked to an account, our server discards these entries and stores nothing.

Uninstall feedback

When you remove the extension, Chrome opens https://call2me.app/extension-uninstall/. The address carries only the extension's version and your language. Filling in the page is optional: if you choose, you can leave one or more reasons, a comment and an email address (if you would like a reply). We store what you send, the version, the language and a one-way hash of your IP address (used only to limit abuse), and use it only to improve the product.

The microphone

The assistant uses your microphone only during an assistant session you started. Chrome asks for microphone permission once, in a small extension window; you can withdraw it at any time in Chrome's settings. While a session runs, the microphone is muted whenever you switch to another window and turned back on when you return. Switching tabs in the same window does not mute it, because the assistant follows the tab you are looking at. The microphone stops when the session ends: when you close the assistant, after about 60 seconds of silence, after the session time limit, when the page changes, or when your balance runs out.

On a Pro account, if you type a command in the side panel while the assistant is off, the extension starts the voice assistant for you so it can carry out the command; this opens the microphone just as pressing “Start the assistant” would, and the usual first-use consent still applies.

Your voice is streamed to Call2Me's servers to be turned into text and answered. Call2Me does not record or save it.

If you use the phone call feature: during a phone call you started, the microphone is used only for that call and stops when the call ends. Your voice and the other person's voice are relayed in real time between your browser and the phone network. Call2Me does not record either side of the call.

If you use interpreted phone calls: both your voice and the other person's voice are streamed in real time to Call2Me's speech recognition, translation and speech services so that each side hears the other in their own language. Call2Me does not record the audio and does not keep a transcript of interpreted calls; the recognised and translated text exists only while the call runs.

What we collect

Your voice, page text and conversation — not saved by Call2Me

Your voice (assistant), the page text, title, address and captions the assistant reads, and your spoken or typed questions and the assistant's answers are used only to answer you, then discarded. Call2Me does not record or save them. Questions and answers are shown in the extension's side panel; they are not saved by the extension either: they exist only in the open side panel and, while a session runs, in the browser's temporary session memory, and are gone when the session ends and the side panel is closed. When you start the voice assistant, the latest typed messages with the assistant (at most 4,000 characters; 1,000 in the free taste) are sent to it once as context.

AI chat — saved only in your browser

Your AI chat messages, the open page (only while “Include page” is on; at most 12,000 characters, none on sensitive sites) and attached files are used only to get an answer from the model you chose and are not stored or logged by Call2Me. Your chats are saved only in this browser, without page text and without image or PDF content (only the page title and file names); you can delete each one. Call2Me keeps only the model used and the cost, for billing.

Screenshots and other tabs — not stored

Only when you ask the assistant to look, a shrunk screenshot of the visible area is sent once for a single analysis and discarded; the extension and Call2Me do not store it. Only when you ask it to compare tabs, the titles and addresses of the open tabs in the same window and the readable text of the tabs you refer to are used for that answer and then discarded.

Collected tables (CSV) — stay on your device

Only when you ask (Pro), the assistant collects the listings, products or table on the page you are on — and, if you ask for several pages, the next pages of the same site, up to 10 — into a table in the side panel. The table is built and kept in your browser; the CSV file is created on your device when you press “Download CSV”. Call2Me does not store the table. A short sample of the first rows is passed to the assistant only so it can tell you what it found, and is not saved.

Interpreted conversation rooms — not recorded

When you open a conversation room, the link you share lets the other person join from their browser. The audio of both people is sent to Call2Me and its speech and translation providers only to interpret it in real time. Rooms are not recorded, and no transcript is kept unless the room's owner has turned on “Save transcripts” for that interpreter in their Call2Me dashboard (it is off by default); then the transcript appears in the owner's call history. We store the room's two languages, its link and the minutes used, for your room list and billing.

Tab audio — not recorded by Call2Me

Processed while the translation session runs, then discarded. The translated text and speech are delivered to you in real time, and Call2Me does not keep a transcript of your translation. The translation service's technical logs record only the length of recognised and translated lines, never their content. If you keep a transcript, it is saved only in your browser and you can download or delete it.

If you use the phone call feature — audio not recorded

The call audio is relayed in real time and not recorded or saved by Call2Me. To place and bill a call we process the number you dialled, the number you called from, the start time, the duration and the cost; these appear in your Call2Me account's call history and are kept for as long as your account exists. The last 10 numbers you dialled are kept only on this device, in the extension's local storage, so you can call them again; you can remove them with “Clear” under the list.

Numbers on the page — found on your device. While the Call tab is open, the extension looks for phone numbers on the page you are viewing (links that start a call and numbers in the page text) so it can offer them as suggestions. This happens only on your computer: the page and the numbers found on it are not sent anywhere. The number you call is sent to Call2Me when you press Call. If “Call with interpreter” is on, the number is sent beforehand to show the price of the interpreted call; like every request to our API, that price request (including the number) appears in our technical request log linked to your account. When you click a tel: link on a signed-in device, the site's name and the masked number are written to the local activity log described above.

Local activity log — one year

Only on signed-in devices linked to an account: the actions listed under “Local activity log”, with the device identifier, account, time, extension version, IP address and browser user agent, for security and abuse prevention. No content; numbers and email addresses are masked. Not stored for devices without a linked account.

Uninstall feedback — only if you fill it in

The reasons you pick, your comment, an optional email address, the extension version, your language and a one-way hash of your IP address — to improve the product (the IP hash only limits abuse). Deleted on request.

Page-button settings — only on this device

The sites where you hid the page button and its position are kept only in the extension's local storage, to remember your choice.

Session metadata — up to 90 days

• Start and end time, duration, and active speaking time
• The languages involved
• A random device identifier generated in your browser
• A one-way hash of your IP address
• The cost of the session

This is kept only to bill your balance correctly, to enforce free-usage limits fairly and to detect abuse.

Email link requests — up to 30 days

When you link a device with an emailed link, we keep the request (the email address, the random device identifier, a one-way hash of your IP address and the time) so the link works once and expires. For security, the type of browser that made the request and only the first two parts of its IP address (for example 85.105.x.x) are shown on the confirmation screen; these two hints are deleted as soon as the request is confirmed or expires, and the request records are kept for at most 30 days.

Free assistant taste record

The random device identifier, a one-way hash of your IP address and how many seconds of the free taste were used — so the free taste is given once, not repeatedly. Kept while needed to prevent repeated free tastes; deleted on request.

Balance transactions — while your account exists

Amount charged, type (for example “Assistant: text”) and time, so your balance and billing history are correct. No page text, questions or answers are included.

Account data — only if you sign in

• Your email address — to identify your Call2Me account so your minutes, balance and history follow you. Kept for as long as your account exists.
• Your password — only if you sign in with a password instead of an emailed link or Google. It is sent once over HTTPS to our sign-in endpoint and is never stored by the extension. Your account password is stored only by Call2Me, hashed.
• Sign-in tokens — to keep you signed in without asking again. Kept in your browser's extension storage until you sign out.

Signing in is optional for the translator, which works without an account on a limited free allowance. The assistant offers a short free taste without an account; after that it needs a signed-in Call2Me account with a verified email address.

What we do not collect

Your name, your browsing history, the contents of pages you did not open the assistant on or add to an AI chat message, form data, cookies, or any identifier that follows you across sites. The exceptions are tab comparison — only when you ask for it, the titles and addresses of the open tabs in the same window and the readable text of the tabs you refer to (except sign-in and payment pages) are read for that one answer and not saved — and the local activity log, which records the name of a site only when you click a tel: link on it.

The device identifier is a random value created on your machine. It is not derived from your hardware or account, and you can clear it at any time by removing the extension's storage. For session metadata, the free-taste record and uninstall feedback, your IP address is never stored in raw form: it is hashed with a secret salt before it is written, so it can be used to spot abuse from a single source but cannot be read back or matched against a list of addresses. Two security records store the IP address of the request as received: the local activity log, and our technical request log of requests to our API. In addition, when you link a device with an emailed link, the type of browser that made the request and only the first two parts of its IP address (for example 85.105.x.x) are shown on the confirmation screen for security; they are deleted as soon as the request is confirmed or expires, and the records are kept for at most 30 days.

Payments

The assistant and the translator are paid from your prepaid Call2Me balance. The assistant is billed per minute of active conversation — while you are speaking or the assistant is answering; silence is not billed. Typed questions and summaries are charged per request, and the assistant asks before a costly job. The balance is shown to you as remaining time. The balance never goes below zero, and the extension never charges your card on its own; automatic top-up happens only if you switch it on yourself in your Call2Me account.

If you add funds from inside the extension, the payment form is opened in a separate window and handled by our payment provider, iyzico. Card details are entered on the provider's own form and are never seen by the extension or by Call2Me. We receive only the confirmation that a payment succeeded and the amount, so we can credit your balance.

If you use the phone call feature: phone calls are paid only from balance you added with real money (the welcome credit cannot be used for calls), per minute at the rate shown before you call, and only from a phone number you own in your Call2Me account. Each call has a maximum length, shown before you call.

Who processes the data

These features need specialist providers. The data, or text derived from it, is sent to:

• LiveKit — real-time audio transport between your browser and Call2Me
• Deepgram — speech recognition (your voice; the translated tab's audio)
• OpenRouter — routes requests to the language model: for the assistant, Google's Gemini model, which receives your question and the page text or captions and, when you ask it to look, a shrunk screenshot of the visible area (and, when you start the voice assistant, your recent typed messages with the assistant as context); for the translator, the text to translate; for AI chat, your chat messages, the open page when you add it with “Include page”, and the files you attach, to the model you picked (Anthropic, OpenAI, Google, xAI, Alibaba/Qwen, Moonshot, Mistral or DeepSeek)
• Cartesia — speech synthesis for the assistant's voice
• ElevenLabs — speech synthesis for the translator's voice
• Telephone carriers (Telnyx or Twilio, whichever provides your number) — only if you use the phone call feature; they connect your call to the phone network; they receive the dialled number, your caller number and the call audio while the call is in progress, as any phone call requires

We send them the data only so they can return the result to you. What we can configure, and do:

• The language models (assistant, translator and AI chat): every assistant request to OpenRouter (spoken and typed), every translator translation request and every AI chat message (with any page text and attached files) is sent with the provider preference data_collection: deny, so OpenRouter routes it only to model providers that do not retain or train on the data.
• Speech recognition (both features): the live speech-recognition stream to Deepgram is opened with Deepgram's model-improvement opt-out (mip_opt_out), so your voice and the translated tab's audio are not used to train Deepgram's models.
• Cartesia receives the text to be spoken (the assistant's answers) and ElevenLabs the text to be spoken for the translator (the translations); each turns it into speech. We make no further promise about how Cartesia or ElevenLabs handle that text beyond their own terms.
• Screenshots: the screenshot request is sent with the same data_collection: deny preference; the screenshot is not stored by the extension or Call2Me.

How the data is used — Limited Use

Our use of the data complies with the Chrome Web Store User Data Policy, including the Limited Use requirements.

Concretely, this means:

• Data is used only to provide the assistant, live translation, AI chat and phone call features you started, to bill your balance, to enforce the free-usage limits, for security and abuse prevention (the local activity log) and, for feedback you choose to leave after removing the extension, to improve the product.
• We do not sell or transfer the data to third parties for advertising, personalisation, credit assessment, lending, or data brokerage.
• We do not use the data for advertising of any kind.
• We do not use your voice, page text, questions, answers or transcripts to train AI models. Which providers are additionally set not to retain or train on the data is listed under “Who processes the data”.
• Nobody at Call2Me reviews your voice, page text, questions, answers or translations. Call2Me does not save the assistant's page text, questions or answers, or the translator's recognised and translated lines; our technical logs record only their length, never their content.

What is shown to you

• Before the first use, the one-time consent screen described above (shown again once when the scope grows, as in version 0.5.0).
• Every action the assistant takes on the page appears as a line in the extension's side panel, and critical actions wait for your confirmation as described above. On the page itself the extension shows only a brief outline around an element the assistant is acting on and the small page button on the right edge (which can be hidden per site or turned off).
• While the assistant is open, the side panel shows its state (listening, thinking, speaking) and the page it is looking at, and the extension's toolbar icon normally shows “MIC”. When it closes, the microphone stops. If a session ends on its own (silence, time limit, balance), the side panel says why.
• Chrome displays its own recording indicator on any tab being captured by the translator. The extension never hides or works around that indicator. The side panel states, before you start, that the tab audio will be sent to Call2Me to be translated, and shows which tab is being translated.

Protected content

Some sites (for example DRM-protected streaming services) cannot be captured by browser extensions. In those cases no audio reaches us, and the extension tells you that no sound is arriving.

Your choices

• Close the assistant with Alt+J or its stop button; the microphone page reading and any actions stop immediately.
• Stop the translator at any time using the Stop button; capture ends immediately.
• Withdraw microphone access in Chrome's settings for the extension.
• Hide the page button on a site with its ×, or turn it off in the settings; stop tel: links from opening the Call tab in the settings.
• Delete AI chats one by one in the chat list; leave out the page by not pressing “Include page”.
• Remove the extension to end all processing. Clearing its storage removes the device identifier.
• Request deletion of session metadata or the free-taste record associated with your device identifier, or of uninstall feedback, by emailing [email protected]. Activity-log entries are security records and are kept for the year described above.

Children

The extension is not directed at children under 13 and we do not knowingly collect their data.

Changes

If this policy changes materially we will update the date above and, where the change affects how audio or page content is handled, notify users in the extension before the change takes effect.

Contact

[email protected] · call2me.app