Privacy Policy

• Your conversations, memories, skills and settings stay in your browser.
• With Browser Agent Cloud, requests pass through our server to Anthropic, the model provider. We keep usage records, not your messages, page content or the model’s replies.
• With your own API key, requests go straight from your browser to your provider.
• Signing in with Google or GitHub is optional, except before you buy a plan. If you sign in, we store your account ID with that provider and your verified email.
• No ads, no third-party analytics, and we don’t sell your data.

Stored only in your browser (chrome.storage.local), and deleted when you remove the extension:

• Your last 30 conversations, including page text that was part of them, research sources and files the agent created
• Memories: short facts you asked it to remember
• Skills and settings
• Your API keys and endpoint addresses, if you use your own key
• A random user ID and a device token that identify this installation to Browser Agent Cloud

You can delete conversations and memories in the extension at any time.

In either mode, a request can contain:

• Your messages and the conversation so far
• Content of the current tab that the agent reads for your request: its text, the list of buttons and fields, the page title and address
• Text you selected, if you asked about a selection
• A PDF you asked it to read. A scanned PDF with no text layer is sent as the whole file to Anthropic (Claude), and only after you click Allow on a confirmation card
• When researching: search results and the summaries or text of pages it read
• Your saved memories (when memory is on) and the names and descriptions of your skills

Page content is read and sent only when you ask the agent to do something, except for home suggestions, described below.

When the extension is installed, before you open it, it registers your installation with our server. It sends your browser language and the extension version, and our server sees your IP address. We store a random user ID, a hash of the device token, your language and the region code in it (for example TW from zh-TW), and a SHA-256 hash of your network address: the whole address for IPv4, or only its first half (the /64 network) for IPv6. We use that hash only to limit how many new installations one network can register and how many of them get free credits. Our server also sets a cookie on its own domain, __Host-ba_bid (see “This website”), so it can recognize the same browser if you reinstall the extension. The extension then opens our welcome page with that user ID. This happens in both modes, and it doesn’t include any page content.

While you use it, the extension asks our server for your plan and remaining credits, sending your device token. Each request to the AI model goes to our server with your device token, your reading-length setting and counters for the task (pages read, actions taken, characters read and searches; no addresses or content). Our server forwards the request to Anthropic through an API gateway operated by [[API gateway operator]] and streams the reply back to you.

What our servers keep:

• Your installation record: user ID, the hashes above (device token, browser cookie and network address), language, region, plan, whether it gets free credits, and the referral link it came from, if any. If you sign in to an account that already exists, we also note which account the old record was merged into
• If you sign in: the provider (Google or GitHub), your account ID there and your verified email, if any
• Usage records for each task and model call: time, model, thinking depth, reading length, token counts, response time, cost, credits, the counters above, whether the call succeeded, why it stopped, and the error type if it failed

What our servers don’t keep: your messages, page content, research summaries or the model’s replies. They pass through only to reach the model provider and bring the reply back. The API gateway: [[What the API gateway logs and for how long]].

The free plan works without signing in. Before you buy a plan, you sign in from Settings in the side panel with Google or GitHub. Your plan and credits then belong to that account, and you can use them on another computer by signing in there.

We ask only for what’s needed to identify you: the openid and email scopes from Google, and read:user and user:email from GitHub.

We store the provider, your account ID there (Google’s “sub”, or your numeric GitHub user ID) and your email address if the provider marks it as verified; otherwise we store no email. We use these to recognize you when you sign in again and to show in Settings which account you’re signed in with.

While you sign in, our server sets a cookie named __Host-ba_auth so that only the browser that started a sign-in can finish it. It lasts at most 10 minutes and is deleted as soon as it’s used.

If you sign out in Settings, this installation is disconnected from the account, and the extension registers a new installation that isn’t signed in and doesn’t get free credits.

Requests go directly from your browser to the provider you choose (Anthropic, OpenAI, Google Gemini or OpenRouter) or to an endpoint you configure, together with your API key. Your tasks aren’t sent to our servers and no credits are counted. The only contact with our server is the registration made when the extension was installed (described under Browser Agent Cloud). The provider’s own privacy policy and terms apply.

When answering needs other pages, the agent uses your own browser:

• Searches open as Google searches in background tabs, or Bing if Google asks to verify you’re human. The search engine receives the keywords, which the AI model writes from your question, plus what your browser normally sends, such as your IP address and cookies. If you’re signed in, the searches may be saved in that account’s activity.
• Pages it reads open in background tabs with your cookies and signed-in sessions, as if you opened them, and close after reading. Before opening a page, it may send a short HEAD request (with your cookies) to check whether the address is a PDF. PDFs are downloaded directly, also with your cookies.
• Each page’s text, up to your reading-length setting and at most 80,000 characters, is condensed by a small Claude model (Haiku) before the main model sees it. In Cloud mode this goes through our server like any other request; with your own Anthropic key it goes straight to Anthropic; with other providers there’s no summary step and the page text goes to your provider.
• To avoid reading pages on your local network unless you typed the address, the extension checks each page’s address and the server address a background tab loaded from. That server address is used only for the check and isn’t stored or sent anywhere.

While the side panel shows the start screen of a new conversation, it suggests things to ask about the page you’re on. To make them, the page’s title, address and first 800 or so characters are sent to the AI model when you open the panel, switch tabs or a page finishes loading. This is on by default in Cloud mode, and you can turn it off in Settings.

The first time you open Browser Agent, it shows a short summary of what’s sent and where and asks you to agree. Nothing is sent to an AI model, including home suggestions, until you agree. The installation registration described above happens at install, before this screen.

Plans are bought on our upgrade page, which loads Paddle’s checkout script from cdn.paddle.com. Paddle, our reseller and merchant of record, collects and processes your payment details; we never receive your full card number. Paddle sends us notifications about your subscription. We store your Paddle customer and subscription IDs, plan, billing interval, subscription status and dates, your email address (which we get from Paddle to contact you about your subscription) and the notifications as received. Paddle handles your data under its own privacy policy.

Our website has no analytics or advertising cookies. Our server sets two cookies of its own, both HttpOnly so page scripts can’t read them: __Host-ba_bid, a random value kept for 400 days that lets us recognize the same browser when the extension is reinstalled, so free credits can’t be claimed again (we store only its SHA-256 hash), and __Host-ba_auth, which exists for at most 10 minutes while you sign in. If you open a referral link (/ref/…), we count the click and set a cookie named ba_ref for 60 days, which the welcome page reads to credit the referral to your installation. The upgrade page loads Paddle’s checkout script, which may set its own cookies when you check out. Our hosting provider, [[Hosting provider]], may keep server logs such as IP addresses for [[Server log retention]].

• Anthropic, the model provider, processes request content in Cloud mode under its commercial terms and privacy policy
• The operator of our API gateway, [[API gateway operator]], passes requests to Anthropic
• Paddle receives what you enter at checkout
• Google or GitHub, if you sign in with them: they handle the sign-in and send us your account ID and email address
• Our hosting provider processes the data above on our behalf
• Search engines and the websites the agent opens receive what your browser normally sends

We don’t sell your data, and we don’t use your conversations or page content to train AI models.

We keep installation, sign-in, usage and subscription records for [[Record retention period]], to count credits and for accounting.

• Turn off home suggestions in Settings
• Use your own API key so tasks don’t go through our servers
• Delete conversations and memories in the extension, or remove the extension to delete everything stored in your browser
• Ask us to access or delete the records tied to your installation or account by writing to [[Contact email]]. For paid plans, include the email you used at checkout or to sign in so we can find your records.

Browser Agent isn’t meant for children under [[Minimum age]].

The use of information received from Google APIs will adhere to the Chrome Web Store User Data Policy, including the Limited Use requirements.

When this policy changes, we update the date at the top.

Privacy requests: [[Contact email]]. [[Company legal name]], [[Company address]].