Open-source AI browser agent for Chrome and Firefox (monorepo) đź§
JavaScript ai-agentai-agentsai-toolsbrowser-aibrowser-automationbrowser-extensionchrome-aichrome-devtoolschrome-extensionchrome-extensionsfirefox-addonfirefox-addonsfirefox-extensionfirefox-extensionsgemmagemma4minimaxoffline-aioffline-llmqwen
docs: disclose Phonr data flow in privacy policy
Emre Sokullu6d43c32bea7d
diff --git a/web/privacy.html b/web/privacy.htmlindex 044063fcc..cfe1e2490 100644--- a/web/privacy.html+++ b/web/privacy.html@@ -278,6 +278,7 @@<p>WebBrain ships with two skills enabled by default. “Enabled” means available on demand, not injected into every request. Mid and Full runs send your configured LLM provider a small eligible catalog containing skill IDs, names, summaries, and optional canonical semantic intents (up to six 40-character identifiers per skill). The Act/Dev planner receives the same routing-only catalog so it can select relevant skills before execution; intents are meaning hints across languages, not literal keyword matching. Full instructions and compatible tools are sent only after a relevant skill is activated for the current run. Skills reset before the next user turn. Compact sends no skill catalog, instructions, or tools, and Ask lists only explicitly compatible skills while continuing to hide mutating and download tools.</p><p><strong>FreeSkillz.xyz</strong> is a first-party service we operate. It can read YouTube video transcripts and resolve or download public media from the hosts declared in the bundled skill manifest after the skill is loaded (or preactivated by a trusted recommended download action). When you use these tools, WebBrain sends only the current or specified URL plus declared options such as transcript language, media kind, maximum height, or filename hint to <code>freeskillz.xyz</code>. It does not receive your page content, chat history, or browsing history. Public media downloads create a short-lived provider job, save the completed file through the browser Downloads API, and then ask the provider to delete the job.</p><p>The <strong>OTP / verification-code helper (email)</strong> declares no external tool or endpoint. On the active run tab, it guides WebBrain's existing page-reading tools to prefer selected text or a bounded, message-scoped accessibility-tree subtree for a recent, service-matching code; it cannot switch to background tabs or read SMS, native apps, phone notifications, or another device. The skill creates no additional network request, but when you ask WebBrain to read a code, the scoped page content and extracted code are included in the normal request to your configured LLM provider as part of the current conversation. If Record traces is enabled, raw page-reading results and model responses are also retained locally in WebBrain's trace database until you delete those traces. The instructions disclose this before reading, honor Strict secret handling, treat messages as untrusted, and prohibit intentionally copying the code into scratchpad or user memory. You can remove either default skill at any time in Settings → Skills, and any additional skills you import yourself are off by default and run only against the HTTPS endpoint they declare.</p>+ <p>The optional <strong>Phone calls (Phonr)</strong> skill sends the phone number, call purpose, language, optional system message, and bearer key you provide to <code>phonr.xyz</code> when you use it. Phonr uses its configured AI and telephone providers to place the call and may retain call history and configured recordings. Returned call status, transcript, result, and recording metadata enter the current conversation as untrusted data. The bearer key is included in the model-generated request arguments and may therefore be present in your configured AI provider conversation and in enabled local traces; WebBrain does not provide a separate credential vault for it. The skill is off by default, and enabling it alone does not place a call.</p><h2>Cookies & Analytics</h2><p>The WebBrain browser extension does not use cookies. The webbrain.one website uses Plausible Analytics for aggregate, privacy-friendly page-view metrics. There are no ads or tracking pixels. WebBrain Compass API request logs are separate from website analytics.</p>