Pi Browser Agent

A Chrome-native AI assistant for the page you are viewing.

Pi Browser Agent brings an AI assistant into Chrome’s Side Panel, helping you understand and work with the page you are viewing without switching tabs. Key features: • Ask questions about visible page content or selected text • Summarize, explain, rewrite, and extract information • Attach images for supported multimodal models • Dictate prompts using Chrome’s speech recognition • Discover visible page controls and, with your approval, click, type, or navigate • Capture the visible viewport after permission is granted • Search or review recent bookmarks after explicit confirmation • Create, rename, resume, and delete conversation sessions • Choose from supported AI providers and models • Customize instructions, fonts, and text size Pi Browser Agent operates only on the active, visible HTTP or HTTPS tab. Sensitive actions—including form submissions, downloads, cross-origin navigation, screenshots, bookmark access, and WebMCP calls—require confirmation. Password and file inputs are excluded from browser automation. Provider credentials are stored in Chrome’s local extension storage and are not exposed to websites or saved in conversation transcripts. When you send a prompt or approve an operation, relevant content may be sent to your selected AI provider. This may include your prompt, visible page content, selected text, attached images, approved screenshots, or approved bookmark results. Voice input uses Chrome’s Web Speech service. Only the editable transcript is submitted to the selected AI provider when you choose Send. To get started, select an AI provider and model, add an API key or sign in with an eligible ChatGPT account for OpenAI Codex, then open a webpage and enter your request in the Side Panel. Requirements: • Chrome 116 or later • An API credential for a supported provider, or a ChatGPT Plus/Pro account with Codex access Pi Browser Agent runs entirely as a Chrome extension. It does not require a local agent process, native host, shell, filesystem access, or local server. Source code: https://github.com/narumiruna/pi-chrome