Orca: Browsing at Scale Through User-Driven and AI-Facilitated Orchestration Across Malleable Webpages

TL;DR

Orca tackles the challenge of information foraging on the modern, distributed Web by redefining the browser as a malleable workspace and webpages as malleable materials. It blends user-driven interaction with AI-facilitated orchestration to enable viewing, navigating, organizing, extracting, operating, and synthesizing across many pages at scale. The paper introduces a spatial Web Canvas, extraction and synthesis mechanisms, and parallel agent-based automation, supported by a modular feedforward prompting system. An initial laboratory evaluation indicates enhanced exploration, flexible task organization, and increased user trust in results, while highlighting potential cognitive load and coordination challenges for multi-agent workflows. Overall, Orca points to a promising direction for future browser designs that tightly integrate AI to augment, rather than replace, human sensemaking across the Web.

Abstract

Web-based activities are fundamentally distributed across webpages. However, conventional browsers with stacks of tabs fail to support operating and synthesizing large volumes of information across pages. While recent AI systems enable fully automated web browsing and information synthesis, they often diminish user agency and hinder contextual understanding. Therefore, we explore how AI could instead augment users' interactions with content across webpages and mitigate cognitive and manual efforts. Through literature on information tasks and web browsing challenges, and an iterative design process, we present a rich set of novel interactions with our prototype web browser, Orca. Leveraging AI, Orca supports user-driven exploration, operation, organization, and synthesis of web content at scale. To enable browsing at scale, webpages are treated as malleable materials that humans and AI can collaboratively manipulate and compose into a malleable, dynamic, and browser-level workspace. Our evaluation revealed an increased "appetite" for information foraging, enhanced user control, and more flexibility in sensemaking across a broader information landscape on the web.

Figures (8)

• Figure 1: The Orca interface features a Web Canvas (b), where webpages can be spatially arranged and organized, and a side panel (a), which can be toggled (1) or set to fullscreen (4). The side panel shows pinned pages or summaries---either for individual webpages or multiple selected ones (3). Users can zoom out the viewport to view all pages at once (2). When a webpage is pinned, its instance on the canvas turns purple (6). Users can pin a set of pages and switch among them (5).
• Figure 2: In Orca, webpages can be arranged side by side. Users control individual pages via bottom controls---e.g., update URL (2) or navigate back or forward (3)---like in traditional browsers. They can also act on selected pages using the contextual menu (top), indicated by the selection box (8). The viewport can be adjusted to show one or multiple selected pages (1). Pages can be pinned to the sidebar for static view (4), arranged in a grid (5), batch closed (6), or extracted via the Page Extraction toggle to surface key information (7).
• Figure 3: Webpages can be organized into a grid (a) or stack (b) layout. Upon creation, Orca assigns the group a short, customizable name based on page content (1), along with its total page count (2). Users can switch between grid and stack layout (3) or convert the group into a table or visualization (4). Pages can be sorted (5) or selectively chosen (6) based on custom criteria. Users can adjust the number of columns in the grid layout to control its width (8). When no longer needed, groups can be dissolved (7).
• Figure 4: Users can extract and surface information across pages at scale for quick scanning. Users start by selecting a set of webpages (a), entering a simple extraction query (b), and pressing the Extract button (1). Orca then displays the extracted information in a consistent format across pages. Users can continue customizing the views of pages to surface the information they care about (c). Extraction queries are saved (2), and hovering over a query highlights the corresponding extracted information for all pages. Users can further close or sort pages (3) to continue exploration.
• Figure 5: Orca facilitates in-depth exploration by allowing users to batch open multiple links from a webpage. The Batch Open menu appears minimized by default at the top right of selected pages (a). On hover, users can pick from suggested queries or enter a custom one (b). Orca then searches the page and compiles matching links, showing the total count to support query refinement (c). Users can drag the capsule onto the Web Canvas to open all links, automatically arranged in a grid (d).